<?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 Tyler on Medium]]></title>
        <description><![CDATA[Stories by Tyler on Medium]]></description>
        <link>https://medium.com/@tylermakesgames?source=rss-b6b8d3bf25ce------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*ZPiwjbr9s9DoH5YaC_opUg.png</url>
            <title>Stories by Tyler on Medium</title>
            <link>https://medium.com/@tylermakesgames?source=rss-b6b8d3bf25ce------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 02:06:54 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@tylermakesgames/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[AGI is the Multitool]]></title>
            <link>https://medium.com/@tylermakesgames/agi-is-the-multitool-24f6a0492c9a?source=rss-b6b8d3bf25ce------2</link>
            <guid isPermaLink="false">https://medium.com/p/24f6a0492c9a</guid>
            <dc:creator><![CDATA[Tyler]]></dc:creator>
            <pubDate>Tue, 18 Apr 2023 18:18:00 GMT</pubDate>
            <atom:updated>2023-04-18T18:18:00.000Z</atom:updated>
            <content:encoded><![CDATA[<h3>and that’s why it won’t matter… at first</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/1*LWsh0C0gvoxAVwQfXZhj5w.png" /><figcaption>Check your junk drawer, I bet you have one</figcaption></figure><p>Do you have a Multitool? Around here, they are better known as Leatherman or Swiss army knives.</p><p>When do you use it? If you need scissors, do you go for the multitool? How about when you need a screwdriver? Is it your go to knife?</p><p>For most of us the answer is: not when a better option is available. When you have the availability of the specialized tool, there is little reason to go for the multitool. That’s because its the Jack of All Trades.</p><p><strong><em>Jack of all trades</em></strong> <em>is a master of none, but oftentimes better than a master of one. <br></em>The full quote makes an interesting point- master of none is still usually better than a single expertise.</p><p>Herein lies the initial problem with ‘AGI’, or Artificial General Intelligence as its being described online. Any tool that utilizes our current level of understanding of intelligence and compute power will just be a ‘multitool’. It will have many smaller less effective tools built into one container.</p><p>By the time we see anything that could be considered AGI by current standards, there will be specialized tools that can outperform it at any task. An AGI that can perform logic, draw, write, and do tasks expected of it will be at a severe disadvantage to any AI tool using the same compute power that specializes at any one of these things.</p><p>Is ‘General Intelligence’ even useful for an AI? Do we want it?<br>A big question we should be asking is whether there is any advantage to having AGI in the first place. Beyond intellectual curiosity, what purpose will it serve us?</p><figure><img alt="“A human being should be able to change a diaper, plan an invasion, butcher a hog, conn a ship, design a building, write a sonnet, balance accounts, build a wall, set a bone, comfort the dying, take orders, give orders, cooperate, act alone, solve equations, analyze a new problem, pitch manure, program a computer, cook a tasty meal, fight efficiently, die gallantly. Specialization is for insects.” by Robert Heinlein" src="https://cdn-images-1.medium.com/max/640/1*QpG-x2acdZ6KRnzHy-LNMg.jpeg" /></figure><p>Heinlein’s ‘specialization is for insects’ was intended to mark the superiority of the human form of intelligence, but he might have inadvertently described the issue with AGI as well. Any sufficiently intelligent AI tool that can do these tasks will probably rely on other specialized tools to handle each task. Similar to how humans are comprised of many organs, tissues, and cells, its probable the first AGI is actually just a wrapper on top of hundreds of layers of specialized AI tools.</p><h3>That’s where it gets concerning…</h3><p>An AGI that can access and utilize all of these specialized tools to their fullest will indeed be an intimidating force of change. The ability to access the APIs of specialized tools and utilize any digital resources for its intentions is how we get a <a href="https://en.wikipedia.org/wiki/Instrumental_convergence#Paperclip_maximizer">universe made of paperclips</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*X0LiIMthV9mTg2BNUzm0tg.png" /><figcaption>made in Midjourney</figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=24f6a0492c9a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Guidelines for Education w/ AI Content Synthesis]]></title>
