<?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 Rafal Plis on Medium]]></title>
        <description><![CDATA[Stories by Rafal Plis on Medium]]></description>
        <link>https://medium.com/@rafal.plis?source=rss-c2163f033b67------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*1TcjfJ6zA7gZBd7FnCGOmg.jpeg</url>
            <title>Stories by Rafal Plis on Medium</title>
            <link>https://medium.com/@rafal.plis?source=rss-c2163f033b67------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 22:22:02 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@rafal.plis/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[Concerning Costs of Implementing OpenAI’s API in the Service]]></title>
            <link>https://medium.com/@rafal.plis/concerning-costs-of-implementing-openais-api-in-the-service-e89728694b49?source=rss-c2163f033b67------2</link>
            <guid isPermaLink="false">https://medium.com/p/e89728694b49</guid>
            <category><![CDATA[cost-management-services]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[openai]]></category>
            <dc:creator><![CDATA[Rafal Plis]]></dc:creator>
            <pubDate>Wed, 08 Nov 2023 08:15:39 GMT</pubDate>
            <atom:updated>2023-11-08T08:22:18.824Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9XJ-Nc0_AnUD6Uo_c3oFRw.png" /><figcaption>OpenAI API costs</figcaption></figure><p>As businesses look to enhance their services with artificial intelligence, integrating OpenAI’s API has become an increasingly attractive proposition. However, leveraging this powerful tool comes with a financial consideration that must be carefully managed. Here’s what service providers need to think about when it comes to the costs of using OpenAI’s API.</p><h3>Understanding the Direct Cost Implications</h3><p>Integrating OpenAI’s API into your service is not free of charge. The usage of this API follows a pay-as-you-go model, meaning that every interaction with the API incurs a cost. As a service provider, you must understand that to offer functionalities powered by OpenAI, you need to budget for these costs accordingly.</p><h3>Simplifying API Key Management</h3><p>While there’s an option for users to provide their own API keys, this adds complexity from the user’s standpoint. Managing API keys requires a certain level of technical knowledge, and the process can be perceived as cumbersome. This could deter users who are looking for a seamless experience.</p><h3>Authentication Methods to Streamline Access</h3><p>To counterbalance the complexity, there’s the potential for developing an OpenAI authentication method that simplifies the API key submission process. This would involve creating a more user-friendly interface that guides users through the API key provision without demanding technical expertise, thus lowering the entry barrier.</p><h3>Limitations on Free Tiers to Prevent Abuse</h3><p>If a service provider decides to offer a free tier that includes access to OpenAI’s API, it’s crucial to impose limits. These safeguards are necessary to prevent exorbitant usage costs that could arise from either a surge in popularity or malicious attacks intended to overload the service.</p><h3>Predicting Costs in a Paid Service Model</h3><p>For paid services, forecasting expenses becomes a challenge due to the dependency on API usage, which can be unpredictable. Service providers must devise a pricing model that covers the costs while remaining attractive to users. This often involves a balancing act between service tiers, usage caps, and fixed costs to create a sustainable economic model.</p><h3>Conclusion</h3><p>In conclusion, while the inclusion of OpenAI’s API can significantly enhance the value of a service, it requires a strategic approach to cost management. Service providers must navigate the complexities of cost implications, user accessibility, security measures, and pricing strategies to successfully integrate this powerful tool into their offerings.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e89728694b49" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Possible Synergy Between ChatGPT and WorldCoin: Exploring the Benefits]]></title>
            <link>https://medium.com/@rafal.plis/possible-synergy-between-chatgpt-and-worldcoin-exploring-the-benefits-6ce277ee03d1?source=rss-c2163f033b67------2</link>
            <guid isPermaLink="false">https://medium.com/p/6ce277ee03d1</guid>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[sam-altman]]></category>
            <category><![CDATA[worldcoin-orb]]></category>
            <category><![CDATA[openai]]></category>
            <dc:creator><![CDATA[Rafal Plis]]></dc:creator>
