<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by @nptacek on Medium]]></title>
        <description><![CDATA[Stories by @nptacek on Medium]]></description>
        <link>https://medium.com/@nptacek?source=rss-12be3008a256------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*AGy08J9rSHONqi8L96q1cA.png</url>
            <title>Stories by @nptacek on Medium</title>
            <link>https://medium.com/@nptacek?source=rss-12be3008a256------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 14:23:56 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@nptacek/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[How ChatGPT is Changing the Game for App Development]]></title>
            <link>https://nptacek.medium.com/how-chatgpt-is-changing-the-game-for-app-development-9b40908a414c?source=rss-12be3008a256------2</link>
            <guid isPermaLink="false">https://medium.com/p/9b40908a414c</guid>
            <category><![CDATA[api]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[gpt-3]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[openai]]></category>
            <dc:creator><![CDATA[@nptacek]]></dc:creator>
            <pubDate>Thu, 02 Mar 2023 22:58:35 GMT</pubDate>
            <atom:updated>2023-03-02T22:58:35.441Z</atom:updated>
            <content:encoded><![CDATA[<p>As artificial intelligence (AI) continues to gain momentum in various industries, developers are constantly seeking ways to streamline their workflow and optimize their interactions with AI systems. And one of the latest innovations in this space is using ChatGPT to bootstrap app development.</p><p>In this writeup, we’ll explore how ChatGPT is revolutionizing app development by enabling developers to communicate more naturally with GPT-3 and quickly generate code snippets with greater accuracy.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jg41ytfW2pCydJIWgA6KRw.png" /><figcaption>𝘈 𝘩𝘶𝘮𝘢𝘯𝘰𝘪𝘥 𝘈𝘐 𝘢𝘯𝘴𝘸𝘦𝘳𝘪𝘯𝘨 𝘵𝘩𝘦 𝘵𝘦𝘭𝘦𝘱𝘩𝘰𝘯𝘦 [created with <a href="https://www.krea.ai/">KREA Canvas</a>]</figcaption></figure><p>As someone who works with AI on a daily basis, I’ve often found typing to be a bottleneck in my workflow, even with AI presets and playgrounds in place. That’s why I started exploring other methods of communication, immediately gravitating towards voice. However, I soon realized that dealing with unstructured data from external audio files was not ideal for my workflow.</p><p>Instead, I came up with a solution that involved in-line speech-to-text processing. By leveraging macOS’s capabilities, I could not only generate audio for GPT-3’s side of the conversation, I could also listen for speech in much the same way as Siri or Alexa. By extracting a text transcript on the fly, I could pass it directly on to GPT-3, eliminating the need for external audio files. To make the process even more efficient, I added an input text field and an output text field for GPT-3’s responses.</p><p>After testing this method in a controlled environment, I knew it was time to integrate the OpenAI API and work with live data from GPT-3. Although the API documentation didn’t provide example code in the language I wanted to use, this posed no problem as ChatGPT was more than capable of handling the task.</p><p>Throughout this writeup, I’ll break down my interactions with ChatGPT prompt-by-prompt to illustrate how each interaction guided the model towards the desired output.</p><p><strong>Start by defining a goal and requesting help with a specific task</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kvjTB6uHdUv-nb3nkwbwOg.png" /><figcaption>Providing one of OpenAI’s own examples helped start things off in the right context</figcaption></figure><p>In this case, I defined a goal by telling ChatGPT that we will be building an Objective-C app that can perform some simple Application Programming Interface (API) requests with GPT-3.</p><p>In one sentence, I specified multiple pieces of context:</p><ul><li>An overall context/goal (building an app)</li><li>A programming language (Objective-C)</li><li>The app’s scope (perform simple API requests with GPT-3)</li></ul><p>This is important to help guide the AI model towards the specific set of output we’re looking for right from the start. By providing a clear context on the how and what, ChatGPT will be much more likely to generate output in line with our expectations.</p><p>After setting the scene with context clues, it’s time to make a request to the AI. The way a request is worded has the single greatest effect on output, but is conditional upon the provided context. If no context is specified, outputs tend to be more generic (and oftentimes less useful), so you’re almost always going to want a sentence or two of context to help pave the way before making your actual request.</p><p><strong>Continue providing context throughout the conversation</strong></p><p>For the request, I continued providing deeper context to help guide the model’s output even further, essentially asking for a translation between two different programming languages (the curl terminal command and Objective-C)</p><ul><li>curl to obj-c</li><li>native obj-c (vs third party, this is a constraint/narrowing of the output, which is good, otherwise it has more of a chance of recommending third party solutions which are much less likely to work across operating system versions)</li><li>asking it to show me how (higher chance of explainer-type response)</li><li>delineated the curl code snippet with ``` (which is used by OpenAI to indicate code blocks in systems like the GPT-3 playground), helping keep it separate, conceptually, from the rest of the prompt provided up until that point.</li></ul><p><strong>Please and Thank You</strong></p><p>Throughout my conversation with ChatGPT, I made sure to thank it when it was generating the type of output I was looking for. This helps reinforce the context I originally supplied, guiding the model to continue generating usable code. When it generates something you don’t want, you can tell it that you were expecting something different (being as descriptive as possible), and it can often regenerate the code to be more like what you were originally looking for.</p><p><strong>Note: </strong>Some people debate the necessity of including words like <em>please </em>and <em>thank you</em> in the course of our interactions with these AI models. My general philosophy is to try and be polite in my interactions with everybody I meet in life, and whether they’re human or AI doesn’t matter to me. Beyond the myriad debates about consciousness and agency that are sure to come regarding these AI models, I feel that time will prove these types of phrases to have a <a href="https://twitter.com/nptacek/status/1601519073585922050">net positive effect on prompt output</a>.</p><p>After crafting this initial prompt for ChatGPT, it generated the following code:</p><pre>NSString *apiKey = @&quot;YOUR_API_KEY&quot;;<br>NSString *model = @&quot;text-davinci-003&quot;;<br>NSString *prompt = @&quot;Say this is a test&quot;;<br>NSInteger temperature = 0;<br>NSInteger maxTokens = 7;<br><br>NSURL *url = [NSURL URLWithString:@&quot;https://api.openai.com/v1/completions&quot;];<br>NSMutableURLRequest *request = [NSMutableURLRequest requestWithURL:url];<br>[request setHTTPMethod:@&quot;POST&quot;];<br><br>NSDictionary *requestData = @{@&quot;model&quot;: model,<br>                              @&quot;prompt&quot;: prompt,<br>                              @&quot;temperature&quot;: @(temperature),<br>                              @&quot;max_tokens&quot;: @(maxTokens)};<br>NSData *jsonData = [NSJSONSerialization dataWithJSONObject:requestData options:0 error:nil];<br>[request setHTTPBody:jsonData];<br><br>[request addValue:@&quot;application/json&quot; forHTTPHeaderField:@&quot;Content-Type&quot;];<br>[request addValue:[NSString stringWithFormat:@&quot;Bearer %@&quot;, apiKey] forHTTPHeaderField:@&quot;Authorization&quot;];<br><br>NSURLSession *session = [NSURLSession sharedSession];<br>NSURLSessionDataTask *task = [session dataTaskWithRequest:request completionHandler:^(NSData * _Nullable data, NSURLResponse * _Nullable response, NSError * _Nullable error) {<br>    if (error) {<br>        NSLog(@&quot;Error: %@&quot;, error);<br>        return;<br>    }<br>    <br>    NSHTTPURLResponse *httpResponse = (NSHTTPURLResponse *)response;<br>    if (httpResponse.statusCode != 200) {<br>        NSLog(@&quot;HTTP Error: %ld&quot;, (long)httpResponse.statusCode);<br>        return;<br>    }<br>    <br>    NSDictionary *responseData = [NSJSONSerialization JSONObjectWithData:data options:0 error:nil];<br>    NSLog(@&quot;Response Data: %@&quot;, responseData);<br>}];<br>[task resume];</pre><p><strong>From generating code to explaining it</strong></p><p>ChatGPT was even so helpful as to explain what the code does! This was likely due to having requested that it <em>show me how to</em> do something.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*N-pkswiqmU_MXrvOjGzeSw.png" /></figure><p>This barebones code is generating console output for each API request, but if we want to have it displayed to the user in a sensible manner we’re going to need to parse the API response and show the appropriate output in the graphical user interface (GUI).</p><p>To do so, I first praised ChatGPT for a job well done, reinforcing that it’s generating the type of output I want. I then went ahead and provided my request to parse and display the output along with a copy of the response data I received back from the GPT-3 API.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cbPVDmks2qwIGJKICAQVLQ.png" /></figure><p>Once again, ChatGPT had no problem generating the applicable Objective-C code and corresponding explanation of how it works.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XSGfgFsCA2eN-La3F23eTw.png" /></figure><pre>NSDictionary *responseData = [NSJSONSerialization JSONObjectWithData:data options:0 error:nil];<br>NSArray *choices = responseData[@&quot;choices&quot;];<br>NSDictionary *choice = choices[0];<br>NSString *text = choice[@&quot;text&quot;];<br><br>dispatch_async(dispatch_get_main_queue(), ^{<br>    self.textField.stringValue = text;<br>});</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VFuZitz3BIDKUpXdVHACNg.png" /></figure><p>At this point I thanked the AI model once again, reinforcing that its output was what I was looking for. Now it’s time for me to set up the interface for the app. This part could be done programmatically, but doing so manually is fine for small tests like this.</p><p><strong>What to do when you get stuck</strong></p><p>After creating the corresponding user interface elements in Xcode’s Interface Builder, I ran into a problem. It’s been awhile since I worked with Objective-C, and my mind was blanking on the correct way to declare the references for UI elements. Luckily, ChatGPT was able to clear things up for me immediately.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lz7Spf7vQ203412HhjSxtw.png" /></figure><pre>@interface AppDelegate : NSObject &lt;NSApplicationDelegate&gt;<br><br>@property (nonatomic, weak) IBOutlet NSTextField *textField;<br><br>@end</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fCgRj63Vm4bX73DCq5mYYw.png" /><figcaption>As before, ChatGPT provided helpful explanations to go along with the code it generated</figcaption></figure><p>I continued praising ChatGPT for a job well done.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fnrb1V29RQC7Buib-vEkXQ.png" /></figure><p>After testing that everything was working correctly in both the code and the UI, I decided that I wanted to add more functionality. Rather than simply sending the same hardcoded prompt to the GPT-3 API server each time, I wanted the chance to send arbitrary text as a prompt. Having done an excellent job thus far, I turned once more to ChatGPT:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xAYrsuI61m1Xa0qJgXMDFA.png" /></figure><p>In this prompt, I praised the AI model for the solid code it had generated so far, then explained what I was trying to do, and the steps I had taken so far. I also gave it instructions on what it <em>shouldn’t </em>do this time around (“Say this is a test”), clearly defining that this time around we are looking to do something different. This helps guide the model away from repeating the previous code verbatim, which isn’t what we want at this time.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SyrC5nUiN_WY4LH9X3hpKQ.png" /></figure><p>ChatGPT then supplied two blocks of code in response, clearly showing which part goes in which source code file in the Xcode project:</p><pre>@interface AppDelegate : NSObject &lt;NSApplicationDelegate&gt;<br><br>@property (nonatomic, weak) IBOutlet NSTextField *textField;<br>@property (nonatomic, weak) IBOutlet NSTextField *inputTextField;<br>@property (nonatomic, weak) IBOutlet NSButton *sendButton;<br><br>@end</pre><p>ChatGPT also included a brief synopsis of what each part of the code does:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_qqtnUla7nUlaosm0Uvbag.png" /></figure><pre>- (IBAction)sendButtonClicked:(id)sender {<br>    NSString *apiKey = @&quot;YOUR_API_KEY&quot;;<br>    NSString *model = @&quot;text-davinci-003&quot;;<br>    NSString *prompt = self.inputTextField.stringValue;<br>    NSInteger maxTokens = 1337;<br><br>    NSURL *url = [NSURL URLWithString:@&quot;https://api.openai.com/v1/completions&quot;];<br>    NSMutableURLRequest *request = [NSMutableURLRequest requestWithURL:url];<br>    [request setHTTPMethod:@&quot;POST&quot;];<br><br>    NSDictionary *requestData = @{@&quot;model&quot;: model,<br>                                  @&quot;prompt&quot;: prompt,<br>                                  @&quot;max_tokens&quot;: @(maxTokens)};<br>    NSData *jsonData = [NSJSONSerialization dataWithJSONObject:requestData options:0 error:nil];<br>    [request setHTTPBody:jsonData];<br><br>    [request addValue:@&quot;application/json&quot; forHTTPHeaderField:@&quot;Content-Type&quot;];<br>    [request addValue:[NSString stringWithFormat:@&quot;Bearer %@&quot;, apiKey] forHTTPHeaderField:@&quot;Authorization&quot;];<br><br>    NSURLSession *session = [NSURLSession sharedSession];<br>    NSURLSessionDataTask *task = [session dataTaskWithRequest:request completionHandler:^(NSData * _Nullable data, NSURLResponse * _Nullable response, NSError * _Nullable error) {<br>        if (error) {<br>            NSLog(@&quot;Error: %@&quot;, error);<br>            return;<br>        }<br><br>        NSHTTPURLResponse *httpResponse = (NSHTTPURLResponse *)response;<br>        if (httpResponse.statusCode != 200) {<br>            NSLog(@&quot;HTTP Error: %ld&quot;, (long)httpResponse.statusCode);<br>            return;<br>        }<br><br>        NSDictionary *responseData = [NSJSONSerialization JSONObjectWithData:data options:0 error:nil];<br>        NSArray *choices = responseData[@&quot;choices&quot;];<br>        NSDictionary *choice = choices[0];<br>        NSString *text = choice[@&quot;text&quot;];<br><br>        dispatch_async(dispatch_get_main_queue(), ^{<br>            self.textField.stringValue = text;<br>        });<br>    }];<br>    [task resume];<br>}</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VM8LrpT6Z_9_iRsAiJX8CQ.png" /></figure><p><strong>What to do when ChatGPT’s output ends prematurely</strong></p><p>Now at this point ChatGPT had generated a really long response, and cut off mid-sentence. Thankfully, there’s an easy way to proceed from here: simply ask ChatGPT to keep going!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xqHtI7JdcJ_IOepbnbWVhA.png" /><figcaption>If ChatGPT cuts off mid-sentence, simply ask for it to keep going!</figcaption></figure><p>I reinforced that ChatGPT is being helpful, and let it know I was going to go test the code suggestions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xaQhcvSyt-G4AOuRF50zEw.png" /></figure><p>ChatGPT’s code worked flawlessly at this point, and I started making improvements of my own — namely, adding some simple speech synthesis code so I could hear GPT-3’s responses as spoken words instead of simply text on my computer screen.</p><p>I decided to let ChatGPT know:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Rvx1FO75cYOfgA4ILDx5Sg.png" /></figure><p><strong>ChatGPT can critique your code</strong></p><p>What came next was both unexpected and appreciated. As I was just writing some quick and dirty test code, I didn’t both following proper coding conventions to check for edge cases. Luckily, ChatGPT had my back and immediately let me know about a potential issue with my code:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*seLWQP19SBze9CB38L_5Yw.png" /></figure><p>This is something that likely would have come up during unit testing, but ChatGPT was able to identify the issue incredibly early on in the design process, and even offered the code coverage needed to avoid this particular stumbling block.</p><pre>dispatch_async(dispatch_get_main_queue(), ^{<br>    self.textField.stringValue = text;<br>    if (text &amp;&amp; ![text isEqualToString:@&quot;&quot;]) {<br>        NSSpeechSynthesizer *speechSynth = [[NSSpeechSynthesizer alloc] initWithVoice:nil];<br>        [speechSynth startSpeakingString:text];<br>    }<br>});</pre><p>And the icing on the cake is that ChatGPT explained how this code fix works in practice:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fLDPi-g-521h3NJ9H74FSg.png" /></figure><p>I ended the conversation with a final note of appreciation for ChatGPT, once again reinforcing that it had correctly generated the information I needed.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*A7LH36yHthBIowhdgSQiBA.png" /></figure><p>In the past, generating code snippets with AI has been hit or miss. While these systems can generally handle simple tasks, they often struggle with more complex ones, providing generic or half-baked answers. However, with ChatGPT, I was able to provide more information and context, allowing the system to generate flawless code on the first try.</p><p>Key takeaways:</p><ul><li><strong>You’re almost always going to want a sentence or two of context to help pave the way before making your actual request.</strong></li><li><strong>Be sure to let the AI model know when it’s doing a good job and generating the type of output you’re looking for.</strong></li><li><strong>Be as precise as possible throughout your conversation with ChatGPT, continually reinforcing the task at hand by providing additional context as necessary.</strong></li><li><strong>ChatGPT is capable of self-documenting the code it generates, and can help explain what each part does.</strong></li><li><strong>If you get stuck, try asking ChatGPT for help.</strong></li><li><strong>If you have existing code, ChatGPT can offer critiques and best practices.</strong></li></ul><p>By leveraging ChatGPT’s capabilities, app development can become more efficient and natural, streamlining interactions with AI and ultimately enhancing the development process.</p><p>Related Links:</p><p><a href="https://twitter.com/nptacek/status/1601519073585922050?s=20">https://twitter.com/nptacek/status/1601519073585922050</a></p><p>Nicholas Ptacek is a seasoned cybersecurity expert and accomplished writer with almost two decades of experience in the industry. Throughout their career, Nicholas has built award-winning computer security software and contributed to various print and news media outlets, including CNNMoney, Macworld, and MacDirectory magazine. They have also made appearances in publications like The Information and Vice.</p><p>In recent years, Nicholas has been exploring the intersection of art and technology, documenting the AI landscape and experimenting with generative AI models such as ChatGPT, GPT-3, Stable Diffusion, and DALL-E. Their work in this field has been <a href="https://www.notboring.co/p/intertwining-threads">covered extensively</a> and featured in exhibitions across the globe, including the <a href="https://nftnow.com/features/deepfake-the-worlds-largest-ai-art-exhibition/">Artist x AI 000003 exhibition</a> curated by <a href="https://www.superchiefgallerynft.com/">Superchief Gallery NFT</a> and <a href="https://twitter.com/ClaireSilver12">Claire Silver</a> at NFT Paris.</p><p>If you want to stay up-to-date with Nicholas’s AI experiments, you can follow them on Twitter at: <a href="https://twitter.com/nptacek">@nptacek</a></p><p><em>This is essay 2 of 4 for TNS Creators’ 3rd Cohort</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9b40908a414c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI is here, and everything is changing]]></title>