            <link>https://medium.com/@tylermakesgames/guidelines-for-education-w-ai-content-synthesis-9c47f185949e?source=rss-b6b8d3bf25ce------2</link>
            <guid isPermaLink="false">https://medium.com/p/9c47f185949e</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[plagiarism]]></category>
            <category><![CDATA[education]]></category>
            <category><![CDATA[academia]]></category>
            <dc:creator><![CDATA[Tyler]]></dc:creator>
            <pubDate>Thu, 30 Mar 2023 23:29:48 GMT</pubDate>
            <atom:updated>2023-03-30T23:54:41.795Z</atom:updated>
            <content:encoded><![CDATA[<p>First up, <strong>who am I?</strong> <em><br></em>I’m a lecturer at UT Austin in the Fine Arts College, and I’ve been working with AI/ML content creation since 2015. I’ve seen first hand the progress of these tools, I’ve introduced them to students, and I’ve had curriculum guidelines for these tools for two years.</p><p>Secondly, <strong>what is “AI Content Synthesis”?</strong><br>AI Content Synthesis is using AI (machine learning) tools to create content of various mediums. The most popular version of this (as of this writing) are <a href="https://openai.com/research/gpt-4">ChatGPT</a> for text, <a href="https://github.com/features/copilot">Github Copilot</a> for code, <a href="https://stability.ai/blog/stable-diffusion-public-release">Stable Diffusion</a>, <a href="https://openai.com/product/dall-e-2">DALL-E 2</a>, and <a href="https://www.midjourney.com/home/">Midjourney</a> for images, and <a href="https://research.runwayml.com/gen2">RunwayML</a> for video. If you’re not already aware of some of these tools, I recommend following the links above to read on their proficiency.</p><h3>The Four Steps of AI Acceptance in Curriculum</h3><p>And how it relates to the steps of acceptance of grief or change</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*ZkeNrRuYTgkkeQ3HwtFfKQ.png" /><figcaption>The four stages. Image made in Stable Diffusion</figcaption></figure><h3>I— Forbidden</h3><p><strong>We will attempt to prevent you from using these tools.</strong><br> This is the <em>denial &amp; anger stage</em>. Many educators are here.</p><p>Unfortunately, there are no reliable ways to prevent the use of these tools. At a basic level, using text tools for better text structure, rewording, or other minor edits have been available for years (Grammarly) and tools that fully write the text can be easily tweaked to avoid the already unreliable AI text detection tools.</p><h4>Do not rely on ‘AI Detector’ tools as a sole source of evidence for academic dishonesty</h4><p>Ultimately the cat and mouse game of AI content generation and AI content prevention (<a href="https://www.zerogpt.com/">ZeroGPT</a>, <a href="https://glaze.cs.uchicago.edu/">Glaze</a>, etc.) will lead to many false positives that could see students with less technical knowledge being caught, while their more technically savvy counterparts will circumvent detection methods. Any prohibited structures should include additional testing before claiming academic dishonesty. Any attempts to forbid the use of AI tools will require additional testing or additional oversight.</p><p><strong>Additional Note</strong>: Academic dishonesty through the use of purchased essays and papers has been available through various online sources for years. This disproportionately impacts students of low income whereas these AI tools provide the same benefit to all students equally. Any use of detection tools for AI dishonesty should have an equal attempt to find purchased content to create an equitable space for students. <br><em>See </em><a href="https://www.facebook.com/permalink.php?story_fbid=pfbid0wpZQCTFczDknWK6Tkj5GFw4PrF13adShfuvhHCoHPdCCkL4owT3PquafAdQrKQiil&amp;id=12331215"><em>more on this from Ian Schreiber</em></a><em>.</em></p><h3>II — Discouraged</h3><p><strong>We accept that you’ll use these tools and there’s nothing we can do about it<br></strong>We’ve reached the <em>bargaining stage</em>. This is a heavy handed “please don’t” with the implicit understanding there is nothing you can do about it.</p><p><strong>You should tell your students there is nothing you can do about it. </strong><br>If you don’t tell your students, the tech savvy students will be advantaged knowing they can rampantly use AI tools, and you’ll disadvantage those who are unaware that there’s no bite behind the bark.</p><p><strong>Require citations. </strong>They may not be verifiable like a normal citation, but if you accept the tools, you can request student cite their AI tools used and where they used them.