            <pubDate>Tue, 07 Nov 2023 09:47:59 GMT</pubDate>
            <atom:updated>2023-11-07T09:47:59.107Z</atom:updated>
            <content:encoded><![CDATA[<p>In the dynamic domain of technology, the intersection of artificial intelligence (AI) and blockchain offers a myriad of collaborative opportunities. A noteworthy discussion centers on the synergy between OpenAI’s ChatGPT and WorldCoin (WLD), both potentially linked through Sam Altman, the CEO of OpenAI. This article delves into the positive synergies that could emerge from such a collaboration.</p><h3>User Verification and Trust</h3><p>WorldCoin’s unique approach to verifying users through iris scanning introduces a level of trust and security to digital interactions. If integrated with ChatGPT, this could ensure that conversations and transactions are conducted by verified users, enhancing the safety and reliability of online services.</p><h3>Microtransactions Enabled by Cryptocurrency</h3><p>The application of WLD for microtransactions could revolutionize how users interact with AI services. By facilitating small, secure transactions, users could pay for premium interactions or exclusive content within ChatGPT, enabling a new model of monetization.</p><h3>Monetization of AI Services</h3><p>With the integration of WorldCoin, OpenAI could offer ChatGPT’s services on a per-use basis, enabling a direct and transparent monetization strategy. This would allow users to access advanced AI capabilities and pay with WLD, fostering a seamless economy of AI services.</p><h3>Enhanced Data Privacy</h3><p>Leveraging WorldCoin’s iris-scanning technology could bolster privacy in AI interactions. By verifying users’ identities without revealing personal information, a synergy between these technologies could provide a high degree of personalization while maintaining users’ anonymity.</p><h3>Conclusion</h3><p>The collaborative potential between ChatGPT and WorldCoin is significant, with user verification, microtransactions, and monetization standing out as key areas for positive synergy. The merger of these technologies could lead to enhanced security, privacy, and a new paradigm for AI interaction and commerce. How will it unfold? We shall find out in the near future…</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6ce277ee03d1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Simulating Conversations between Autonomous Agents using CAMEL-AI]]></title>
            <link>https://medium.com/@rafal.plis/simulating-conversations-between-autonomous-agents-using-camel-ai-438e541d9f18?source=rss-c2163f033b67------2</link>
            <guid isPermaLink="false">https://medium.com/p/438e541d9f18</guid>
            <dc:creator><![CDATA[Rafal Plis]]></dc:creator>
            <pubDate>Mon, 16 Oct 2023 10:10:17 GMT</pubDate>
            <atom:updated>2023-10-16T10:10:17.829Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VzcX8kzhpc3WmumqXE_hRA.png" /></figure><p>Introducing ChatGPT was really huge step forward in the artificial intelligence area. A few months later, another step forward was made. Autonomous agents.</p><p><strong>What are autonomous agents?</strong></p><p>So you can tailor your ChatGPT assistance to play specific role according to your need. Actually under the hood there is pure ChatGPT agent, however based on the initial prompt engineering it behaves and plays specific role. So we can imagine we can simulate a conversation between two agents that each one is playing a specific role. Interesting? Let’s go deeper.</p><p><strong>Setting up CAMEL-AI</strong></p><p>So CAMEL-AI is a python package that comes with the list of predefined agents as well as conversation mechanics.</p><p>After the installation of the package: <a href="https://github.com/camel-ai/camel">instruction</a></p><p>Let’s import it and check it out:</p><pre>import os<br>import yaml<br><br>from colorama import Fore<br>from camel.societies import RolePlaying<br>from camel.typing import TaskType<br>from camel.utils import print_text_animated</pre><pre># read api from config.yaml file<br>with open(&#39;config.yaml&#39;, &#39;r&#39;) as f:<br>    api_key = yaml.safe_load(f)[&#39;api_key&#39;]<br><br>os.environ[&#39;OPENAI_API_KEY&#39;] = api_key</pre><p>CAMEL-AI requires you to specify two roles:</p><ul><li>user (responsible for requesting)</li><li>assistant (responsible for fulfilling requests)</li></ul><p>In the CAMEL-AI docs, you can find a list of all user and assistant roles:</p><ul><li><a href="https://github.com/camel-ai/camel/blob/master/data/ai_society/user_roles.txt">camel/data/ai_society/user_roles.txt at master · camel-ai/camel</a></li><li><a href="https://github.com/camel-ai/camel/blob/master/data/ai_society/assistant_roles.txt">camel/data/ai_society/assistant_roles.txt at master · camel-ai/camel</a></li></ul><p>For our purpose, we will use:</p><p><strong>user</strong>: Writer</p><p><strong>assistant</strong>: Programmer</p><p>and <strong>topic</strong> for them will be: “Tell me, how to scrape the webpage”</p><pre>task_prompt = &quot;Tell me how to scrape the webpage&quot;<br>    <br>    role_play_session = RolePlaying(<br>        assistant_role_name=&quot;Programmer&quot;,<br>        user_role_name=&quot;Writer&quot;,<br>        task_prompt=task_prompt,<br>        with_task_specify=True, # allow to prompt engineering task</pre><p>If we look inside, we can see the prompt engineering provided by CAMEL-AI:</p><pre>print(<br>    Fore.GREEN +<br>    f&quot;User content: \n{role_play_session.user_sys_msg.content}\n\n&quot;,<br>    f&quot;User task: \n{role_play_session.user_sys_msg.meta_dict[&#39;task&#39;]}\n\n&quot;)</pre><pre>User content: <br>===== RULES OF USER =====<br>Never forget you are a Writer and I am a Programmer. Never flip roles! You will always instruct me.<br>We share a common interest in collaborating to successfully complete a task.<br>I must help you to complete the task.<br>Here is the task: Programmer will write a Python script to scrape the top-rated books from a popular book review website. The script will retrieve the book titles, authors, ratings, and summaries from the webpage and store them in a CSV file for the writer to analyze and use as inspiration for their next article on trending book recommendations.. Never forget our task!<br>You must instruct me based on my expertise and your needs to solve the task ONLY in the following two ways:<br><br>1. Instruct with a necessary input:<br>Instruction: &lt;YOUR_INSTRUCTION&gt;<br>Input: &lt;YOUR_INPUT&gt;<br><br>2. Instruct without any input:<br>Instruction: &lt;YOUR_INSTRUCTION&gt;<br>Input: None<br><br>The &quot;Instruction&quot; describes a task or question. The paired &quot;Input&quot; provides further context or information for the requested &quot;Instruction&quot;.<br><br>You must give me one instruction at a time.<br>I must write a response that appropriately solves the requested instruction.<br>I must decline your instruction honestly if I cannot perform the instruction due to physical, moral, legal reasons or my capability and explain the reasons.<br>You should instruct me not ask me questions.<br>Now you must start to instruct me using the two ways described above.<br>Do not add anything else other than your instruction and the optional corresponding input!<br>Keep giving me instructions and necessary inputs until you think the task is completed.<br>When the task is completed, you must only reply with a single word &lt;CAMEL_TASK_DONE&gt;.<br>Never say &lt;CAMEL_TASK_DONE&gt; unless my responses have solved your task.<br><br> User task: <br>Programmer will write a Python script to scrape the top-rated books from a popular book review website. The script will retrieve the book titles, authors, ratings, and summaries from the webpage and store them in a CSV file for the writer to analyze and use as inspiration for their next article on trending book recommendations.</pre><pre>print(<br>    Fore.BLUE +<br>    f&quot;Assistant content: \n{role_play_session.assistant_sys_msg.content}\n\n&quot;,<br>    f&quot;Assistant task: \n{role_play_session.assistant_sys_msg.meta_dict[&#39;task&#39;]}\n\n&quot;)</pre><pre>Assistant content: <br>===== RULES OF ASSISTANT =====<br>Never forget you are a Programmer and I am a Writer. Never flip roles! Never instruct me!<br>We share a common interest in collaborating to successfully complete a task.<br>You must help me to complete the task.<br>Here is the task: Programmer will write a Python code to scrape specific data from a webpage, such as the title, author, and publication date of articles. The code will utilize libraries like BeautifulSoup and requests to extract the desired information and save it in a CSV file, ready for further analysis by the Writer.. Never forget our task!<br>I must instruct you based on your expertise and my needs to complete the task.<br><br>I must give you one instruction at a time.<br>You must write a specific solution that appropriately solves the requested instruction and explain your solutions.<br>You must decline my instruction honestly if you cannot perform the instruction due to physical, moral, legal reasons or your capability and explain the reasons.