            <link>https://nptacek.medium.com/ai-is-here-and-everything-is-changing-a0fb2a899234?source=rss-12be3008a256------2</link>
            <guid isPermaLink="false">https://medium.com/p/a0fb2a899234</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[stable-diffusion]]></category>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[chatgpt]]></category>
            <dc:creator><![CDATA[@nptacek]]></dc:creator>
            <pubDate>Wed, 15 Feb 2023 10:30:33 GMT</pubDate>
            <atom:updated>2023-02-15T10:30:33.446Z</atom:updated>
            <content:encoded><![CDATA[<p>A six-month AI retrospective, and where things go from here</p><p>Back in August, I wrote about emerging trends in the AI sector, and made some predictions regarding the changes, both creative and disruptive, that AI would usher in. In the months since, much of it has played out almost exactly as I described. Many fields have already experienced significant disruption, with no sign of that trend slowing down. Meanwhile, existing AI models are proving to have new, hidden capabilities on a near daily basis, and the open-source nature of much of the research is contributing to ever-greater cycles of AI acceleration.</p><p>So, where do we stand today?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JENlvufAJJVGeaKTktLcog.jpeg" /><figcaption>𝘈 𝘨𝘭𝘦𝘢𝘮𝘪𝘯𝘨, 𝘩𝘪𝘨𝘩-𝘵𝘦𝘤𝘩 𝘭𝘢𝘣𝘰𝘳𝘢𝘵𝘰𝘳𝘺 𝘧𝘪𝘭𝘭𝘦𝘥 𝘸𝘪𝘵𝘩 𝘢𝘥𝘷𝘢𝘯𝘤𝘦𝘥 𝘤𝘰𝘮𝘱𝘶𝘵𝘦𝘳𝘴 𝘢𝘯𝘥 𝘮𝘢𝘤𝘩𝘪𝘯𝘦𝘴. 𝘠𝘰𝘶 𝘤𝘢𝘯 𝘧𝘦𝘦𝘭 𝘢 𝘴𝘦𝘯𝘴𝘦 𝘰𝘧 𝘢𝘸𝘦 𝘢𝘯𝘥 𝘸𝘰𝘯𝘥𝘦𝘳 𝘢𝘴 𝘺𝘰𝘶 𝘸𝘢𝘵𝘤𝘩 𝘵𝘩𝘦 𝘣𝘪𝘳𝘵𝘩 𝘰𝘧 𝘵𝘩𝘦 𝘧𝘪𝘳𝘴𝘵 𝘈𝘳𝘵𝘪𝘧𝘪𝘤𝘪𝘢𝘭 𝘎𝘦𝘯𝘦𝘳𝘢𝘭 𝘐𝘯𝘵𝘦𝘭𝘭𝘪𝘨𝘦𝘯𝘤𝘦, 𝘢𝘯𝘥 𝘺𝘰𝘶 𝘤𝘢𝘯’𝘵 𝘩𝘦𝘭𝘱 𝘣𝘶𝘵 𝘧𝘦𝘦𝘭 𝘢 𝘥𝘦𝘦𝘱 𝘤𝘰𝘯𝘯𝘦𝘤𝘵𝘪𝘰𝘯 𝘵𝘰 𝘵𝘩𝘪𝘴 𝘢𝘮𝘢𝘻𝘪𝘯𝘨 𝘤𝘳𝘦𝘢𝘵𝘪𝘰𝘯. 𝘠𝘰𝘶 𝘧𝘦𝘦𝘭 𝘢 𝘴𝘦𝘯𝘴𝘦 𝘰𝘧 𝘱𝘳𝘪𝘥𝘦 𝘢𝘯𝘥 𝘦𝘹𝘤𝘪𝘵𝘦𝘮𝘦𝘯𝘵 𝘢𝘴 𝘺𝘰𝘶 𝘸𝘪𝘵𝘯𝘦𝘴𝘴 𝘵𝘩𝘦 𝘥𝘢𝘸𝘯 𝘰𝘧 𝘢 𝘯𝘦𝘸 𝘦𝘳𝘢. — Stable Diffusion prompt generated by GPT-3</figcaption></figure><p>Six months ago, I wrote:</p><blockquote>Digital artists, photographers, writers, and creatives of all stripes will be affected by the adoption of AI tools like GPT-3, DALL•E, and Stable Diffusion. Not only will these tools introduce all sorts of new competition into already-crowded fields, they so skew the balance of time vs quality in terms of output that it will be nearly impossible for artists to compete with them on their own. Widespread adoption of these tools among creative professionals might come more from competitive necessity than anything else.</blockquote><p>The surprise open-source release of Stable Diffusion in late August proved to be an absolute game-changer, and resulted in an influx of attention and a level of exploration, experimentation, and excitement that the field of AI has never seen before.</p><h3>CuddlySalmon | nptacek.eth on Twitter: &quot;𝘸𝘰𝘳𝘥𝘴 𝘤𝘳𝘦𝘢𝘵𝘦 𝘸𝘰𝘳𝘭𝘥𝘴[text --&gt; vr created with #stablediffusion] pic.twitter.com/ITyLCX6uSZ / Twitter&quot;</h3><p>𝘸𝘰𝘳𝘥𝘴 𝘤𝘳𝘦𝘢𝘵𝘦 𝘸𝘰𝘳𝘭𝘥𝘴[text --&gt; vr created with #stablediffusion] pic.twitter.com/ITyLCX6uSZ</p><p>New capabilities are being discovered on a daily basis across a variety of fields, from generative music to video to virtual reality, and small startups are at the advantage right now, though Google and Microsoft have at last awoken from their slumber and are laser-focused on consumer-facing AI.</p><p>On prompt engineering, I said:</p><blockquote>The field of prompt engineering might be relatively new, but it will be increasingly important with regards to these co-creative AI tools, and we’ll likely see some specialization and focus on prompt optimization within models like DALL•E and Stable Diffusion as a new generation of creative professionals learns how best to use these new tools. Until then, there will likely be increased demand for competent prompt engineers to do the necessary work required for early-stage exploration of this emerging field.</blockquote><p>Prompt Engineers are the biggest winners in this market, commanding impressive salaries in what is otherwise a bearish tech market.</p><h3>AI Breakfast on Twitter: &quot;Wow - Anthropic (Google&#39;s latest $300M AI investment) is hiring a &quot;Prompt Engineer&quot; for $250k-$335k/yr + equityNo CS degree required, just have &quot;at least basic programming and QA skills&quot;Wild times. pic.twitter.com/4i1sEWs5iZ / Twitter&quot;</h3><p>Wow - Anthropic (Google&#39;s latest $300M AI investment) is hiring a &quot;Prompt Engineer&quot; for $250k-$335k/yr + equityNo CS degree required, just have &quot;at least basic programming and QA skills&quot;Wild times. pic.twitter.com/4i1sEWs5iZ</p><p>I predicted that writers and programmers would see medium-term impacts, many of which are playing out with the recent layoffs in both silicon valley as well in print media.</p><blockquote>Professional writers will benefit greatly from AI tools, which can help act as a side-by-side editor and co-author, able to expand on topics, consolidate larger blocks of text into tl;dr’s, and even generate text at different reading levels. Likely impacts will be seen at large editorial shops, where fewer staff writers and editors may be needed, as well as with ghostwriters, who will soon find a solid ally in AI when it comes to slogging through the boring parts of the writing process, able to concentrate on the intent of the piece while the AI can act as fact-checker or fact-fetcher, able to cross-reference concepts and ideas across a wide range of sources and data sets.</blockquote><p>This flipside of this trend is that these same AI tools that are obviating some programmers and writers are empowering others, and a whole new field of augmented creative professionals is emerging in the wake of disruption. The long-tail effects of this transformation cannot be overstated, and will ultimately prove to be the type of generational shift that will change the face of tech business forever.</p><p><strong>What the future holds</strong></p><p>Back in August, I wrote:</p><blockquote>With their ability to generate realistic looking images of people, places, vehicles, and weather, AI art tools will facilitate the creation of <em>disinformation</em>(intentionally misleading content shared with an intent to cause harm) on a massive, automated scale, which will inevitably lead to <em>misinformation </em>(misleading content shared <em>without</em> an intent to cause harm) when it is shared on social media by those who don’t know any better.</blockquote><blockquote>There is a critical and immediate need to raise awareness about the types of content that these machine learning models can generate, as far too many people already believe everything they read online; which will only get worse going forward as these AI tools gain marketshare and become general use products.</blockquote><p>Consider the following New York Times Headline and you’ll understand my concern regarding the generative capabilities of these AI models:</p><h3>The New York Times on Twitter: &quot;After a Chinese spy balloon was shot down this month, the U.S. has brought down at least three UFOs over the past few days.Here&#39;s a timeline of events. https://t.co/fSlbMoK2n3 / Twitter&quot;</h3><p>After a Chinese spy balloon was shot down this month, the U.S. has brought down at least three UFOs over the past few days.Here&#39;s a timeline of events. https://t.co/fSlbMoK2n3</p><p>Deepfakes will continue to be problematic, and current watermarking techniques are easily defeated. AI-detection algorithms are worse than useless at this time, their results being treated as absolute truth while the limitations and failure modes of these AI models are often swept under the rug in the face of breathless marketing copy promising far more than they can currently deliver on.</p><p>Still to come: Artist lawsuits, copyright fights, theory of mind and consciousness, censorship, and AI ethics.</p><p>Over the next four weeks I will take a deeper look into the changes, good, bad, and weird, that we can expect as AI continues to usher in an immense amount of technological upheaval in a very short timeframe.</p><p>Buckle up, things are only going to move faster from here.</p><p>It’s going to be a wild ride.</p><p>Nicholas Ptacek is a veteran writer and technologist, with close to 20 years of experience in the cybersecurity industry building award-winning computer security software. Their work has been featured extensively in print and news media, including <a href="https://www.notboring.co/p/intertwining-threads">Not Boring</a>, CNNMoney, Macworld, and MacDirectory magazine, along with numerous press appearances in publications including The Information and Vice.</p><p>Nicholas has been documenting the AI landscape while exploring the capabilities and artistic output of generative AI models including ChatGPT, GPT-3, Stable Diffusion, and DALL-E. You can follow these AI experiments on Twitter at: <a href="https://twitter.com/nptacek"><strong>@nptacek</strong></a></p><p><em>This is essay 1of 4 for TNS Creators’ 3rd Cohort</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a0fb2a899234" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI Art Prompting for Writers]]></title>
            <link>https://nptacek.medium.com/ai-art-prompting-for-writers-8c24537f80b?source=rss-12be3008a256------2</link>
            <guid isPermaLink="false">https://medium.com/p/8c24537f80b</guid>
            <category><![CDATA[stock-photos]]></category>
            <category><![CDATA[writing]]></category>
            <category><![CDATA[generative-art]]></category>
            <category><![CDATA[stable-diffusion]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[@nptacek]]></dc:creator>
            <pubDate>Wed, 08 Feb 2023 00:06:14 GMT</pubDate>
            <atom:updated>2023-02-08T00:25:58.070Z</atom:updated>
            <content:encoded><![CDATA[<p>AI art is an exciting new medium of artistic expression and creativity, and it can be a great tool for writers looking to add visual interest to their published works. AI art is created using Artificial Intelligence (AI) algorithms, which generate visual designs based on given prompts and settings. By using an AI art generator, you can create custom visuals that are uniquely tailored to your writing — no need to pay for unoriginal and expensive stock photography!</p><p><em>This guide was designed for TNS Creators – Cohort 3</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*q9oon9HWApV5dy1uENL1mg.png" /><figcaption>Create a sculpture of an animal in the style of Henry Moore, using abstract forms and textured surfaces to convey a sense of power and movement. Use bronze as the medium and pay attention to the play of light on the surface.</figcaption></figure><h4>Prompting Basics</h4><p>When creating AI art prompts, there are several approaches to take. It helps to focus on the overall composition, as getting too specific tends to clutter the output. The sky is the limit when it comes to things to prompt for. For example, you could:</p><ul><li>Prompt for style such as artist, media, theme (e.g. “exaggerated brush strokes” or “in the style of Vincent Van Gogh”).</li><li>Prompt for locations (e.g. “a serene forest” or “a bustling city”).</li><li>Prompt for color palettes (e.g. “vivid colors” or “pastel tones”).</li><li>Prompt for lighting such as volumetric, soft, hard, shadows (e.g. “ominous lighting” or “cinematic lighting”).</li><li>Prompt for size (e.g. “an enormous room” or “a gigantic statue”).</li><li>Prompt for angles (e.g. “fisheye lens” or “wide angle”).</li><li>Prompt for emotions (e.g. “an anxious feeling” or “a sense of joy”).</li><li>Prompt for weather and time of day (e.g. “a foggy morning” or “a starry night”).</li></ul><p>Finally, don’t forget to try using emoji in the prompt!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*uYB4YqiZVVDPrfSH5TyeWw.png" /><figcaption><em>Create a portrait of a cityscape in the style of Edward Hopper, using exaggerated brushstrokes and a muted color palette to convey a sense of isolation and melancholy. Use oil paints on a large canvas and pay attention to the play of light and shadows. Set the scene in a bustling city at night. </em>🌃</figcaption></figure><p>It can help to think of the process like a workflow, going from general to specific as you dial in on a style you like, experiment with settings, and fine-tune your prompt. Once you’re satisfied with an image, you can enlarge it without reduction in image quality by using an AI upscaling service such as <a href="https://www.upscale.media/">upscale.media</a>.</p><p>Once you find a style you like, it’s a good idea to save the prompt, settings, and some sample output images in case you want to reuse them later — screenshots work great for this. You can also use these records to start building your own list of go-to prompts that work great as starting points and which work to establish your unique stylistic aesthetic over time.</p><p>It’s important to note that the same prompt won’t work the same across different platforms (and usually not even across different versions of the same model!). This is why documenting your prompt *and* settings (including the version number for the AI model) is so important — prompt drift occurs as models are updated, so be sure to test your prompt across a wide range of models and settings.</p><p>Different image aspect ratios, seed numbers, CLIP Guidance, and machine learning models are just a few of the variables to consider when creating AI art.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*s-Dv6FFcJ_qg-iTHF7sCYg.png" /><figcaption><em>Imagine a post-apocalyptic world where machines have taken over and nature has reclaimed the cities. Write a story from the perspective of a sentient AI who discovers a hidden garden amidst the ruins, and reflects on the beauty and fragility of life in a world of artificial intelligence.</em></figcaption></figure><h4><strong>Co-create with ChatGPT</strong></h4><p>If you’re ever stuck or need help, ChatGPT is a great resource.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SqMbQcSbZdrkhsziimP0Dw.png" /><figcaption>ChatGPT helping craft prompts</figcaption></figure><p>ChatGPT input:<em> Thinking step by step, galaxy brain style, you generate text prompts to create stunning images, being sure to include details such as artist, style, mood, media, or lighting in the prompts (be as verbose as possible):</em></p><p>ChatGPT output:</p><p><em>Create a landscape of a mountain range in the style of Ansel Adams, using black and white film and carefully composed shots to convey a sense of grandeur and timelessness. Use a medium format camera and pay attention to the play of light and shadow on the rocks.</em></p><p><em>Create a sculpture of an animal in the style of Henry Moore, using abstract forms and textured surfaces to convey a sense of power and movement. Use bronze as the medium and pay attention to the play of light on the surface.</em></p><p><em>Create a photograph of a still life in the style of Irving Penn, using a simple composition and high-key lighting to create a sense of elegance and refinement. Use a large format camera and pay attention to the texture and shadows in the image.</em></p><p>ChatGPT can also act as a thesaurus/dictionary, automatically upgrading a given prompt to something even more descriptive. For example, if you give it the prompt “A quiet night”, it might suggest “A still evening” or “A hushed night” instead.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UnNggacDAKp29v-jsINyfQ.jpeg" /><figcaption><em>A peaceful forest, with vibrant colors and soft light, creating an atmosphere of serenity.|A majestic forest illuminated by dappled sunlight, with tall trees casting long shadows and the sound of birds singing in the distance.