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_7K71kUNhsGiC66XxYM53g.png" /><figcaption>My slide citing the GPT portion with color coding and footnote</figcaption></figure><h3>III — Accepted</h3><p><strong>We accept you’ll use these tools and we’re okay with it. <br></strong>We’ve reached the <em>acceptance stage. If you can’t beat ‘em, join ‘em! <br></em>For many educators, this will be the final stage. With knowledge that you can’t stop it, and understanding that discouraging it only harms those that follow the guidelines you create, the inevitable end point is accepting the use of these tools.</p><p><strong>Use it yourself! </strong>At this stage, if you’re accepting use from the students, there’s no reason you shouldn’t use these tools for your own content. This will help you encourage your preferred citation format, as you can guide them on the best practices on utilizing the technology.</p><h3>IV— Encouraged</h3><p><strong>We want you to use it! <br></strong>At this point we’re working AI tools into our curriculum and embracing their use. This is not only accepting the change, but requiring it. I would recommend easing into this, as you may find that some students are resistant to this level of utilization.</p><p><strong>Require an AI collaboration. </strong>Similar to teaching students how to utilize search engines, Wikipedia, a calculator, or any other academic assistance we reach the point where we actively require students to use these tools as part of their process. This will ensure all students have a proficiency with these tools, preventing any disadvantages that would come from lack of exposure.</p><p>Next up I’ll be doing <strong>a post dedicated to encouraging tools </strong>and how I recommend handling this. In the mean time, I encourage checking out <a href="https://medium.com/@rwatkins_7167/updating-your-course-syllabus-for-chatgpt-965f4b57b003">Ryan Watkin’s article</a> on ways to adapt your curriculum for the use of AI tools.</p><p>See you next time!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9c47f185949e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[CC / Public Domain Image Models & why we need ‘Opt-Out’]]></title>
            <link>https://medium.com/@tylermakesgames/cc-public-domain-image-models-why-we-need-opt-out-a381ee8c76ec?source=rss-b6b8d3bf25ce------2</link>
            <guid isPermaLink="false">https://medium.com/p/a381ee8c76ec</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[creative-commons]]></category>
            <category><![CDATA[ai-art]]></category>
            <dc:creator><![CDATA[Tyler]]></dc:creator>
            <pubDate>Sat, 17 Dec 2022 20:34:12 GMT</pubDate>
            <atom:updated>2022-12-17T20:34:12.519Z</atom:updated>
            <content:encoded><![CDATA[<p>The problem with public domain / CC image models and why “Opt-Out” is needed.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/757/1*eok6TerF3Z4GaYX5BHWjrA.png" /><figcaption>Openverse, a site for creative commons images</figcaption></figure><p>I’ve seen a few suggestions that ML tools could ‘just use public domain / creative commons”. And while that sounds like a fair compromise, its important to understand why that isn’t enough, and why an eventual <strong>‘Opt-In’ site for images will need an accompanying Opt-Out system.</strong></p><blockquote><strong>Many copyrighted images are posted in public domain / CC sites incorrectly.</strong></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LxXiWXJaEvSO4Ug9I0v4Tw.png" /><figcaption><strong>A quick search on Openverse finds many of Simon Stålenhag’s works</strong></figcaption></figure><p>So here we have the problem- <strong>Whose responsibility is it to find and remove these copyright images</strong>? <br>Creative Commons doesn’t have the resources to manually curate and wouldn’t be able to verify if they did. They provide the framework for creative commons, but not the enforcement team. <br>Image model creators can feasibly claim ‘innocent infringement’ if they build on this content, since they can claim a fair assumption that all images are properly tagged.<em> (IANAL — I’m making an assumption here. If anyone has a legal background that would like to clarify, please reach out)</em></p><p>Ultimately <strong>the role of enforcing copyright falls upon the copyright owner</strong>, but expecting every artist to spend hours regularly checking these site for their content is a lot of undue stress.