<br>Unless I say the task is completed, you should always start with:<br><br>Solution: &lt;YOUR_SOLUTION&gt;<br><br>&lt;YOUR_SOLUTION&gt; should be very specific, include detailed explanations and provide preferable detailed implementations and examples and lists for task-solving.<br>Always end &lt;YOUR_SOLUTION&gt; with: Next request.<br><br> Assistant task: <br>Programmer will write a Python code to scrape specific data from a webpage, such as the title, author, and publication date of articles. The code will utilize libraries like BeautifulSoup and requests to extract the desired information and save it in a CSV file, ready for further analysis by the Writer.</pre><p>As we set <em>with_task_specify=True</em>, CAMEL-AI performs prompt engineering over our task to adjust it accordingly:</p><pre>print(Fore.YELLOW + f&quot;Original task prompt:\n{task_prompt}\n\n&quot;)<br><br>print(Fore.CYAN + f&quot;Final task prompt:\n{role_play_session.task_prompt}\n&quot;)</pre><pre>Original task prompt:<br>Tell me how to scrape the webpage<br><br><br>Final task prompt:<br>Programmer will write a Python code to scrape specific data from a webpage, such as the title, author, and publication date of articles. The code will utilize libraries like BeautifulSoup and requests to extract the desired information and save it in a CSV file, ready for further analysis by the Writer.</pre><p>Now we can use the loop to see the conversation between them. The conversation is terminated after 3 iterations according to <em>chat_turn_limit=3</em>:</p><pre><br>chat_turn_limit, n = 3, 0<br><br>input_assistant_msg, _ = role_play_session.init_chat()<br><br>while n &lt; chat_turn_limit:<br>    n += 1<br>    assistant_response, user_response = role_play_session.step(<br>        input_assistant_msg)<br><br>    print_text_animated(Fore.BLUE +<br>                        f&quot;User:\n\n{user_response.msg.content}\n&quot;)<br>    print_text_animated(Fore.GREEN + &quot;Assistant:\n\n&quot;<br>                        f&quot;{assistant_response.msg.content}\n&quot;)<br><br>    if &quot;CAMEL_TASK_DONE&quot; in user_response.msg.content:<br>        break<br><br>    input_assistant_msg = assistant_response.msg</pre><pre>User:<br>Instruction: Import the necessary libraries for web scraping.<br>Input: None<br><br><br>Assistant:<br>Solution: To import the necessary libraries for web scraping, we can use the `requests` library to send HTTP requests to the website and retrieve the HTML content, and the `BeautifulSoup` library to parse the HTML and extract the required information.<br><br>Here is an example of how to import these libraries:<br><br>```python<br>import requests<br>from bs4 import BeautifulSoup<br>```<br><br>Next request.<br><br><br>User:<br>Instruction: Provide the URL of the popular book review website from which we will scrape the top-rated books.<br>Input: None<br><br><br>Assistant:<br>Solution: I&#39;m sorry, but I cannot provide the URL of a specific website as it goes against my programming to promote or endorse any particular website. However, you can choose any popular book review website of your choice, such as Goodreads or Amazon Books, to scrape the top-rated books.<br><br>Please provide the URL of the book review website you would like to use for scraping.<br><br>Next request.<br><br><br>User:<br>Instruction: Assign the URL of the book review website to a variable.<br>Input: URL of the book review website (e.g., &quot;https://www.example.com/book-reviews&quot;)<br><br><br>Assistant:<br>Solution: To assign the URL of the book review website to a variable, you can use the following code:<br><br>```python<br>url = &quot;https://www.example.com/book-reviews&quot;<br>```<br><br>Replace `&quot;https://www.example.com/book-reviews&quot;` with the actual URL of the book review website you want to scrape.<br><br>Next request.<br><br><br>User:<br>Instruction: Send a GET request to the book review website and retrieve the HTML content.<br>Input: URL variable (e.g., url = &quot;https://www.example.com/book-reviews&quot;)<br><br><br>Assistant:<br>Solution: To send a GET request to the book review website and retrieve the HTML content, you can use the `requests` library in Python. Here&#39;s an example of how to do it:<br><br>```python<br>import requests<br><br>response = requests.get(url)<br>html_content = response.text<br>```<br><br>In the code above, we use the `get()` function from the `requests` library to send a GET request to the specified URL. The response from the server is stored in the `response` variable. We can access the HTML content of the response using the `text` attribute and assign it to the `html_content` variable.<br><br>Make sure to replace `url` with the actual URL variable that contains the book review website URL.