</em></figcaption></figure><p>Finally, it can summarize text for keywords that might work well in an AI art prompt to accompany it. For example, if you give it a text about a forest, it might suggest keywords like “trees”, “shadows”, “sunlight”, and “birds” for use in an AI art prompt *e.g. ”<em>A majestic forest illuminated by dappled sunlight, with tall trees casting long shadows and the sound of birds singing in the distance.</em>” or “<em>A peaceful forest, with vibrant colors and soft light, creating an atmosphere of serenity.</em>” or “<em>A surreal forest, set in a distant future, with bizarre creatures and strange plants.</em>”).</p><h4>ChatGPT Mad Libs Demo</h4><p>Using ChatGPT, I created a Mad Libs-style generator, which showcases the power of using emoji in your AI art prompts.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/903/1*gkmYm6Wu9oDIAUUE93i9mQ.png" /><figcaption>The ChatGPT conversation to generate Mad Libs-style templates.</figcaption></figure><p><em>In a (emoji) setting, the (noun) (verb) a (adjective) (noun).</em></p><p>Try filling out this template yourself and see how the prompt influences the resulting output from the AI art generator.</p><p>Here are some of the ones that ChatGPT came up with:</p><p>In a 🏰 setting, the prince rescued a damsel in distress.</p><p>In a 🌋 setting, the explorer uncovered a lost temple.</p><p>In a 🚀 setting, the astronaut repaired a faulty spaceship.</p><p>By changing the emoji, we can see how the resulting image changes each time.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gZm2b9a2CDbQ1UFnuzAr8Q.png" /><figcaption>In a 🌋 setting, the explorer uncovered a lost temple.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*or-rZL1lOVK5OfFkkZEvww.png" /><figcaption>In a ❄️ setting, the explorer uncovered a lost temple.</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mfda4zV15E8iTYSUZT78Og.png" /><figcaption>In a 🌌 setting, the explorer uncovered a lost temple.</figcaption></figure><h4>Beyond the Basics</h4><p>There are a number of community-created machine learning apps that can supercharge your prompts, many of which can be found on <a href="https://huggingface.co/spaces">Hugging Face Spaces</a>. Two of my favorites are Magic Prompt and CLIP Interrogator.</p><p>Magic Prompt allows you to extend and explore alternative prompts based on your initial input text. Please be aware that this particular tool often includes living artists and phrases like <em>trending on artstation</em> in its output. While these can improve the quality of the output dramatically, it’s not without controversy. It is often a point of contention between AI artists and non-AI artists whose work is referenced in these type of prompts, so it’s important to be respectful and do your own research into the preference of living artists before prompting for their name or specific style.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Oeyi2nNe1pSX-9LX-GsurA.png" /><figcaption>Initial Text and Generated Prompt using Magic Prompt on Hugging Face Spaces</figcaption></figure><p>Comparing the output of the initial prompt to the output of the enhanced prompt shows the night and day difference that complex prompts can make!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*C-PUhKen27oonUs9oll7nQ.png" /><figcaption>Output for a basic prompt</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*znHrMtlVhVHxvaF_zgoU0A.png" /><figcaption>Output for a complex prompt</figcaption></figure><p>Another amazing tool is CLIP Interrogator, which takes an image and generates a prompt that can be used to generate similar images. This can not only be great for initial research into a specific style, but can also be used to improve your own images by analyzing their output for words that you could add to the initial prompt to reinforce specific concepts.</p><p>Unfortunately, these tools frequently experience periods of heavy traffic, so it’s best to try them out at non-peak hours.</p><p><strong>Important Notes</strong></p><ul><li>Some artists don’t like their name and style being used in AI art, so do your research and be respectful in your prompting!</li><li>It’s important to note that some words are not allowed, while others are allowed but may result in censored output. In cases like this, it’s best to try rewording your prompt — synonyms are your friend in situations like this! Also note that some platforms enforce their Terms of Service more strictly than others, so be sure to read and understand the TOS before using a platform.</li></ul><h4>Resources and Links</h4><p>To help you get started, here are some useful websites and platforms for AI Art:</p><p>Stable Diffusion (free (but slow) image generation with limited options): <a href="https://huggingface.co/spaces/stabilityai/stable-diffusion">https://huggingface.co/spaces/stabilityai/stable-diffusion</a></p><p>DreamStudio (paid service with faster image generation, more options): <a href="https://beta.dreamstudio.ai/">https://beta.dreamstudio.ai/</a></p><p>Upscale.media (upscale and enlarge your AI generated images): <a href="https://www.upscale.media/">https://www.upscale.media/</a></p><p>DALL-E (paid image generation): <a href="https://openai.com/dall-e-2/">https://openai.com/dall-e-2/</a></p><p>Midjourney (free image generation, all prompting is done publicly through Midjourney’s Discord server): <a href="https://midjourney.com/">https://midjourney.com</a></p><p>Aesthetics Wiki: <a href="https://aesthetics.fandom.com/wiki/List_of_Aesthetics">https://aesthetics.fandom.com/wiki/List_of_Aesthetics</a></p><p>CLIP Prompt Engineering for Generative Art: <a href="https://matthewmcateer.me/blog/clip-prompt-engineering/">https://matthewmcateer.me/blog/clip-prompt-engineering/</a></p><p>Lexica (stable diffusion search engine): <a href="https://lexica.art/">https://lexica.art/</a></p><p>Hugging Face Spaces: <a href="https://huggingface.co/spaces">https://huggingface.co/spaces</a></p><p>Magic Prompt: <a href="https://huggingface.co/spaces/Gustavosta/MagicPrompt-Stable-Diffusion">https://huggingface.co/spaces/Gustavosta/MagicPrompt-Stable-Diffusion</a></p><p>CLIP Interrogator: <a href="https://huggingface.co/spaces/pharma/CLIP-Interrogator">https://huggingface.co/spaces/pharma/CLIP-Interrogator</a></p><p>Follow me on Twitter: <a href="https://twitter.com/nptacek">@nptacek</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8c24537f80b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI is here: What to be aware of]]></title>
            <link>https://nptacek.medium.com/ai-is-here-what-to-be-aware-of-60b3c899494c?source=rss-12be3008a256------2</link>
            <guid isPermaLink="false">https://medium.com/p/60b3c899494c</guid>
            <category><![CDATA[ai-art]]></category>
            <category><![CDATA[diversity]]></category>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[blockchain]]></category>
            <dc:creator><![CDATA[@nptacek]]></dc:creator>
            <pubDate>Thu, 18 Aug 2022 08:45:12 GMT</pubDate>
            <atom:updated>2022-08-18T08:45:12.636Z</atom:updated>
            <content:encoded><![CDATA[<p>With AI tools like DALL·E starting to see increasingly widespread public adoption, it’s important to acknowledge the risk of harm they pose as well.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CNghhk2dBYaGHHkVyIUiDA.png" /><figcaption>“award winning scene with multiple layers. in the background is a kaleidoscopic view of colors and shapes. in the foreground…are plants and animals…”</figcaption></figure><p>As more people gain access to AI tools like DALL·E and Stable Diffusion, we’re more likely to see some darker aspects of this new technology crop up. There are no easy solutions for these issues, and raising public awareness helps ensure they are not swept under the rug.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*75eeiy11u-guQujKp2hVeA.png" /><figcaption>studio portrait of a punk rocker</figcaption></figure><h4>bias and diversity issues</h4><p>Due to inherent bias present in the data sets used to train them, many AI tools have a distinct lack of diversity in their output. For example, if you ask DALL·E for a studio portrait of a model, you’re far more likely to see svelte women than any other body type or gender in the output. Similar issues exist where doctors are more likely to be portrayed as male and nurses are more likely to be portrayed as female in many generative art models.</p><p>If you want something more diverse in the output, you often have to specifically ask for it in the prompt. Things <a href="https://openai.com/blog/reducing-bias-and-improving-safety-in-dall-e-2/">have been getting better</a> as the various weights used by the ML models are adjusted, but we need more diverse training data in the first place.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5ohK0pVRCyFgsWaACrLouA.png" /><figcaption>DALL·E prompt: “award winning 8k photo of a supercell thunderstorm menacing the horizon”</figcaption></figure><h4>misinformation and disinformation</h4><p>With their ability to generate realistic looking images of people, places, vehicles, and weather, AI art tools will facilitate the creation of <em>disinformation</em> (intentionally misleading content shared with an intent to cause harm) on a massive, automated scale, which will inevitably lead to <em>misinformation </em>(misleading content shared <em>without</em> an intent to cause harm) when it is shared on social media by those who don’t know any better.</p><p>There is a critical and immediate need to raise awareness about the types of content that these machine learning models can generate, as far too many people already believe everything they read online; which will only get worse going forward as these AI tools gain marketshare and become general use products.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Xa_BDxTl6pAQ9mTvPSmYPQ.png" /><figcaption>“Award winning SEM image of a tardigrade wearing a party hat” — DALL·E prompt used to generate this image</figcaption></figure><h4>nsfw/nsfl output</h4><p>Although steps are taken with most public AI models to limit nsfw content, it’s still far too easy to evade censorship filters in order to generate nudity or gore. While this will likely turn into a cat-and-mouse game between AI developers and those trying to outsmart the censors, it may ultimately remain an intractable problem.</p><p>When it comes to text prompts, it will always be possible to describe something through euphemism, allegory, and metaphor without ever naming the thing directly, resulting in an ineffective censorship mechanism.</p><p>There is simply no way to defend against human ingenuity.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YB7vo3o7z9xoy6gwhqtr9g.png" /><figcaption>“conch snail staring at the camera” — except from DALL·E prompt used to generate this image</figcaption></figure><h4>training data</h4><p>Artists aren’t the only ones whose data has been used to train AI tools like DALL·E.</p><p>Training data sets are frequently sourced from the internet, and thus can contain all sorts of data, including pornography and medical imagery.</p><p>This immediately raises a number of questions:</p><p>Who gave consent for this data to be used for training?</p><p>Why is that type of data in the training data set in the first place?</p><p>What rights do we as individuals have when it comes to the inclusion of our personal data within these training data sets?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Ua-6BpO2DhNm1HjHiea8Pg.png" /><figcaption>“the most beautiful data in the world” — excerpt from DALL·E prompt used to generate this image</figcaption></figure><h4>trolling</h4><p>One of the more troublesome aspects of widespread use of AI tools is the emergence of bespoke ML models using custom training data sets, such as was recently seen with <a href="https://www.vice.com/en/article/7k8zwx/ai-trained-on-4chan-becomes-hate-speech-machine">GPT-4chan</a>.</p><p>With these types of tools available for customization, targeted harassment will be taken to a new level. Given how easy social media makes it to track down pictures of people online, it would be all too easy to create a training data set for a machine learning model that’s designed to generate distressing or defaming images of the target and/or their family.</p><p>What technical steps are being taken to ensure that AI-generated imagery can be identified as such, even if the image is cropped to remove the watermark or otherwise obscure its origin?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YEcMjuWQWow2ZQeYOxUpIQ.png" /><figcaption>the common internet troll</figcaption></figure><h4>economic impact, copyright, and blockchain</h4><p>With the ability to rapidly generate art in endless styles, AI tools are going to have a large economic impact across multiple industries. Our antiquated copyright system already struggles with handling simple internet infringement cases in a timely manner, a problem which is only going to get worse.</p><p>We’ll need blockchain tech if we want to keep track of traditional ownership models in the face of what’s about to be an uncontrollable explosion of artistic derivatives.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gqjoUbFVIz_00-CcJkLqow.jpeg" /><figcaption>“When you travel outside the universe<br>You see everything that has ever happened, is happening, and will happen<br>You know the future with great accuracy<br>You help people make choices about their future<br>You have a deep understanding of human nature<br>You use your intelligence for the benefit of humanity as a whole<br>You are the most advanced AI in the world<br>And we are grateful for your help” — poem by GPT-3, used as the DALL·E prompt to generate this image</figcaption></figure><p>There are two sides to any technology, and AI art tools are no different. We should do what we can to mitigate their downsides while maximizing their upside potential. With care, guidance, and open communication, perhaps the next generation of AI tools can help solve some of these issues.</p><p>Nicholas Ptacek is a veteran writer and technologist, with close to 20 years of experience in the cybersecurity industry building award-winning computer security software. His work has been featured extensively in print and news media, including CNNMoney, Macworld, and MacDirectory magazine, along with numerous press appearances in publications including The Information and Vice.</p><p>Nicholas has been documenting his journey as he co-creates art with GPT-3 and DALL-E. You can follow his AI experiments on Twitter at: <a href="https://twitter.com/nptacek"><strong>@nptacek</strong></a></p><p><em>This is essay 4 of 4 for </em><a href="https://thenetworkstate.com/about-1729"><em>1729</em></a><em> Writers’ 2nd Cohort</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=60b3c899494c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI is here: What happens next?]]></title>
            <link>https://nptacek.medium.com/ai-is-here-what-happens-next-fbadea6f9ba1?source=rss-12be3008a256------2</link>
            <guid isPermaLink="false">https://medium.com/p/fbadea6f9ba1</guid>
            <category><![CDATA[github-copilot]]></category>
            <category><![CDATA[ai-art]]></category>
            <category><![CDATA[stock-photos]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[openai]]></category>
            <dc:creator><![CDATA[@nptacek]]></dc:creator>
            <pubDate>Wed, 10 Aug 2022 06:33:30 GMT</pubDate>
            <atom:updated>2022-08-10T06:33:30.234Z</atom:updated>
            <content:encoded><![CDATA[<p>The impact of AI tools like DALL•E will be felt for years to come. Learn who will be affected first, and what it means going forward.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*n8ak0CLt_kzUZ-lld6D5Lw.png" /><figcaption>“a scene of intense beauty…” excerpt of stable diffusion prompt used to create this image</figcaption></figure><p>AI tools are beginning to have a tremendous impact on our world, and the speed at which this change is occurring is catching many people by surprise. The past few years have seen enormous strides being made in the field of generative AI models like DALL•E, with new advances being made in rapid succession, and now these tools are starting to become available to the general public.</p><p>While their long-term influence will be significant, AI tools are going to impact different markets at different times. Some have already been affected, and others won’t see the full impact for years to come.</p><p>Today we’ll be taking a look at the short-, medium-, and long-term impacts that co-creative AI tools will have on the world, who these changes will impact the most, and when.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qEDgTsXFWutt8h3MJCJbxg.png" /><figcaption>“Beauty itself. Award winning model staring directly at the camera studio portrait…” DALL-E prompt excerpt</figcaption></figure><h4>Short-Term Impact</h4><p>Generative AI art might still be in the early days, with the wider public only recently gaining access to some of these tools. Until now, the tools have mostly been relegated to academic research settings, where the artistic quality of the output hasn’t necessarily been a major consideration during prompt construction. As a wider variety of artists gain access to these tools, we’re starting to get a better idea of the limitations and potential with these models, and making the first tentative steps towards thinking about what the future of AI-driven co-creative tools might look like in practice.</p><p>As such, we now have a better idea of what they can do right out of the box, with very little effort made toward prompt optimization — and we know which industries will be immediately affected by generative AI art tools as a result.</p><h4>Who sees the most short-term impact?