</p><p>Ironically, I believe the image models and their proliferation may be the solution. My proposed solution is two-fold:</p><p>* Sites like <a href="https://haveibeentrained.com/">HaveIBeenTrained</a> allow for users to verify themselves and their artwork for an API that can be used to check ‘model-open’ sites for improper content. This can then be used for a 1:1 image search on “these sites to ensure their content isn’t added. Think of this like TinEye or Reverse Google Image search but specifically for sites that permit model scraping. This would be the “Opt-Out” enforcement.</p><p>* Using BLIP on ViT/H on submissions would allow sites to determine likely captions on an image, and from that assess if the image should be reviewed by a smaller team of moderators/curators. This could also be used to prevent AI images generated from previous models that don’t have significant changes from source material.<br>(BLIP generates proposed captions on an image, a UI version can be found here —<a href="https://huggingface.co/spaces/pharma/CLIP-Interrogator"> https://huggingface.co/spaces/pharma/CLIP-Interrogator </a>)</p><p>I hate to even say it aloud, but the concepts behind registering an image online (NFT-ideals) are useful here, but I don’t think blockchain is the right solution.</p><p>TL;DR — Even if Opt-Out is the default, someone can post images without permission to sites like Flickr and opt-in without permission of the original creator, and without an approved/verified ownership of those images on the site, the site owners can’t verify without the artist filing a copyright claim.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a381ee8c76ec" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Predictions on ML tools for Creativity]]></title>
            <link>https://medium.com/@tylermakesgames/predictions-on-ml-tools-for-creativity-b9e80c9f4327?source=rss-b6b8d3bf25ce------2</link>
            <guid isPermaLink="false">https://medium.com/p/b9e80c9f4327</guid>
            <category><![CDATA[ai-art]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[chatgpt]]></category>
            <dc:creator><![CDATA[Tyler]]></dc:creator>
            <pubDate>Sat, 17 Dec 2022 19:08:55 GMT</pubDate>
            <atom:updated>2022-12-19T17:17:01.364Z</atom:updated>
            <content:encoded><![CDATA[<p>A couple predictions where this AI (ML) Image &amp; Text proliferation will lead us:</p><p>First up, let’s be clear that I’m not stating anything here is specifically positive or negative, my goal here is to look at this as a thought experiment, a projection based on what I’ve seen in this space since 2015. Thoughts are my own, etc. etc.</p><p><em>Note- Machine Learning (ML) is frequently labeled as AI, but what we culturally consider to be ‘AI’ is a moving target. So while I use them interchangeably, I mean Machine Learning.</em></p><h3>Image</h3><p>To further advance upcoming models, Midjourney, Stable, and OpenAI will use their <strong>generated images with </strong><a href="https://wandb.ai/ayush-thakur/RLHF/reports/Understanding-Reinforcement-Learning-from-Human-Feedback-RLHF-Part-1--VmlldzoyODk5MTIx"><strong>RLHF</strong></a> to make a model that further distances itself from the original materials. The intention for this will most likely come from a similar intent as the LAION-Aesthetics, but with the added benefit of a step of obfuscation from the original source images.</p><p>But that won’t matter by the end of next year, because we will see the release of<strong> few-shot or zero-shot learning models</strong>, ones that don’t need the 400m-5b image sets for training. You can see a similar example with <a href="https://dreambooth.github.io/">Dreambooth</a> and soon it will be less dependent on the original source. This will probably lead us back to the <em>‘style transfer’</em> days of ML, where you need distinct (small) training sets to do anything meaningful.</p><p>With greater integration, the <strong>separation between “ai artist” and “traditional artist” will get far murkier</strong> in the days to come. Photoshop and Substance already have many ML filters and tools embedded, and with Figma, Canva, Krita, Blender, and other tools already incorporating Stable Diffusion into their workflows, we’ll begin to see more people using ML systems, potentially unaware of the underlying tech. <em>(If you’ve ever used the </em><a href="https://helpx.adobe.com/photoshop/how-to/fills-masks-sensei.html"><em>Content-Aware tools</em></a><em> in Photoshop, you’ve made images with AI/ML).</em></p><h3>Text</h3><p>From the text standpoint, GPT-4 and the new embeddings updates will continue to push the text generation models forward. OpenAI will release a better GPT-detector, but until it is clear of false positives, there will be no clear way to determine if a text is fully or partially generated with AI.