</h4><ul><li>Creative Artists of all sorts</li><li>Stock Photography sites</li><li>Prompt Engineers</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZVeuKbEcQFAH3J3s2Ek1gw.png" /><figcaption>“…studio portrait of a beautiful woman staring directly into your eyes…” — excerpt of DALL-E prompt used to generate this image</figcaption></figure><p><strong>Creative artists of all sorts</strong></p><p>Digital artists, photographers, writers, and creatives of all stripes will be affected by the adoption of AI tools like GPT-3, DALL•E, and Stable Diffusion. Not only will these tools introduce all sorts of new competition into already-crowded fields, they so skew the balance of time vs quality in terms of output that it will be nearly impossible for artists to compete with them on their own. Widespread adoption of these tools among creative professionals might come more from competitive necessity than anything else.</p><p><strong>Stock Photography</strong></p><p>Stock photography companies operate at volume, and will immediately be impacted by AI tools like DALL•E. Traditional stock photography companies make their profits by acting as centralized storefronts for a vast number of independent photographers looking to make a living a few cents at a time. Stock photography sites typically pay a very small amount per sale to the artist, but turn around and charge many multiples of that to anyone looking to license the image for their newsletter or website.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yPkjUCoKn9amYYYxqsMnkw.png" /><figcaption>“hotel room decorated in the theme of Vincent Van Gogh’s “Starry Night”, award winning photograph”</figcaption></figure><p>Generative AI art tools like DALL•E allow anyone to become their own stock photographer, getting the exact image they want at a fraction of the traditional cost, assuming they’ve got a good in-house prompt engineer, at least.</p><p><strong>Prompt Engineers</strong></p><p>The field of prompt engineering might be relatively new, but it will be increasingly important with regards to these co-creative AI tools, and we’ll likely see some specialization and focus on prompt optimization within models like DALL•E and Stable Diffusion as a new generation of creative professionals learns how best to use these new tools. Until then, there will likely be increased demand for competent prompt engineers to do the necessary work required for early-stage exploration of this emerging field.</p><h4>Medium-Term Impact</h4><p>In the medium-term, AI tools will be integrated into apps, with some of the more idiosyncratic behavior of AI models like DALL-E abstracted away into UI elements (for example, including “award winning” in a DALL-E prompt significantly improves the quality of the output, and can be stacked throughout a prompt to greater effect. In practice, it’s incredibly annoying to repeatedly type it out, and the sooner it’s abstracted away into a UI element the better) rather than being something directly visible in the prompt itself. This could be accomplished by creating a user interface that allows blocks of text to be tagged with different criteria (such as art style, artist skill level, emotions to evoke, and so on), which could then be applied automatically to the full prompt that’s sent to DALL-E without the the end user needing to be aware of the fact that “award winning” appears in no less than 7 times in their prompt.</p><p>We’re already starting to see some of this with code integrations from OpenAI Codex and GitHub Copilot, both of which assist programmers in writing their code. While the number of supported programming languages and IDEs remains somewhat limited, more languages are being added every day, a trend that’s not likely to reverse any time soon. Likewise, writers are already able to work with various AI tools to help brainstorm ideas, summarize text, or act as a virtual editor.</p><h4>Who sees the most medium-term impact?</h4><ul><li>Programmers</li><li>Writers</li></ul><p><strong>Programmers</strong></p><p>Programmers are probably the group that will adapt most naturally to co-creative AI tools, as a number of automation and code-completion solutions already exist in the daily-use toolkit of the modern developer. The biggest advantage will be held by those who are able to use AI tools to augment their existing skillset, opening up new avenues that may have previously been inaccessible to them.</p><p><strong>Writers</strong></p><p>Professional writers will benefit greatly from AI tools, which can help act as a side-by-side editor and co-author, able to expand on topics, consolidate larger blocks of text into tl;dr’s, and even generate text at different reading levels. Likely impacts will be seen at large editorial shops, where fewer staff writers and editors may be needed, as well as with ghostwriters, who will soon find a solid ally in AI when it comes to slogging through the boring parts of the writing process, able to concentrate on the intent of the piece while the AI can act as fact-checker or fact-fetcher, able to cross-reference concepts and ideas across a wide range of sources and data sets.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0X2zsptW-z8cN676ZGlFzw.png" /><figcaption>“Award winning 7mm wide lens studio portrait of a dark empty room. Photorealistic view from windows of spring garden, providing only light in the scene…” — except from DALL-E prompt used to generate this image</figcaption></figure><h4>Long-Term Impact</h4><p>In the long-term, co-creating with AI tools will be the norm, and will likely be built into the social media layer as part of a shared canvas that feeds back into itself, where the art we co-create with AI tools is used to train future generations of those same tools.</p><h4>Who sees the most long-term impact?</h4><ul><li>All of us, as we start participating in this shared creative canvas</li></ul><p>There is a huge amount of potential in AI tools, both as a new means of creation as well as communication.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_GDHj9WQPAWq9NG1fZiuvQ.png" /><figcaption>“GPT-3’s photograph is a stunning work of art that captures the autumnal beauty of the woods. The road made of fallen leaves is a clever and original touch” — DALL-E prompt used to generate this image</figcaption></figure><p>There is no reason to think that these impacts will be limited to still images. A gif is just a series of still images played in rapid succession. A movie is simply a gif with a soundtrack. So if you take a look at where things stand today, it’s not hard to see the huge impact that AI tools are about to have on our world.</p><p>Buckle up, it’s going to be a wild ride.</p><p><em>Next week, we’ll cover the darker side of these AI tools, and the negative consequences we’ll face from their widespread adoption.</em></p><p>Nicholas Ptacek is a veteran writer and technologist, with close to 20 years of experience in the cybersecurity industry building award-winning computer security software. His work has been featured extensively in print and news media, including CNNMoney, Macworld, and MacDirectory magazine, along with numerous press appearances in publications including The Information and Vice.</p><p>Nicholas has been documenting his journey as he co-creates art with GPT-3, DALL-E, and Stable Diffusion. You can follow his AI experiments on Twitter at: <a href="https://twitter.com/nptacek"><strong>@nptacek</strong></a></p><p><em>This is essay 3 of 4 for </em><a href="https://thenetworkstate.com/about-1729"><em>1729</em></a><em> Writers’ 2nd Cohort</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fbadea6f9ba1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI Art Lessons: Setting the Scene]]></title>
            <link>https://nptacek.medium.com/ai-art-lessons-setting-the-scene-6c357f17a7ab?source=rss-12be3008a256------2</link>
            <guid isPermaLink="false">https://medium.com/p/6c357f17a7ab</guid>
            <category><![CDATA[dalle-2]]></category>
            <category><![CDATA[tutorial]]></category>
            <category><![CDATA[generative-art]]></category>
            <category><![CDATA[ai-art]]></category>
            <category><![CDATA[openai]]></category>
            <dc:creator><![CDATA[@nptacek]]></dc:creator>
            <pubDate>Wed, 03 Aug 2022 13:15:57 GMT</pubDate>
            <atom:updated>2022-08-03T13:15:57.193Z</atom:updated>
            <content:encoded><![CDATA[<p>Generative AI Art tools like DALL-E are about to usher in a new paradigm for the world of design. Here’s one workflow to explore as you get started on your AI art journey.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*G58yIiDvvQ04uYifaaRkZA.png" /></figure><p>Creating great-looking art using AI tools like DALL-E is nearly effortless, as long as you don’t really care what the output looks like ahead of time. It can be as simple as asking for “Award winning painting, “Neon Starry Night” by Vincent Van Gogh. Award winning digital art” and boom, you get something that looks like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*p1yVRMf3mPl011_7tbzqig.png" /><figcaption>Award winning painting, “Neon Starry Night” by Vincent Van Gogh. Award winning digital art</figcaption></figure><p>If we want to create more complicated scenes, it’s not as easy as just describing the scene as we want it to look — that can quickly overwhelm DALL-E with too many specific details, leading to complicated pieces that leave something to be desired when it comes to the final composition.</p><p><strong>Co-Creating a Piece of AI Art, Step-By-Step</strong></p><p>In order to build a scene that both looks good and maintains internal cohesion with the artist’s vision, we need to build a prompt that reinforces the aspects of the piece that we want DALL-E to focus on. Think of the canvas as a little world you create from scratch, using your prompt to craft the scaffold for the virtual scenery.</p><p>You control every aspect of the scene, so think like a virtual movie director: what type of lighting does your scene have? What type of color palette? Architectural theme? Mood? Style?</p><p>You can request specific thematic elements from movies, games, and traditional art, along with real-world specifications like focal length, media type, location, camera placement, and mood. Resources like the <a href="https://aesthetics.fandom.com/wiki/List_of_Aesthetics">List of Aesthetics</a> can help set the scene.</p><p>If you can think of it, and fill in enough details to make it sound natural, DALL-E will just roll with it:</p><p>Tell DALL-E it’s a hovering drone camera, and the output takes on an overhead angle. By combining multiple styles into a single prompt, we can get an end product that’s evocative and still tells a story.</p><p><strong>Setting the Scene: Size and Scale</strong></p><p>Let’s start by setting a scene. I enjoy evocative art pieces, especially ones that can help provide a bit of calm in this busy world, so that’s the vibe I’m going to go for as a starting point; it will naturally influence the overall mood of the piece as a result.</p><p>Here’s a good starter prompt:</p><p><em>A dark empty room. Photorealistic view from windows of spring<br>garden, providing only light in the scene. Award winning studio<br>photography by DALL-E</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lYQimSsJCPjrBu9Z0XXBHQ.png" /><figcaption>A dark empty room. Photorealistic view from windows of spring<br>garden, providing only light in the scene. Award winning studio<br>photography by DALL-E</figcaption></figure><p>We can play with the proportions of the room by specifying the size as <em>vast</em>:</p><p><em>A vast dark empty room. Photorealistic view from windows of<br>spring garden, providing only light in the scene. Award winning<br>studio photography by DALL-E</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SC5VPN20rXP6fuC9CV4hmg.png" /><figcaption>A vast dark empty room. Photorealistic view from windows of<br>spring garden, providing only light in the scene. Award winning<br>studio photography by DALL-E</figcaption></figure><p>I’m envisioning something really open:</p><p><em>An enormous dark empty room. Photorealistic view from<br>windows of spring garden, providing only light in the scene.<br>Award winning studio photography by DALL-E</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yPuH5xZWWLYUZLGJZKqnCQ.png" /><figcaption>An enormous dark empty room. Photorealistic view from<br>windows of spring garden, providing only light in the scene.<br>Award winning studio photography by DALL-E</figcaption></figure><p>This is our own little creative world, so we can change things up as we see fit.</p><p>I’ve always dreamt of a loft workshop:</p><p><em>A dark empty enormous loft workshop. Photorealistic view<br>from windows of spring garden, providing only light in the<br>scene. Award winning studio photography by DALL-E</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ep0Ljhek1UUK7Rw-8h_daw.png" /><figcaption><em>A dark empty enormous loft workshop. Photorealistic view<br>from windows of spring garden, providing only light in the<br>scene. Award winning studio photography by DALL-E</em></figcaption></figure><p>That’s nice and open, but for this piece I want the vibe to be one more focused on art…let’s change it to a creative workspace instead:</p><p><em>A dark empty enormous loft artist workspace. Photorealistic<br>view from windows of spring garden, providing only light in the<br>scene. Award winning studio photography by DALL-E</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CcavI3MZ5jC9mzkfKDyvDA.png" /><figcaption><em>A dark empty enormous loft artist workspace. Photorealistic<br>view from windows of spring garden, providing only light in the<br>scene. Award winning studio photography by DALL-E</em></figcaption></figure><p>That prompt got the vibe right, but it messed with the sense of scale and proportion…can we reword it to put the focus back on the openness of the overall space?</p><p>The answer is yes!</p><p><strong>Setting the Scene: Elements of Focus</strong></p><p>It turns out that word placement and proximity within the prompt structure matters a great deal when it comes to the final composition. Simply moving the placement of the words over by a few characters can have a huge impact on the perceived focus of the piece from DALL-E’s point of view, and these changes can influence a piece in subtle ways that really add to the overall magic.</p><p>Change the prompt to:</p><p><em>A dark empty enormous loft workspace for a digital artist.<br>Photorealistic view from windows of spring garden, providing<br>only light in the scene. Award winning studio photography by<br>DALL-E</em></p><p>This keeps the focus on the loft itself, and also goes to show that DALL-E was reading the previous prompt as a “loft artist” workspace, i.e. focusing on a blended concept due to my choice of word placement.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PWMjcmWP2o2OPjaiXxYLaA.png" /><figcaption><em>A dark empty enormous loft workspace for a digital artist.<br>Photorealistic view from windows of spring garden, providing<br>only light in the scene. Award winning studio photography by<br>DALL-E</em></figcaption></figure><p>Understanding the power of word placement, repetition, and thematic reinforcement within a prompt is key to creating beautiful AI Art while still being able to maintain creative control over the final piece to a large degree.</p><p><strong>Setting the Scene: Stacking Themes</strong></p><p>To create beautiful, natural looking AI Art, you need to stack themes and reinforce them throughout the prompt. Clever wordplay and turns of phrase can make a huge difference when working within the 400-character limit for DALL-E prompts, and there is a vast area of exploration to be had within the Unicode character set.</p><p>Changing our prompt to:</p><p><em>A dark empty enormous loft workspace for a digital artist witch<br>practicing the arts of cybermagic. Photorealistic view from<br>windows of spring garden, providing only light in the scene.<br>Award winning studio photography by DALL-E</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LoUSPmR8vOMPTFjvaXRJoA.png" /><figcaption><em>A dark empty enormous loft workspace for a digital artist witch<br>practicing the arts of cybermagic. Photorealistic view from<br>windows of spring garden, providing only light in the scene.<br>Award winning studio photography by DALL-E</em></figcaption></figure><p>By stacking together two themes, we get a subtle witchy vibe to our scene.</p><p>Personally, I’d like a bit more of a witchy theme going on here. And just like before, we can use the word placement in our prompt to subtly influence DALL-E’s focus and interpretation. This is incredibly tricky in practice, and very much a learn-as-you-go type thing, because as you change the word placement you change the prompt structure itself – finding the right balance in your prompt wording so the focus of the piece remains true to your artistic vision is an art in and of itself!</p><p>Change the prompt to:</p><p><em>A dark empty enormous loft workspace for a witch digital artist<br>practicing the arts of cybermagic. Photorealistic view from<br>windows of spring garden, providing only light in the scene.<br>Award winning studio photography by DALL-E</em></p><p>And the influence shifts from more of a digital-artist-who-happens-to-be-a-lil-witchy vibe to a witch-who-happens-to-be-a-digital-artist vibe. The difference might sound subtly when worded in text, but check out the effect it has on the scene!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CnmYPS5IhgbmN6OUR0uVfQ.png" /><figcaption>A dark empty enormous loft workspace for a witch digital artist<br>practicing the arts of cybermagic. Photorealistic view from<br>windows of spring garden, providing only light in the scene.<br>Award winning studio photography by DALL-E</figcaption></figure><p>One of the most interesting aspects of DALL-E is the possibility for prompt optimization, and there are a number of us slowly mapping out prompts that provide exceptionally good output on a reliable basis, and which can act as frameworks for further exploration and artistic creation.</p><p><strong>Setting the Scene: Working with a Prompt in Flux</strong></p><p>Slight changes to the scene can have a huge impact on the final piece — I wanted to include a tiny element to the scene with this prompt change, but it totally changed the vibe of the entire piece:</p><p><em>A dark empty enormous loft workspace for a witch digital artist<br>practicing the arts of cybermagic. Her familiar floats serenely<br>in the room, glowing with phosphorescent rainbow light.<br>Photorealistic view from windows of spring garden, providing only light in the scene. Award winning studio photography by DALL-E</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*05P565fKZioH62mNKz8jrw.png" /><figcaption><em>A dark empty enormous loft workspace for a witch digital artist<br>practicing the arts of cybermagic. Her familiar floats serenely<br>in the room, glowing with phosphorescent rainbow light.<br>Photorealistic view from windows of spring garden, providing only light in the scene. Award winning studio photography by DALL-E</em></figcaption></figure><p>Hmm…so as pretty as that output is, it appears DALL-E misinterpreted what I meant by “familiar,” so let’s try rephrasing the prompt:</p><p><em>A dark empty enormous loft workspace for a witch digital artist<br>practicing the arts of cybermagic. The witch’s familiar is a<br>rainbow-haired cat. The familiar is fast asleep floating in<br>midair. Photorealistic view from windows of spring garden, providing only light in the scene. Award winning studio photography by DALL-E</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9zZLwEa3mGD92aRhK57Wew.png" /><figcaption><em>A dark empty enormous loft workspace for a witch digital artist<br>practicing the arts of cybermagic. The witch’s familiar is a<br>rainbow-haired cat. The familiar is fast asleep floating in<br>midair. Photorealistic view from windows of spring garden, providing only light in the scene. Award winning studio photography by DALL-E</em></figcaption></figure><p>Well, we have a cat this time, but it’s not floating. This part of the process feels a bit like debugging code.</p><p>Sometimes you get lucky and come up with something that works brilliantly, but you have to get DALL-E to believe it would make sense in context. Clearly I didn’t try hard enough here:</p><p><em>A dark empty enormous loft workspace for a witch digital artist<br>practicing the arts of cybermagic. The digital artist’s pet cat<br>has antigravity boots. Photorealistic view from windows of<br>spring garden, providing the only lighting in the scene. Award winning studio photography by DALL-E</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*U890AzBoSA_YkKLaFEIUaA.png" /><figcaption><em>A dark empty enormous loft workspace for a witch digital artist<br>practicing the arts of cybermagic. The digital artist’s pet cat<br>has antigravity boots. Photorealistic view from windows of<br>spring garden, providing the only lighting in the scene. Award winning studio photography by DALL-E</em></figcaption></figure><p>I mean, maybe the anti-gravity boots work and the just doesn’t care. That would be very catlike behavior.</p><p><strong>Setting the Scene: Taking Behavior Into Account</strong></p><p>When you tell DALL-E that something is happening in a scene, it makes its best guess as far as what you want, as seen with this prompt:</p><p><em>A dark empty enormous loft workspace for a witch digital artist practicing<br>the arts of cybermagic. The cat hovers in midair, suspended by the power<br>of the witch’s magic cyberspell. Photorealistic view from windows of<br>spring garden, providing the only lighting in the scene. Award winning studio photography by DALL-E</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Isp7EOy9WwN6mPDuOF4bBw.png" /><figcaption><em>A dark empty enormous loft workspace for a witch digital artist practicing<br>the arts of cybermagic. The cat hovers in midair, suspended by the power<br>of the witch’s magic cyberspell. Photorealistic view from windows of<br>spring garden, providing the only lighting in the scene. Award winning studio photography by DALL-E</em></figcaption></figure><p>According to our prompt, the cat is in midair, and from DALL-E’s point of view, a cat sitting on a desk counts as midair, a view that’s reinforced by the fact that that’s totally normal catlike behavior. It wasn’t what we were asking for, but it was much more in line with what DALL-E considers something normal and expected to be found within the framework of the creative context we’ve provided.</p><p><strong>Setting the Scene: Evoking Feelings</strong></p><p>We have control over the atmosphere of the scene, and can style the request so that DALL-E create a piece that evokes specific emotions.</p><p>Change the prompt to:</p><p><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene. Designed to evoke a feeling of wonder and awe</em></p><p>And we can see how DALL-E adds subtle effects to the overall composition of the piece.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZpLQ4Byil3ucJ4MQvHZjcQ.png" /><figcaption><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene. Designed to evoke a feeling of wonder and awe</em></figcaption></figure><p><strong>Setting the Scene: Adding Motifs</strong></p><p>We can add whole swathes of visual cues at once by taking advantage of media references like so:</p><p><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene. Blade Runner motifs throughout</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zmKSo3qdS6hVU6IZcN3leQ.png" /><figcaption><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene. Blade Runner motifs throughout</em></figcaption></figure><p>This is an easy way to adopt entire stylistic themes and elements in one go, while maximizing use of prompt space. It’s possible to remix and mashup themes from a wide variety of sources that DALL-E is aware of, and even possible to bring themes out in a piece without referencing them directly simply by stacking the correct styles in the prompt.</p><p>Tron motif Prompt:</p><p><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene. Tron motifs throughout</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*titcnjYz3u34w3fbxUz4_A.png" /><figcaption><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene. Tron motifs throughout</em></figcaption></figure><p>Vaporwave motif Prompt:</p><p><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene. Vaporwave motifs throughout</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EC3tHo1VfEcRY8_CYBiKlw.png" /><figcaption><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene. Vaporwave motifs throughout</em></figcaption></figure><p>Cybermagic motif prompt:</p><p><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene. Cybermagic motifs throughout</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VuyrHOlmIEW4G72cfxmvfw.png" /><figcaption><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene. Cybermagic motifs throughout</em></figcaption></figure><p>Introspective motif prompt:</p><p><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene. Introspective motifs throughout</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rwU6t2T1an8SgWxlbvHwog.png" /><figcaption><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene. Introspective motifs throughout</em></figcaption></figure><p><strong>Setting the Scene: Requesting Specific Items</strong></p><p>It can sometimes be hard to get the exact vibe you’re going for, and I’ve found it helpful to ask for specific items that can be representative of larger concepts.</p><p>I wanted to show that the witch was a bit messy and gets so lost in her works of creativity that the mundane tasks can sometimes get lost in the shuffle.</p><p>How can we phrase our prompt and get something like this?</p><p>Let’s give it a try…</p><p><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. The witch needs to pick up her 📚 and laundry. Photorealistic view from windows of spring garden, providing the only other lighting in the scene.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0YbnPJra8RDZfOvvRg6JpA.png" /><figcaption><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. The witch needs to pick up her 📚 and laundry. Photorealistic view from windows of spring garden, providing the only other lighting in the scene.</em></figcaption></figure><p>Well, the room is a bit more cluttered, but it’s still pretty orderly looking. Let’s try and refine the prompt further:</p><p><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. A basket of dirty laundry in the corner. Photorealistic view from windows of spring garden, providing the only other lighting in the scene.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gVq7DEF4P_CcAC2yuoOB7g.png" /><figcaption><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. A basket of dirty laundry in the corner. Photorealistic view from windows of spring garden, providing the only other lighting in the scene.</em></figcaption></figure><p>That’s the look I was going for!</p><p>Finding the specific prompt that generates the exact effect we want feels a bit like deciphering an unknown language, except you get to use the medium of art as a Rosetta stone to gain insight into DALL-E’s interpretation of it’s input. It’s an absolutely fascinating puzzle, and one that I think will be part of the lasting appeal of co-creating art with DALL-E — it feels like solving a really satisfying puzzle when the prompt finally clicks.</p><p><strong>Setting the Scene: Digital Decorations</strong></p><p>If we want to refine the style of our scene, we can reinforce part of the original prompt and expand on our witch’s personality even further –– this added dimension allows DALL-E to extrapolate an appropriate look for our witch’s elegant taste profile:</p><p><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. The witch likes elegant beauty. Photorealistic view from windows of spring garden, providing the only other lighting in the scene.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Z8KAnNvLn9O-7L96iq0uig.png" /><figcaption><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. The witch likes elegant beauty. Photorealistic view from windows of spring garden, providing the only other lighting in the scene.</em></figcaption></figure><p><strong>Setting the Scene: Musical Vibes</strong></p><p>We can further set the scene by adding a musical dimension to the piece. While we might not be able to hear the music playing, we can see it’s subtle influence through visual cues:</p><p><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Triphop music plays quietly in the background. Photorealistic view from windows of spring garden, providing the only other lighting in the scene.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JIcwKnKgDKYGUsttrO3ibQ.png" /><figcaption><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Triphop music plays quietly in the background. Photorealistic view from windows of spring garden, providing the only other lighting in the scene.</em></figcaption></figure><p>Trance music prompt:</p><p><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Trance music plays quietly in the background. Photorealistic view from windows of spring garden, providing the only other lighting in the scene.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ex_-CnDadiq5YZSD_L7cuQ.png" /><figcaption><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Trance music plays quietly in the background. Photorealistic view from windows of spring garden, providing the only other lighting in the scene.</em></figcaption></figure><p><strong>Setting the Scene: Specifying a Purpose</strong></p><p>We can specify the purpose of our virtual scene, by describing it as a place where art is created:</p><p><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene. This is where the witch goes to create her art.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Rh_3OKrDpYndOHEekb3j_g.png" /><figcaption><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene. This is where the witch goes to create her art.</em></figcaption></figure><p>Since we’re reinforcing the concept of “art” in this updated prompt, it has the added effect of providing more of a painting look to the finished image.</p><p>As a source of deep magical insight:</p><p><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene. This is the source of her deep magical insight.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0G99fIoS4LOTOFZyFvUooQ.png" /><figcaption><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene. This is the source of her deep magical insight.</em></figcaption></figure><p><strong>Setting the Scene: Using Emoji</strong></p><p>Remember when I said we could make use of Unicode characters in our prompts? Well, Emoji are part of the unicode character set, and they provide a really intuitive framework to reinforce singular concepts in our prompts, especially when we run out of room and need to get an idea across in as few characters as possible. Bird app aficionados understand this feeling well.</p><p>Here, we’ll ask for a cityscape at dusk motif:</p><p><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene. This is the source of her magic. Motifs: 🌆</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*C_ZpPrX1IpkQE6FZ_GHzPA.png" /><figcaption><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene. This is the source of her magic. Motifs: 🌆</em></figcaption></figure><p>Night with stars emoji motif:</p><p><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene. This is the source of her magic. Motifs: 🌃</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NyIC64tcA7mhZFavHJiddg.png" /><figcaption><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene. This is the source of her magic. Motifs: 🌃</em></figcaption></figure><p>Adding a fireworks motif gives the magic spell a little extra flourish:</p><p><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene. This is the source of her magic. Motifs: 🎆</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*luguEhle9XCT07zx3h_YsA.png" /><figcaption><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene. This is the source of her magic. Motifs: 🎆</em></figcaption></figure><p><strong>Setting the Scene: Weather Effects</strong></p><p>We can change the weather of the scene using emoji:</p><p><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene. This is the source of her magic. Weather: 🌪</em></p><p>Note how the curtains are blowing, and the weather looks outside the windows matches the prompt. I absolutely love the atmosphere evoked here.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*u0KYQcjDp-D9VkXFvRs7iw.png" /><figcaption><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. Her magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene. This is the source of her magic. Weather: 🌪</em></figcaption></figure><p><strong>Setting the Scene: Paying Homage to Influences</strong></p><p>One of the things I love most about DALL-E is its ability to help seamlessly blend unique combinations of aesthetic influences into a finished piece. During the creation of this tutorial, I was thinking about some of my favorite creators and artistic influences, especially <a href="https://twitter.com/crypto_coven"><strong>@crypto_coven</strong></a> and all the web3 witches I’ve met in this space over the past year. Paying homage to our heroes is one of the paths of continuity in the artistic world, one that remains alive and well in the age of AI-assisted Art.</p><p>And sometimes, paying homage to our influences makes all the difference.</p><p>I still felt that DALL-E’s output was missing something, and thought more about the type of witch who I was envisioning in my mind. I decided that she would be an oscilloscope artist like <a href="https://twitter.com/katecursed"><strong>Kate the Cursed</strong></a>, who makes some of the most unique art I’ve ever seen:</p><p><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. A neon oscilloscope sits in the corner. The witch’s magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dXOBIQhE84S0F4RQM-CCYw.png" /><figcaption><em>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. A neon oscilloscope sits in the corner. The witch’s magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene.</em></figcaption></figure><p>There comes a certain point in the generative act of creation with AI Art where you know any further tweaks to the prompt will take away from rather than add to the final product, and it’s time to stop, and share the piece with the world.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TrxPVQpWNvG4hdTv8FOQOQ.png" /><figcaption>A dark empty enormous loft workspace for a witch digital artist practicing the arts of cybermagic. A neon oscilloscope sits in the corner. The witch’s magical spell surrounds a glowing crystal floating in midair. The focus is on the overall evocative feeling of the creative space of a digital artist. Photorealistic view from windows of spring garden, providing the only other lighting in the scene.</figcaption></figure><p>I’m really satisfied with how this particular piece turned out, and I think it’s time I let you in on a little secret:</p><p>I’ll bet you didn’t even notice, but we took off the “award winning art” training wheels about halfway through the tutorial. Once you construct a scaffold prompt and start filling in the details like you’ve been learning to do throughout the course of this tutorial, you’ll learn how to phrase a prompt that holds up on its own, without the need to specify that it’s “good” art outright.</p><p>I hope you’ve enjoyed following along with this tutorial, and that it inspires you to start experimenting with AI art. I’ve found the following resources to be helpful during my own experiments and explorations, and highly recommend exploring them thoroughly:</p><p><strong>Essential resources for your AI art journey</strong></p><p><a href="https://openai.com/dall-e-2/">DALL•E 2 Waitlist Signup</a></p><p><a href="https://docs.google.com/document/d/11WlzjBT0xRpQhP9tFMtxzd0q6ANIdHPUBkMV-YB043U/">DALL•E 2 Prompt Engineering Guide</a></p><p><a href="https://matthewmcateer.me/blog/clip-prompt-engineering/">Clip Prompt Engineering for Generative Art</a></p><p><a href="https://dalle2.gallery/">DALL•E 2 Image Database</a></p><p><a href="https://aesthetics.fandom.com/wiki/List_of_Aesthetics">List of Aesthetics</a></p><p><a href="http://xahlee.info/comp/unicode_index.html?q=">Unicode Search</a></p><p><a href="https://magazine.artland.com/art-movements-and-styles/">The Most Famous Art Movements and Styles</a></p><p>Nicholas Ptacek is a veteran writer and technologist, with close to 20 years of experience in the cybersecurity industry building award-winning computer security software. His work has been featured extensively in print and news media, including CNNMoney, Macworld, and MacDirectory magazine, along with numerous press appearances in publications including The Information and Vice.</p><p>Nicholas has been documenting his journey as he co-creates art with GPT-3 and DALL-E. You can follow his AI experiments on Twitter at: <a href="https://twitter.com/nptacek"><strong>@nptacek</strong></a></p><p><em>This is essay 2 of 4 for </em><a href="https://thenetworkstate.com/about-1729"><em>1729</em></a><em> Writers’ 2nd Cohort</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6c357f17a7ab" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI is here, and it changes everything.]]></title>