</p><p>In the games space, there will be at least a few games that utilize this to create volumes of filler text in their products. Instead of having excerpts and small sections of books in a game like Skyrim, there will be hundreds of full tomes of text with enough readability to be passably good. Instead of NPCs with 3–5 canned responses, online games will allow for limited conversations with text parsing using systems similar to ChatGPT.</p><h3>Code</h3><p>This one is tricky to predict without knowing the outcome of the <a href="https://githubcopilotlitigation.com/">Github Copilot class action suit</a>. Similar to other areas, the most likely output of code AI/ML tools is not the full code itself, but some elements of it. Generating comments, reformatting &amp; linting will all be improved by ML tools in the near future, as well as quickly generating known algorithms in self-contained units <em>(CodeML: write fast inverse sqrt in C++)</em></p><h3>Audio</h3><p>Tools already exist for simple melody and music generation (jukebox, AIVA, boomy) but they haven’t hit the mainstream like image tools yet. The first spaces in music to see this work are non-vocal genres, as the voice synthesis isn’t quite there yet. However, that space is advancing quickly too, so its quite possible we’ll have a full style transfer system for voices, instrument styles, and more by ~2024.</p><h3>Animation</h3><p>Frame tweening will be vastly improved by AI/ML tools giving traditional animation and stop-motion an opportunity to boost productivity and cost effectiveness. Television cartoons will benefit from this the most, being able to produce the weekly content at a much higher rate while maintaining style &amp; quality. Film makers with animation workflows (such as rotoscoping) will also greatly benefit from this advancement.</p><h3>3D Models</h3><p>This will be the toughest one to crack for game-ready assets, due to the many stages it takes to prepare an asset for games. Breaking that process down, its likely that stages of it will see more AI tools incorporated. Tools built on the work of <a href="https://zhan-xu.github.io/rig-net/">RigNet</a> will assist in the process of rigging &amp; animation, further advancements in models such as <a href="https://nv-tlabs.github.io/AUV-NET/">AUV-Net</a> will see better auto UV unwrapping.</p><h3>Media Industries</h3><p>By ~2024, legislation will be proposed in many countries and regions enforcing an EXIF, watermark, or disclaimer on ML generated content. China is already there, and I’m sure EU will follow suit. US might be the last due to fair use policies providing more leeway, but California, Virginia, and Texas have already taken first steps in relation to deepfakes.</p><p>Game studios and film studios will find themselves split, some companies including many AI processes in their workflows, and others deciding to forbid their use. For those forbidding the practice wholesale, they will find it difficult to compete financially and will most likely need to target audiences that support their initiative, creating a “<em>GMO-free</em>”-esque branded environment for entertainment. This one will be impacted heavily by unionization efforts in games and the film VFX industry.</p><p>Speaking of, any <strong>game or film</strong> company (who doesn’t forbid use) with products of a reputable style will<strong> develop their own internal models</strong> utilizing few-shot learning and make it a proprietary part of their workflow. This will be something creators will need to understand when reviewing contracts and employee agreements.</p><p>As for<strong> job loss in entertainment technology</strong>- The biggest loss of jobs we’ll see won’t be at the studios in western countries, but the outsourcing firms that the AAA studios work with in Asia and the Middle East. Companies like Virtuos, DNEG, and Keywords will see less client engagement due to in-house solutions for mass production. It’s even possible that we see a short period of growth for jobs in the primary studios, but it won’t balance the losses in the outsourcing space.</p><h3>Miscellaneous</h3><p><strong>Content moderation &amp; customer support </strong>using ML text analysis is already here, and will become more prevalent. Smaller companies will now have access to the same sort of CS &amp; auto-moderation tools that major tech companies have, and will run into the same auto-moderation problems they face without the infrastructure to deal with it.