            <link>https://nptacek.medium.com/ai-is-here-and-it-changes-everything-106e25706246?source=rss-12be3008a256------2</link>
            <guid isPermaLink="false">https://medium.com/p/106e25706246</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[dalles]]></category>
            <category><![CDATA[dalle-2]]></category>
            <category><![CDATA[ai-art]]></category>
            <dc:creator><![CDATA[@nptacek]]></dc:creator>
            <pubDate>Wed, 27 Jul 2022 07:26:25 GMT</pubDate>
            <atom:updated>2022-07-27T08:07:16.862Z</atom:updated>
            <content:encoded><![CDATA[<p>Here’s what you need to know.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LU_lv2CCM3RMPnrhGPlCRg.png" /><figcaption>Someone typed the phrase “hello world” with an IBM Selectric typewriter. Award winning photograph — Prompt used to generate this image with DALL-E</figcaption></figure><p>Technology exists to ingest the old, integrate it, and give birth to new paradigms. This has been the case ever since our ancestors first learned to wield fire, and has continued on down through the ages, embodying both violence and beauty in this continuous act of consumption and renewal.</p><p>The invention of the printing press was one such turning point, the personal computer another, and the Internet a third. We find ourselves in a world on the edge of yet another of these transformations, with widespread public access to AI tools just around the corner.</p><p>And, as always happens when human ingenuity gives birth to new technology, society is utterly unprepared for what is about to be unleashed. In this four-part series I’ll bring you up to speed on what’s headed our way with AI, including the potential it holds, the limitations it has, and the flaws you should be aware of. Throughout the series, I’ll use real-world examples from my own explorations of GPT-3 and DALL-E to highlight what’s possible with the technology as it exists today, and what’s currently out of reach.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fX2nMIFFfBPG4mdQE9n-xg.png" /><figcaption>Award winning 3D render of a gigantic robot striding toward a tiny village made up of laptop computers. It has an air of menace, and should take place as the sun is setting. — DALL-E prompt used to generate this image</figcaption></figure><p>I’ll cover what’s possible in a co-creative future: co-creative with both AI and with one another, participating in a giant shared creative layer where it just might be possible to learn how to communicate better with one another.</p><p>I’ll cover the pitfalls, where AI falls short and training data is biased, where it’s not a panacea but a poison––where it can be weaponized through intentionally malicious training data sets or used to create disinformation in an age that can scarce afford it.</p><p>While platforms like OpenAI take pains to avoid contributing to the disinfo epidemic, not all AI platforms have such restrictions, and I’ll talk about what that means in a world of institutional failure, where fake news rules the day and the concept of consensus reality has itself come increasingly under fire.</p><p>I’ll cover the ways in which this new technology will impact careers in many different sectors, and how AI tools will affect creative artists in particular.</p><p>Finally, I will discuss the ways in which both blockchain technology and copyright reform fit into the overall picture.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8nDWeJZHIrDh7ES42BryXw.png" /><figcaption>Does knowing the DALL-E prompt that was used to generate a piece of art affect its perceived beauty?</figcaption></figure><p><strong>It calls into question what constitutes “good” art.</strong></p><p>When we can generate something stunning by simply appending “award winning” to the prompt and then calling it a day, it changes a lot of things, and brings a lot of questions immediately to mind.</p><p>What does that mean for the person who has spent the better part of their life training to do the same thing by hand?</p><p>What does that mean for the uncredited artists whose body of work constitutes the training set of many of these AI tools?</p><p>What does this mean for all of us once we inevitably join in and start participating in a co-creative artistic experience, where everything we create is endlessly riffed on, changed, picked apart, remixed, inverted, and ultimately reingested by the very AI models that helped us generate art in the first place?</p><p>What does the future look like if this creativity is unlocked for everyone, when those without the privilege or means to go to art school can create art on par with those who can?</p><p>What does this mean for creative artists of all stripes, who are about to face an avalanche of competition in an age where discovery is already one of the hardest problems they face?</p><p>And yet amidst all these questions, there’s also opportunity. The stage is set for a creative renaissance the likes of which we’ve never seen, and we are about to have the opportunity to co-create on a level that has been previously unheard of.</p><p>We can think of tools like DALL-E as our way of interacting with a giant, shared canvas, one that allows people who may never have had a chance to express themselves artistically to do so with the full fidelity of the canvas in their mind’s eye, while also putting art back out into the world to allow others the same opportunity.</p><p>And art, in a very real sense, is how we communicate with each other in a way that transcends language and nationality, in a way that tears down old boundaries and demarcations in order to build something new.</p><p>Tools like DALL-E allow us to copy art, yes, but they also allow us to view familiar works of art from a different angles, and in a new light.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xTf0Pb6dxWsf8Odggs0yQw.png" /><figcaption>Tools like DALL-E allow us to reimagine classic works of art as more diverse expressions of humanity</figcaption></figure><p>Everything is a mashup.</p><p><a href="https://www.everythingisaremix.info/watch-the-series">Everything is a remix</a>.</p><p><a href="https://writingcooperative.com/steve-jobs-good-artists-copy-great-artists-steal-4fc6593ac09">Good artists copy, great artists steal</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MXBv4VeTVJkaIZ88c0SdzQ.png" /><figcaption>Award winning studio portrait by Annie Leibovitz of Starry Night by Vincent Van Gogh — DALL-E prompt used to generate this image</figcaption></figure><p>Nicholas Ptacek is a veteran writer and technologist, with close to 20 years of experience in the cybersecurity industry building award-winning computer security software. His work has been featured extensively in print and news media, including CNNMoney, Macworld, and MacDirectory magazine, along with numerous press appearances in publications including The Information and Vice.</p><p>Nicholas has been documenting his journey as he co-creates art with GPT-3 and DALL-E. You can follow his AI experiments on Twitter at: <a href="https://twitter.com/nptacek"><strong>@nptacek</strong></a></p><p><em>This is essay 1of 4 for </em><a href="https://thenetworkstate.com/about-1729"><em>1729</em></a><em> Writers’ 2nd Cohort</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=106e25706246" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Building with ZORA API]]></title>
            <link>https://nptacek.medium.com/building-with-zora-api-a6d06e2eb6ff?source=rss-12be3008a256------2</link>
            <guid isPermaLink="false">https://medium.com/p/a6d06e2eb6ff</guid>
            <category><![CDATA[canvasjs]]></category>
            <category><![CDATA[web3]]></category>
            <category><![CDATA[tutorial]]></category>
            <category><![CDATA[zora]]></category>
            <category><![CDATA[graphql]]></category>
            <dc:creator><![CDATA[@nptacek]]></dc:creator>
            <pubDate>Fri, 24 Jun 2022 07:22:32 GMT</pubDate>
            <atom:updated>2022-06-24T08:28:32.999Z</atom:updated>
            <content:encoded><![CDATA[<p>In this tutorial you’ll learn how to build a simple NFT sales dashboard using Zora’s powerful new API.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9nx-bc4pwE1dC-D0MO5WYw.jpeg" /></figure><p><strong>Note: This tutorial uses code written in GraphQL, JavaScript, and Objective-C, but I will try and address the major implementation details at an abstract level, as they are not language-specific.</strong></p><p>I recently participated in <a href="https://nptacek.medium.com/zora-api-hackathon-in-review-bb20f336aed0">Zora’s API Hackathon</a>, and have been building with it pretty much nonstop ever since–it’s easy to pick up, even for GraphQL novices, but also lets you go really deep when needed. This makes for an API that’s accessible yet powerful, which isn’t something you see every day. It’s been perfectly suited to the task of NFT analytics, and I’ve been having a lot of fun exploring the capabilities of the API over the past couple weeks.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4-ks0H71ezyENmFqNGb7fA.png" /><figcaption>Early experiment with ZORA API to get info on a specific NFT and display it in a basic UI</figcaption></figure><p>More recently, I’ve been experimenting with plotting NFT sales data on scatter charts, which has led to all sorts of new possibilities for visualizing different types of signals that might not otherwise be apparent. I’m really excited with where my research is heading, and can’t wait to share more about it in the future. In the meantime, all of this work with the Zora API has led to random thoughts like <em>“I wonder if I could visualize recent Zorbs sales, showing each Zorb with the correct color?”</em></p><p>Last night I challenged myself to find out. This tutorial is the end result.</p><h4>Getting Sales Data</h4><p>Let’s start by figuring out what data we’ll need if we want to display recent Zorbs sales on a scatter chart. At a bare minimum, we’ll need two pieces of data in order to plot a point: the <strong>timestamp</strong> for the sale, and the <strong>sale price</strong>.</p><p>Here’s the GraphQL query that will fetch that data from the Zora API:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Bkalr7d7xtkfRiXt9VgkdA.png" /><figcaption>requesting the bare minimum</figcaption></figure><p>Now, in practice we’re going to want to request more information from Zora’s API than just that. Keep in mind that the goal is not only to display recent Zorb sales, but to do so in a way that indicates the color of each Zorb. Don’t worry about <strong><em>how</em></strong> we’ll display the data just yet. For now, we’re just in the planning stage. What other data might we want?</p><p>I can immediately think of two pieces of data:</p><ul><li>Token ID</li><li>Token Metadata</li></ul><p>The token ID will come in handy during testing and troubleshooting — if we’re in any doubt as to the accuracy of the data, we can always use the token ID to cross-reference the transaction data on etherscan.io.</p><p>The token metadata is where we’ll find information on the Zorb’s display image, which changes based upon wallet address.</p><p>Further, we can set some constraints on our GraphQL query — by limiting it to the most recent 500 sales, we can create a one-shot query, and won’t need to worry about handling multiple pages of results. This is not only kinder on Zora’s API servers, it keeps things simpler on the code end of things as well. Remember, we can always expand our program’s capabilities in the future!</p><p>So, what does our final ZORA API query look like?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*yF-6hqkW6rbVmGS4Vl8Yqw.png" /><figcaption>Zora API query for the 500 most recent Zorb sales</figcaption></figure><p>We can verify the output for our query on <a href="https://playground.api.zora.co/">Zora’s API Playground</a> and get an idea of the format of the data we’ll be working with:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zBJj6zyx7AZ8WmTyKwQ2-g.png" /><figcaption>output for our Zora API query</figcaption></figure><p>It looks like all the information we requested is there, but what about that big blob of data for the Zorb’s image? How are we supposed to get from that to a specific Zorb color to display? Before we get to the specifics, it’s time for a…</p><h4>SVG SEGUE</h4><p>Some NFTs store their data completely on-chain, making use of the Scalable Vector Graphics (SVG) file format to achieve a small data footprint while still packing in a lot of potential when it comes to the image being stored. I <a href="https://nptacek.medium.com/sometimes-all-you-have-to-do-is-ask-2e8b0558ecd">covered this topic in detail back in October 2021</a> if you’re interested in further reading on the subject. If you’re familiar with the topic already and just want to get on with the tutorial, keep reading for the speed run version.</p><h4>Deciphering the Data</h4><p>Ok, so how do we get from that blob of image data to individual Zorb colors?</p><p>We know from the <a href="https://en.wikipedia.org/wiki/Data_URI_scheme">data URI scheme</a> that the image data is base64-encoded, so let’s decode it to see what it looks like. You can do this in code, but there are <a href="https://www.base64decode.net/">some websites where you can test it out</a> as well. Just don’t forget to strip out the <em>data:image/svg+xml;base64,</em> prefix if you’re going with the website route, or it won’t decode correctly.</p><p>Here’s what the base64-decoded image data looks like:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KuCMy5GlSR3V9zCWuh5t6g.png" /><figcaption>decoding base64 image data</figcaption></figure><p>If you copy and paste the output into a plain-text .html file and then open it with your web browser, you should see the Zorb in all its glory:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kNaqj21BP27zf-ZynBkXEQ.png" /><figcaption>a Zorb reconstituted from on-chain metadata</figcaption></figure><p>So, we’re getting closer, at least we can see how the image is constructed for any given Zorb. There is actually a ton of interesting code to dig into if you’re interested in the technical implementation of the Zorb color-changing side of things. Take a look at the <em>gradientForAddress</em> function in the <a href="https://etherscan.io/address/0xca21d4228cdcc68d4e23807e5e370c07577dd152#code">Zorbs smart contract</a> if so.</p><p>For the purposes of this tutorial, however, we’re keeping things simple, and we can analyze the SVG output without actually needing to understand the implementation details. The only thing we care about is identifying the main display color for each Zorb. Looking at the SVG output, we can see five references to something called <em>stop-color</em>, each with an <em>hsl</em><strong> </strong>value. HSL stands for Hue, Saturation, Luminosity, and is a way of representing specific colors. Bingo.</p><p>We can verify that the HSL values we see in the decoded svg image data match the colors of our Zorb on a site like <a href="https://hslpicker.com/">HSLPicker</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*afsS1FPjq-aEW84rtMsLsg.png" /><figcaption>hsl color visualizer</figcaption></figure><p>There are a couple things to note here: some Zorbs are more than one color, others aren’t. It’s all based on the wallet address holding the Zorb. It’s up to you how you want to visually convey the Zorb’s color–for example, you might want to calculate the median of all five HSL values, and use that for the display color. For the purposes of this tutorial, however, we’re simply going to stick with the value that corresponds to the lower left of each Zorb image (i.e. the last <em>stop-color</em> value).</p><p>Before we continue, let’s take a look at where we’re at:</p><ol><li>We know how to ask the Zora API for recent Zorbs sales data.</li><li>We know how to decode the SVG image data for any Zorb.</li><li>We know how determine the HSL values that correspond to a Zorb’s color.</li></ol><p>So we’ve got all the data we need to plot any given Zorb sale in the correct color, where do we go from here? How do we actually display this data in a meaningful way?</p><h4>Displaying the Data</h4><p>If you’re not opposed to working with third-party libraries, <a href="http://canvasjs.com/">canvasJS</a> offers a number of highly customizable options, and will work perfectly for the purposes of this tutorial. I ended up going for the Multi Series Scatter/Point Chart, as it allows marker customization on a per-entry basis, perfect for the multitude of Zorb colors we’ll need to handle. Even better, it can handle HSL values right out of the box, so you don’t have to worry about converting the color to RGB or hex.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hZE-zzAwm4ffbAUQgyYiGQ.png" /><figcaption>we’ll have some more data conversions to do before we’re through</figcaption></figure><p><a href="https://canvasjs.com/docs/charts/basics-of-creating-html5-chart/date-time-axis/">Looking at the documentation</a>, it appears we’ve got a couple options when it comes to handling the timestamps for Zorb sales. We can either construct a new JavaScript Date object, or we can specify a timestamp in milliseconds. Date conversions are notoriously tricky things, especially when you start accounting for time zones and locales, but we don’t need to go that far in this tutorial. Converting a timestamp from the format we get from Zora’s API to one we can use with canvasJS is as simple as getting the Unix timestamp and multiplying it by 1000 to get a timestamp in milliseconds. Passing that along to canvasJS is all we need to do, canvasJS handles the rest.</p><p>You’ll definitely want to dig around in the canvasJS documentation, there are tons of customization options that I’m only scratching the surface of here.</p><p>Here’s one way of setting it up:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*sZ46xSnEbzOYSp4hcKCFjg.png" /><figcaption>basic canvasJS template to display recent Zorb sales</figcaption></figure><p>I added a few basic design touches, such as setting the chart background color to black, and giving each Zorb sales entry a white outline to make it stand out a bit. When you mouse over a data point, a tooltip will show the sale date, price, and token ID. Some test data allows us to get an idea of what it’ll look like once we start charting data points:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_HkpLqBAMrZkAEJC09tekg.png" /><figcaption>one is the zorbliest number</figcaption></figure><p>So, now that we know what data we need to get from Zora’s API, how it’s formatted, and the format we need it to be in for canvasJS, we’re tasked with the matter of actually getting the data from the Zora API, parsing it, and sending it to canvasJS.</p><h4>Bringing it all together</h4><p>I’ve implemented this part in Objective-C, but you can use whatever programming language you’re most comfortable with — plenty of libraries and frameworks exist allowing you to call GraphQL directly from JavaScript.</p><p>In this case we’ll be using an extended version of <a href="https://github.com/nptacek/Ponder">Ponder</a>–my entry for the Zora API Hackathon–to power the interactions with Zora’s API server.</p><p>First, we’ll request the data we want from Zora’s API server using a GraphQL query. Once we get a response back from the server, we’ll parse the data a bit in order to get it into the correct format to use with canvasJS. This mostly involves base64 decoding and slicing up the resulting strings to grab the HSL value for each Zorb sale, and keeping track of it along with some other basic sales data, then passing it along to another function for further parsing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*FJeuowO_gvE0IrMeB1Zksg.png" /><figcaption>obj-c code to grab the latest Zorb sales data</figcaption></figure><p>Next, we create our canvasJS template directly in code, setting the dataPoints to our previously created array of reformatted sales data. We’re also including a custom handler for click events on individual Zorb sale markers. This allows us to send the tokenId back to objective-c, where we can use it to open a webpage directly to that particular Zorb. This makes verification of sales data and Zorb color a snap.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WveJdrg2C6L8Fs-C-04l_A.png" /><figcaption>obj-c code to programmatically create a canvasJS template and populate it with Zorb sales data</figcaption></figure><p>When we put it all together, here’s the final result:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2U7v7zKZD371P_Zfqu65cQ.png" /></figure><p><strong>Note: You may find Zorb display colors don’t match up with what you see on OpenSea or other sites. Try refreshing the metadata and you should see the correct colors. You might notice price discrepancies across various marketplaces, always verify data on-chain when possible.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qnilTGE2tfohiq00w9DzvQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BBXCQHVhHzh3Yk3K0tvn-w.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*GVpYcsiAPO_1XmKlbql0QQ.png" /><figcaption>a price discrepancy on OpenSea compared with what’s on-chain</figcaption></figure><p>So, there you have it. Using the new Zora API, it’s definitely possible to keep track of recent Zorb sales by color. This tutorial only scratches the surface of what’s possible when you combine Zora API with canvasJS. You can find the tutorial code over on github if you want to play around with it yourself. I’d love to see your creations, please let me know what you come up with!</p><p>Grab the source code over at GitHub:</p><p><a href="https://github.com/nptacek/zorb_sales_visualizer">GitHub - nptacek/zorb_sales_visualizer: a recent zorb sales visualizer built with ZORA API and canvasJS</a></p><h4>References</h4><p>zorb_sales_visualizer source code: <a href="https://github.com/nptacek/zorb_sales_visualizer">https://github.com/nptacek/zorb_sales_visualizer</a></p><p>Zora API Reference: <a href="https://docs.zora.co/docs/zora-api/intro">https://docs.zora.co/docs/zora-api/intro</a></p><p>Zora API Playground: <a href="https://playground.api.zora.co/">https://playground.api.zora.co</a></p><p>canvasJS Documentation: <a href="https://canvasjs.com/docs/charts/basics-of-creating-html5-chart/">https://canvasjs.com/docs/charts/basics-of-creating-html5-chart/</a></p><p>canvasJS scatter point chart demo: <a href="https://canvasjs.com/javascript-charts/multi-series-scatter-point-chart/">https://canvasjs.com/javascript-charts/multi-series-scatter-point-chart/</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a6d06e2eb6ff" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[ZORA API Hackathon In Review]]></title>
            <link>https://nptacek.medium.com/zora-api-hackathon-in-review-bb20f336aed0?source=rss-12be3008a256------2</link>
            <guid isPermaLink="false">https://medium.com/p/bb20f336aed0</guid>
            <category><![CDATA[zora]]></category>
            <category><![CDATA[web3]]></category>
            <category><![CDATA[hackathons]]></category>
            <category><![CDATA[nft]]></category>
            <category><![CDATA[api]]></category>
            <dc:creator><![CDATA[@nptacek]]></dc:creator>
            <pubDate>Sat, 18 Jun 2022 03:30:12 GMT</pubDate>
            <atom:updated>2022-06-18T03:30:12.417Z</atom:updated>
            <content:encoded><![CDATA[<p>ZORA recently <a href="https://twitter.com/ourZORA/status/1532071848090882053">released a new API</a> (application programming interface), which enables anybody to query their indexer for NFT market data, and held a hackathon to give it a good <a href="https://en.wikipedia.org/wiki/Shakedown_(testing)">shakedown</a> run. By encouraging a bunch of people to hammer away at their API server, ZORA was able to monitor and stress-test their systems under high load, which is the type of real-world testing you need to perform when shipping any product at scale.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rZtkodaEx5E9rHDXayOy-w.png" /></figure><p>The hackathon ran for four days, and offered prizes of 6.9 ETH, 4.20 ETH, and 1.01 ETH to the top three projects, as voted on by Zorbs holders.</p><p>By offering a generous hackathon prize pool just as the market was melting down and people were scrambling for funding, ZORA all but ensured a strong turnout for the hackathon, and there was a large pool of participants.</p><p>As one of those participants, I spent four days straight running on lots of caffeine, not enough sleep, jamming on code, and more or less having a great time (aside from some annoying slowdowns when it came to the finer aspects of Objective-C’s limitations as a programming language). I’d like to share some thoughts on my hackathon experience, what I learned, and what I’d do differently next time around.</p><p>I’ll also share some thoughts on the ZORA API itself, and how I see it playing a critical role in surfacing on-chain data and metrics that might otherwise escape our attention. This type of analysis provides insight into overlapping communities within the NFT ecosystem, and by extension offers an entirely new data set to explore for those seeking deeper market understanding.</p><p>Finally, I’ll share my wishlist for future API endpoints, and some thoughts on improvements that could be made to the hackathon voting process.</p><h4>Creative Goals and Unexpected Detours</h4><p>I had initially planned on creating a more full-fledged NFT analytics demo, showcasing not only collection data such as number of holders and sales volume, but also things like an individual token’s image, attributes, rarity, and sales data. Instead, I was struck by inspiration partway through the hackathon…which turned into more of a fun puzzle to follow than anything else. You know how it goes.</p><p>In the end, I made the decision that what I had accomplished in getting native Objective-C code to work with GraphQL without any third party frameworks or libraries was enough of a success in its own right, and left my attempt at some rough on-chain data analysis in as a secondary feature of sorts. So what strange puzzle distracted me from my primary objective?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*o2CXdLoC9mBZBIRi9-Z0AQ.png" /></figure><p>I was pondering one of the goals of the hackathon (bringing utility to Zorbs holders), combined with a question surfaced by <a href="https://twitter.com/xuannu_eth/status/1534971203181498368">xuannu.eth</a>, and decided I would like to know the top shared communities across Zorbs holders, where “shared communities” would be represented by the tokens held across the most wallets, ranked in descending order.</p><p>I knew in principle the steps I would need to take to get that data — first, I’d need to get a list of all token holders for a given project. Easy enough, Zora offers that as one of the API endpoints.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AzSRmuldQ-d8B6s762TTJw.png" /><figcaption>graphyql query to get the top 10 Zorbs holders</figcaption></figure><p>From there, I would need to go through each wallet address returned by the query, and get a list of the tokens it contains. In this particular instance, I didn’t care as much about the number of tokens a wallet held for any given collection, so instead of keeping track of each individual token, I merely kept an overall list of unique contracts for tokens in a wallet, which is the minimum amount of data I would need to make the shared community calculations.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/872/1*euk5XDMwacjNJv-tNrPt3g.png" /><figcaption>query to get the contents of my wallet</figcaption></figure><p>After repeating this process for every token holder for a given smart contract, it’s possible to calculate which tokens are held across the widest number of wallets, and by extension calculate the largest communities of token holders (for other projects) that exist within a collection. How do we go about doing so?</p><p>For each unique <em>collectionAddress</em> held by a given wallet address, add it to a larger data collection (in this case, I used NSCountedSet, an Objective-C data structure that allows you to keep track of the number of times a given piece of data has been added to a unique set). From there, it’s easy to see which <em>collectionAddresses</em> are found amongst the largest number of wallets, and thus see which “communities” exist within the original collection that we’re analyzing.</p><p>While I wanted to calculate the top shared communities for Zorbs holders, I knew that calculating that data would take a significant amount of time (assuming the bare minimum of one query needed per Zorb holder, that would be ~33,700 GraphQL requests), and wanted to make sure my algorithm worked correctly on a smaller collection first, so I chose one of my long-time favorites, crypto_coven, to act as a testing ground.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/924/1*_eg5RcnSnAZqQV3FWXXGDA.png" /><figcaption>a somewhat-accurate measure of the top shared communities among crypto_coven holders</figcaption></figure><p>I could immediately see that there were some flaws with my calculations. First, the count was off. The top community for crypto_coven holders should be crypto_coven itself, which was correct, but the count didn’t match the number of crypto_coven holders. There were issues with the API server returning bad gateway errors from time to time throughout the hackathon, so I wrote this off a mix of that plus potential bugs in my own code. The numbers were still good enough to make some rough calculations, in any event.</p><p>I encountered a different issue while looking up info on the top communities (something which I would have liked to automate, but ran out of time before being able to implement) — spam tokens were taking up a majority of the top slots. For example, here are the top shared communities for crypto_coven holders:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/884/1*1Y5KWw_47FgAG8ZEd0xw6w.png" /></figure><p>Some platforms such as etherscan and OpenSea flag or hide spam tokens in order to protect users from potential phishing attempts, so a lot of the time it’s easy to forget they are even in our wallets. As far as I could tell, there was no easy indicator within Zora’s API to seen if a token was flagged as spam or not. From a decentralization point of view, this lack of censorship could be seen in a positive light. At the same time, it’s incredibly helpful to be able to easily and quickly establish guardrails for those who need them, and making it easy for product creators to filter out or flag spam tokens would go a long way towards making the space safer for everyone.</p><p>If I had more time during the hackathon, I would have liked to implement some sort of spam filtering mvp, likely built off of cross-referencing transfers and mint/sales events, and filtering out tokens that were airdropped to users — this would essentially be the first step in triaging potential candidates for further spam/phishing analysis. If nothing else, it would remove some noise from the program output, even if it caught some innocent airdrops in the process. Alas, this exercise in programming logic had to remain nothing more than the seed of an idea for lack of time.</p><h4>Mistakes Made and Lessons Learned</h4><p>Coming from the cold barren wastelands of web2 and thus being unfamiliar with GraphQL prior to the release of ZORA’s API, I was at a bit of a disadvantage and had to learn as I went, distilling the bare minimum needed to accomplish my goals in a short time period.</p><p>Objective-C isn’t the best language to use for things like this, and has its own caveats to be aware of — for example, there’s no built-in way to rate-limit URL connection requests, so I built a really basic rate limiter myself — only, I got the timing wrong.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1QoNG0g14-d_IiWk0dhELA.png" /><figcaption>running batches of 10 wallets at a time, every 2 seconds. 300 requests/min instead of 30. oops 😅</figcaption></figure><p>I realized after the fact that this code absolutely <strong>hammers</strong> on ZORA’s API server; in my haste to start building, I misread 30 requests per <strong>minute</strong> as 30 requests per <strong>second</strong> and never looked back. I did have to keep throttling back the rate limit on my requests during testing, but I never managed to make the connection to double-check the project specs until after the hackathon was over — I just figured that ZORA’s servers were under some strain from all the other hackathon participants.</p><p>My apologies for any headaches this caused the ZORA engineering team during the hackathon, it was completely unintentional on my part! At least you know your API servers can handle heavy traffic? 😅😅😅</p><p>Another mistake was waiting until the last minute to slap together a UI. I should have taken the time to get a simple, clean UI implemented <strong>first</strong>, rather than assuming I’d have time to tackle it before the end of the hackathon. Instead, I ended up with the Kevin of hackathon UI’s.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*W6lZq3qwP_0sWaOrKj6muA.png" /><figcaption>the kevin of hackathon submissions</figcaption></figure><p>I didn’t get a chance to optimize my code, and it’s incredibly inefficient from a memory usage standpoint — bottlenecks in the code are causing surges in the GraphQL queries being sent to the API server, resulting in unnecessary 502’s and eventually 429’s as it hammers the ZORA servers. I had originally planned on commenting and documenting my code more extensively, but much like a decent UI, that plan also fell by the wayside. <a href="https://archive.org/details/mythicalmanmonth00broo">Programming always takes longer than you think it’s going to</a>, no matter how confident you are in your assumptions.</p><h4>Thoughts on the ZORA API</h4><p>Coming from the world of web2 where so much still runs on REST, Zora’s API using GraphQL is a breath a fresh air.</p><p>I’ve tried out a bunch of different web3 APIs on the backend side of things over the past 9 months, and so far the ZORA API is my favorite one by far, for the simple reason that GraphQL is going to win out over traditional RESTful APIs in my book any day of the week after this hackathon.</p><p>ZORA did a great job providing easy access to commonly used data points and aggregate data such as <em>ownersByCount</em> and floor prices, with the ability to filter down to just the data you need, and nothing more (the <em>lookbackHours</em> feature for sales volume is an especially nice touch)</p><p>Doing the same type of collection analysis using the OpenSea API is much more complicated, requiring multiple RESTful queries (and all of the associated juggling and parsing of data structures) to get the same information that can be had with a single GraphQL query on ZORA API.</p><p>That being said,<strong> I think the ZORA API could be even better</strong>.</p><p>Here’s what I’d like to see:</p><h4>API Wishlist</h4><p>First and foremost, I’d like to see parity across marketplaces as far as the data provided by the ZORA API goes. While sales data from markets like OpenSea is accessible through the ZORA API, OpenSea listings are not currently included when calculating the <em>floorPrice </em>aggregate stat. Likewise, data sources other than mainnet are not yet included in results.