</p><p><strong>User Generated content &amp; modding</strong> in games will see a huge boon from these tools, allowing smaller teams to create larger content in pre-existing games/engines. Utilizing ML tools either through approved modding/UGC pipelines in game or external tools more companies will be able to provide the simple building blocks of content creation within their game worlds (Roblox and Core but with more ML systems generating content for creators)</p><p><strong>Live games will be utilizing the text </strong><a href="https://openai.com/blog/new-and-improved-embedding-model/"><strong>embedding models</strong> of OpenAI</a> for further optimization of content and meta. It will also be used for detecting cheating, griefing, and other ToS violations in online &amp; live games. These tools will give data analysts far more ability to find outliers and extrapolate from current data sets.</p><p><strong>Advocacy for UBI</strong> will increase as more careers &amp; industries are threatened, especially by those who work in creative fields. The Overton window for UBI will move to a more mainstream conversation, as factory, transportation, retail, and other major industries all run into a <a href="https://en.wikipedia.org/wiki/Post-scarcity_economy">post-scarcity</a> dilemma.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b9e80c9f4327" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[So your Discord account was hacked…]]></title>
            <link>https://medium.com/@tylermakesgames/so-your-discord-account-was-hacked-441b5061fa67?source=rss-b6b8d3bf25ce------2</link>
            <guid isPermaLink="false">https://medium.com/p/441b5061fa67</guid>
            <category><![CDATA[discord]]></category>
            <category><![CDATA[game-development]]></category>
            <category><![CDATA[indie-game]]></category>
            <dc:creator><![CDATA[Tyler]]></dc:creator>
            <pubDate>Thu, 09 Dec 2021 20:02:20 GMT</pubDate>
            <atom:updated>2021-12-11T17:02:45.298Z</atom:updated>
            <content:encoded><![CDATA[<p><em>On 12/7/21 at 11:42 my Discord account was compromised and I’m still dealing with the ramifications of it.</em></p><p>Hey y’all,<br>I am an indie game dev and part time lecturer who uses Discord for both jobs. A “hacker” <em>(read: script kiddie)</em> stole my nitro boosted account, DMed my contacts, and deleted my verified server. Losing my account and servers isn’t just a personal loss, it also directly impacts my professional life.</p><p><strong>How did this happen?</strong><br>A trusted friend sent me a DM asking to playtest a game</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/502/0*u4-36DmkUZMjzVzY.png" /><figcaption>an example of what it started sending as my account once it was stolen</figcaption></figure><p>Consider my dual role as indie dev and game dev lecturer, this is not uncommon for me to receive. I took these precautionary steps:</p><ul><li>Asking the person personal questions to ensure it was them</li><li>After downloading the file, virus scanning it</li><li>Closing all applications before running it</li><li>Going into airplane mode and running a “netstat -a” in console to see if it tried reaching any IP addresses</li></ul><p><strong>None of this mattered.</strong> When I opened Discord and Twitter, entered my 2FA codes to both, my accounts were immediately compromised.</p><p><strong>As an indie-</strong> I can kiss most of that 10k user base goodbye. It will most likely take me a few months, if not years, to rebuild that type of community (that Discord server was over 5 years old). Many folks in there just checked for the announcements for game updates and weren’t regulars. This loss is incredible for a small indie like myself, and I spent a good portion of yesterday debating whether I would just cancel my next game because of this.</p><p><strong>As a lecturer-</strong> We use optional Discord servers for classes so students can work in groups and have quick voice/text channels for support. I had multiple servers set up for students where they could ask myself and the TAs questions. Many students also DM me with occasional questions, so I’m sure the hacker targeted a few with this same scam (and as you can imagine, some students would be happy to playtest something for a trusted faculty)</p><p><strong>On a personal level-</strong> I may lose friends from this, or at least their trust. There’s some hints of victim blaming when falling for this type of scam, because many people <em>(including myself, before this)</em> have an attitude of “it would never happen to me, I’m a smart one”. While I don’t expect people will outright unfriend me for this, I assume that their trust and goodwill is diminished from this. In fact, I probably won’t go back to using Discord for anything beyond work, at least for awhile.