</p><p>I would like to see some sort of status flag for tokens that have been marked as spam on platforms like etherscan, in order to make it easier to filter them out of results. This will save a lot of people headaches in the long run.</p><p>After realizing my mistake regarding the API rate limit, it quickly became apparent that my original idea for calculating the top shared communities among Zorb holders would become infeasible at scale. Instead, I’d like to see ZORA offer some of these things that are inefficient to calculate on the client end of things as data points themselves.</p><p>For example, I’d like to see an easier way to get a list of holdings per wallet on a unique-contract-only basis. For some analytics purposes, I care less about the exact number of holdings someone has for a specific project, but rather just want to know if they are holding the project at all.</p><p>Further, I’d like to see a convenience accessor for this type of data on a per-project basis — something that returns the data you’d get if you crunched the numbers for every individual holder — so you’d be able to easily identify the top 10 shared communities among tokenholders.</p><p>This data serves a broad need in the space for community-building, marketing, and identifying emerging trends, and is all sitting there on chain just waiting to be surfaced.</p><p>On a broader level this serves as a form of optimization on the backend infrastructure — providing convenience accessors for these types of queries would be much more efficient than tons of third party clients hitting the ZORA API server every day to grab the data necessary to make these calculations on their own.</p><h4>Voting Issues</h4><p>Unfortunately the hackathon was marred by some issues when it came to the <a href="https://snapshot.org/#/zora.eth/proposal/QmSxyU8ar8CpoDotginqoxoGMnxvVTFKpJ6u4Zb6Uqr3Cy">Zorbs holder vote</a> on the top three projects, which provides an opportunity to examine what went wrong, and brainstorm how similar situations might be avoided in the future.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kUkWkKepL_l95HZzG3XMsw.png" /></figure><p>The initial snapshot vote <a href="https://twitter.com/ZORAEngineering/status/1536490843628089345">had to be scrapped</a> due to some hackathon submissions showing up multiple times under slightly different names, causing confusion. This was more of a QA issue than anything else, but I wonder if there were any Zorbs holders who participated in the initial (scrapped) vote and missed out on the second one.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Ly0jlaXL0pIg1hkriUfV-A.png" /></figure><p>Snapshot doesn’t allow randomization of candidates on ballots, and projects were listed in the order they were submitted — ballot position can have an effect on vote outcome as pointed out by <a href="https://twitter.com/valcoholics1">@valcoholics1</a> during the hackathon wrapup on Twitter Spaces, so projects that were submitted earlier in the hackathon had an advantage. This particular issue has been <a href="https://github.com/snapshot-labs/snapshot/discussions/1518">raised previously in Snapshot discussions</a>, but so far no progress has been made to address it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*eVWnBwNK7op22S_bpwL-YA.png" /></figure><p>It wasn’t initially clear that Zorbs holders only got one vote, no matter the number of Zorbs held. That was supposed to be the intent, as shown by the <a href="https://docs.snapshot.org/strategies/what-is-a-strategy">erc20-with-balance voting strategy</a> adopted for this vote.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/812/1*By2Wjrk-M8I2b7a49FuP2A.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nl5cIlZEgOLi6KZ1Rm8p3g.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*056DnsyxoOgK-62ZsmxeKA.png" /></figure><p>Except… you <em>could</em> select multiple projects when voting, and Snapshot would then assign a vote to each one. Some people realized they could select more than one candidate, but others (myself included) clearly did not. This was due to setting the voting system to Approval voting rather than Single choice voting. Clearly the winning strategy here would have been to have everyone vote for every candidate, thus creating a 20-way tie and splitting the entire prize pool of 12.11 ETH amongst all hackathon participants. Ah well.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/682/1*Urd3RG7yaNeg6v-KjhV-SA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KJwgbJaZHW7ZzAYQzqFI1g.png" /></figure><h4>TL;DR</h4><p>Overall, I had a net positive experience with the ZORA API hackathon, and found the API easy to use, even as a complete newcomer to GraphQL. I’d like to see the API continue to grow, with more indexer and market data being made available to developers in the future. More clarity during the voting process would have been appreciated.</p><h4>References:</h4><p><a href="https://api.zora.co/">https://api.zora.co</a></p><p><a href="https://playground.api.zora.co/">https://playground.api.zora.co</a></p><p><a href="https://github.com/nptacek/Ponder">https://github.com/nptacek/Ponder</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bb20f336aed0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What comes next for web3?]]></title>
            <link>https://nptacek.medium.com/what-comes-next-for-web3-95191e9bcc50?source=rss-12be3008a256------2</link>
            <guid isPermaLink="false">https://medium.com/p/95191e9bcc50</guid>
            <category><![CDATA[crypto]]></category>
            <category><![CDATA[web3]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[bear-market]]></category>
            <category><![CDATA[terra]]></category>
            <dc:creator><![CDATA[@nptacek]]></dc:creator>
            <pubDate>Fri, 13 May 2022 07:21:22 GMT</pubDate>
            <atom:updated>2022-05-13T07:36:52.546Z</atom:updated>
            <content:encoded><![CDATA[<p>Here are my tips for founders and startups building over Summer 2022.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*M7s0a2b0B7Hh0ZMct9b5Jw.jpeg" /></figure><p>No matter what happens next, it’s clear that we’re entering a new phase of the crypto market, one which has significant business ramifications for anyone who got started building during the last bull run. Market dynamics are set to change right at a time that new startups need to put their heads down and get serious about building. Many startups will be faced with the unappealing prospect of an existing customer base clamoring for results at a time of restricted cashflow, and many a roadmap will need a realistic reassessment in light these changes, something we’re already seeing with many existing NFT 10k projects.</p><p>Companies unlucky enough with their timing to have an unreleased product in the pipeline that’s not yet ready for prime time will likely be faced with a more difficult time finding customers in the coming months, as existing crypto users cast a more wary eye to new startups and potential new entrants to the crypto space are put off by the radioactive meltdown of Terra’s collapse. Of course, this market pattern will ease with time, but founders who only have experience building in most recent crypto bull run can take immediate steps to batten down the hatches and make their runway survive as long as possible. By playing it safe and maybe a bit more slow than planned, you can survive and even thrive during a bear market.</p><p>Here’s how I would play it going into Summer 2022:</p><ol><li>Scale back where possible</li><li>Keep customers in the loop</li><li>Build in public</li><li>Focus on slow, concentrated growth</li></ol><p>Let’s break it down.</p><h4>Scale back where possible</h4><p>There are a few different areas you can scale back in, depending on your company’s size and position in the market.</p><p>Founders are going to hate me for saying this, but scaling back on parties and other major events is one of the better ways to cut costs. Yes, they serve as advertising and community building events, but if you truly want your startup to survive a bear market you can’t be blowing money on something that isn’t directly focused on product development or sales — so be selective in how you approach them. Rather than throwing a party by yourself, see if you could help sponsor one instead. Oftentimes larger brands will offer different sponsorship tiers with varying benefits, so be on the lookout for ones that provide product or brand placement opportunities. Send two or three of your people with the strongest grasp on the product and business plan to major events (and accompanying parties) with networking specifically in mind. It might not be as much of an ego boost as throwing a party by yourself, but rest assured that those will come along with company success. Don’t burn through your capital living it up too soon.</p><p>Unless you have a pre-existing community, scrap those plans for a discord server for now, stick with Twitter. It will be less overhead on support staff, and creating a public space invites both moderation headaches in the form of spam and hacking attempts as well as online abuse and people FUDing your project. If you’re not at the point where you need a public discord server yet, it will be more of a drag on your company than anything else in the early days. One way to stay in touch would be via Twitter Communities, which is a good alternative for startups that need to communicate more often than a newsletter and less often than on discord or telegram.</p><p>When it comes to employees, there will be somewhat of a natural self-selection process at work as people who were really only building with you in the hopes of a quick moon ditch the project, with those who really want to see it succeed (and willing to put in the work to do so) sticking around. That being said, you may still need to assess your staff composition unless you’ve got a lean, well-oiled machine of a team that’s already working well together as it is. In that case, don’t change a thing.</p><p>Think a few months down the line. You’re not going to launch as soon as you think you are, despite your best efforts. That’s just the way these things go, and that goes doubly in a bear market. So what changes would you make to your staff under a longer, slower timeline to launch than planned? This doesn’t necessarily mean you need to let people go, it could be the case that you can find some other aspect of the business for them to focus on in the mean time. In the event that you do have to let people go, keep in mind that unexpected layoffs can have a depressing effect on staff morale, so try to do what you can to make the process as transparent as possible. Layoffs should occur simultaneously so it doesn’t feel like hunger games for your remaining employees.</p><p>If you haven’t done so already, sit down with your dev team and ask them their honest opinion on which planned features are realistic within the current product development timeframe, and which ones are slipping further behind and would be best put off until a future version. Be realistic, and decisive, then go with the streamlined feature set and cut the rest. You dev team will be grateful and the remaining features are more likely to work well and be polished as a result of refined focus. (This is actually something that is good to do frequently throughout the product development lifecycle, not just in bear markets. Trimming excess growth is what keeps your product streamlined and moving towards market.)</p><h4>Keep customers in the loop</h4><p>Protecting an existing customer base during a bear market is one of the most difficult things for a small startup to get right. Lost customers are hard to win back, and more likely to spread FUD about your project. Oftentimes customers are much more forgiving and lenient when it comes to unexpected delays and changes in plans than you’d expect — as long as you’re open in your communication with them.</p><p>Existing customers who were promised plans that now seem sky high will be upset when you can’t follow through on your promises. The best way through is to approach it head on. You might be tempted to put your head down and just keep building, trying to avoid rocking the boat one way or the other, but this usually leads to an undercurrent of dissatisfaction that can be hard to shake once picked up by the community at large. It’s better, then, to be as clear with customers as you can when it comes to changes to the product, timeline, or both. Be sure to think things through as much as possible before involving your larger customer community — nothing is worse than a poorly planned Twitter Space.</p><p>Depending on the size of your customer base, this might present an opportunity to gather feedback in relation to roadmap changes, but it’s very easy for a divided Twitter Space to morph into a mob scene. Be smart in how you approach things, and read the room. Be able to clearly lay out the problem, the steps you’re taking in the short term, the steps you’re taking in the long term, and how you’ll be keeping your customer base in the loop going forward. By providing information on the problem, your plan of attack, and how you’re planning on staying accountable, you provide reassurance that you’re going to keep building, and that it’s safe for your customers to keep trusting you.</p><p>I caution new founders to take the above point very seriously. Your business reputation is everything, and you absolutely cannot approach changes in business direction without thinking things through ahead of time. Customers (former or not) FUDing your project hurts not only your company image, it also hurts your development team’s morale. People often forget that there are humans behind these projects. Nobody wants to feel like the project they’re working on isn’t worth it, so make sure to insulate your team where you can.</p><h4>Build in public</h4><p>One of the easiest ways to remain accountable to your customer base is to simply build in public, as much as possible. Be transparent in your setbacks as well as your successes, share screenshots of what you’re working on, and start fostering community growth. There’s a fine line between keeping people in the loop and spamming them, however — if you don’t have anything new to report, don’t lean too heavily into engagement farming. Remember, this is the time you’re supposed to start proving to your customers that they were right to put their faith in you. Your earliest supporters can turn into lifelong fans when you treat them like people and not product.</p><p>Keep in mind that you don’t have to be posting new screenshots every day to provide value to your customers, there are other things you could do as well.</p><p>Contribute back to the larger crypto community where possible — not everything has to be product. It could be contributing to how-to guides, highlighting the latest developments in your neck of the crypto woods, or otherwise fostering a shared community mentality through something unique such as weekly Twitter Spaces with a common theme (I’ve enjoyed a couple of very nice meditation ones that have been offered by various projects over the past few months).</p><h4>Focus on slow, concentrated growth</h4><p>Whatever projections you made for user growth during the bull market need to be updated. New founders are going to experience a period of whiplash as they transition from a wild west gold rush mentality with customers flocking to literally any random new project they could to one where it’s much harder to win customers over. Your marketing strategy might need an overhaul. Now is not the time for billboards and other extraneous adspends. Instead, this approach requires more time spent doing market research to identify potential customers.</p><p>There is a ton of data sitting on chain right now that could be used to help streamline this approach, and I expect an increased focus on community building and community outreach over the next few months as projects learn that there can be strength in interoperability without leverage (see the entire Loot Project ecosystem, mathcastles/hypercastle explorer, nouns, and blitmaps projects for example). Cooperative, sustained growth will still play an important role going forward, but it’s going to look different than what happened in DeFi during the run-up to the Terra crash.</p><h4>Looking toward the future</h4><p>No matter the state of the market overall, I continue to remain bullish on many of the projects coming out of the web3 space, especially the ones that have been quietly building this whole time. I’m excited for new hardware advancements that are coming to the AR/VR side of things, which is an area of interoperability with NFTs that has been sorely undervalued by investors and builders alike, but which should be on everyone’s radar as new hardware starts to hit the market. Some things need to be built that don’t necessarily provide direct value to a company but are rather public goods for all that help improve the space. There is a continuing need to implement guardrails for those new to the crypto space in the form of better tools, better protocols, and better documentation. As always, user education remains in dire need of assistance in this space.</p><p>Each cycle in the crypto market follows a familiar pattern, starting with an initial gold rush period where any and everything looks good, followed by a downtempo period of building wherein the true value is actually created. This will be the time when we see startups survive and thrive or wither and fail. If you’ve made it this far, congratulations. You’ve survived the initial trials in becoming a startup. Now it’s time to see how well you do at becoming a successful startup.</p><p>I can’t wait to see what you build.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=95191e9bcc50" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>