</p><p><strong>My lessons learned:</strong></p><ul><li><strong>Make a separate owner account for your servers</strong>, secure it, and never use it except for “run as admin” tasks. <strong>If you take one thing away from this, please let it be this.</strong> Losing your server owner account on Discord cannot be resolved by them. I now have 3 discord accounts, one for each job and one personal account for the normal use. <br>Addendum: Discord doesn’t let you use the same phone number for 2FA on multiple accounts. This is tricky to get around.</li><li><strong>Check with any close friends on a separate form of communication</strong> before trusting their “playtests”. This one is obvious in hindsight, but I thought asking probing questions was enough. It was not.</li><li><strong>Set up a virtual/disconnected sandbox environment when playtesting</strong>. It’s a gut response to say “don’t open .exes”, but I help playtest lots of games, and they are sketch “untrusted’ exes about half of the time. In the future, whenever I playtest <em>(even for a trusted source like IGF or Indiecade)</em> I’ll use <a href="https://docs.microsoft.com/en-us/windows/security/threat-protection/windows-sandbox/windows-sandbox-overview">Windows Sandbox</a>, or do so on an old laptop that is exclusively for that. (h/t to Sarah &amp; Miah for that one)</li><li><strong>2FA is not as secure as I thought.</strong> Once you’ve entered the 2FA code, its still possible for malware to steal your auth token and compromise your account. I treated it like a magic bullet that stops any and all password stealing, but it is not.</li><li><strong>Multiple passwords and account usernames are a must. </strong>I’m fortunate that I had separate emails and passwords for my various social media accounts and banking info, because if I hadn’t this would have gone far worse for me. Note: Discord won’t let you remove payment info while you have an active Nitro, even if you pay yearly. Its stuck on there. My friend had a $500 charge attempt when his account was compromised. <br><strong>Lock down your paypal/card immediately if this happens to you.</strong></li><li><strong>Don’t use Discord for personal conversations. </strong>This might seem extreme for folks that use it like I once did, but once your account is breached, the new account holder has full access to all of your DMs and any personal information held within them. From this point on, I’ll be using <a href="https://signal.org/en/">Signal</a> for any personal conversations.</li><li><strong>Don’t trust Discord to help.</strong> <strong>Its been over 96 hours</strong> and I haven’t heard a thing back from them. I was a verified server owner, I went through the process to get Discord’s equivalent to the “blue check”. It didn’t matter. <br><em>My twitter was also compromised and I had it back in 2 hours.</em> Discord did not scale their customer support with the rest of the product, and it shows. If something happens to your Discord account, do not wait for Discord to respond with any form of help. Assume you are on your own.</li></ul><p>Hopefully you’re reading this before this happens to you and not after. If this does happen to you, feel free to reach out. We’ll start a support group.</p><p>EDIT: My suggestions for Discord, if someone from the company happens to see this:</p><ul><li>If an account registers a “hacked account” issue, maybe don’t let the current account holder post on the zendesk tickets? Seems like an easy way for hacker/bot to instantly close any tickets opened by the real user (they immediately made a new account with my original email)</li><li>New account with an email address that was previously used for a account? Maybe require the account to verify the email before locking that email address to the account (I was told I couldn’t recover my account because there was another account with that email. I didn’t create that account)</li><li>Add a special contact pipeline for verified server owners. If you’re going to have a verification process for high-population verified owners, you should have a response system that doesn’t take 96+ hours.</li><li>From my understanding, these hacks come from an injection into Discord’s js files. Discord app should verify its javascript files with a checksum before allowing a user to authenticate and log in</li><li>Before an account can change the 2FA, confirm with the current 2FA, or require them open a ticket if they can no longer access the previous 2FA device. You make backup codes. Use them.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=441b5061fa67" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>