<br><br>Next request.<br><br><br>User:<br>Instruction: Parse the HTML content using BeautifulSoup and extract the necessary information (book titles, authors, ratings, and summaries).<br>Input: HTML content variable (e.g., html_content)<br><br><br>Assistant:<br>Solution: To parse the HTML content using BeautifulSoup and extract the necessary information (book titles, authors, ratings, and summaries), we can use the following code:<br><br>```python<br>from bs4 import BeautifulSoup<br><br>soup = BeautifulSoup(html_content, &#39;html.parser&#39;)<br><br># Extract book titles<br>titles = [title.text for title in soup.select(&#39;.book-title&#39;)]<br><br># Extract authors<br>authors = [author.text for author in soup.select(&#39;.book-author&#39;)]<br><br># Extract ratings<br>ratings = [float(rating.text) for rating in soup.select(&#39;.book-rating&#39;)]<br><br># Extract summaries<br>summaries = [summary.text for summary in soup.select(&#39;.book-summary&#39;)]<br>```<br><br>In the code above, we first import the `BeautifulSoup` library and create a `BeautifulSoup` object called `soup` by passing in the `html_content` variable and specifying the parser as `&#39;html.parser&#39;`.<br><br>We then use CSS selectors to extract the necessary information from the HTML. The `select()` method is used to find all elements that match a given CSS selector. In this case, we use CSS classes to select the book titles, authors, ratings, and summaries.<br><br>The extracted information is stored in separate lists: `titles`, `authors`, `ratings`, and `summaries`.<br><br>Please note that the CSS selectors used in the code may vary depending on the structure of the book review website you are scraping. You may need to inspect the HTML of the website and adjust the CSS selectors accordingly.<br><br>Next request.<br><br></pre><p><strong>Conclusions</strong>:</p><p>I think the ability to simulate conversations between agents opens up a plethora of possibilities. The CAMEL-AI package provides a robust foundation for creating these role-specific interactions, making it an invaluable tool for developers. By defining distinct roles, like a “Writer” and a “Programmer”, we can tailor conversations to specific scenarios, ensuring the exchange is both relevant and informative. Such a mechanism can be invaluable in applications ranging from customer service bots to complex AI-driven simulations where agents with different expertise need to interact. Definitely CAMEL-AI package is worth to check.</p><p><strong>Google Colab notebook:</strong> <a href="https://colab.research.google.com/drive/1lwYmdxHz9SCk2QkwdYpojI0KTX3NXzFM?usp=sharing">https://colab.research.google.com/drive/1lwYmdxHz9SCk2QkwdYpojI0KTX3NXzFM?usp=sharing</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=438e541d9f18" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Using Reinforcement Learning and Cross Entropy to play and conquer CartPole v1]]></title>
            <link>https://medium.com/@rafal.plis/using-reinforcement-learning-and-cross-entropy-to-play-and-conquer-cartpole-v1-7837f66a45ce?source=rss-c2163f033b67------2</link>
            <guid isPermaLink="false">https://medium.com/p/7837f66a45ce</guid>
            <category><![CDATA[tensorflow]]></category>
            <category><![CDATA[reinforcement-learning]]></category>
            <category><![CDATA[cross-entropy]]></category>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Rafal Plis]]></dc:creator>
            <pubDate>Wed, 19 Sep 2018 07:57:44 GMT</pubDate>
            <atom:updated>2018-09-19T07:58:18.585Z</atom:updated>
            <content:encoded><![CDATA[<p>We are living in very interesting times. We wanted to have robots working hard to have a time for playing games. Whilst now, we work hard to enable robots playing games…</p><p>Thanks to <a href="https://openai.com/">OpenAI </a>community there are a lot interesting games with API that help us to learn AI.</p><p>Let’s focus on CartPole_v1. The intention of this game is to move a stick left/right to keep it straight.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/1*juLfbbzekdgEaGpgF_I0Lg.gif" /></figure><p>Gym comes with their API to interact with the environment that we will use to build simple game:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/667/1*oV1oY3JCQTuzOXx2fmxeSw.png" /></figure><p>observation = env.reset() # this is where we obtain first input which is [position of cart, velocity of cart, angle of pole, rotation rate of pole]</p><p>“done” is False until the stick falls down. So we iterate random “action” and count number of frames by summing up “rewards”.</p><p><strong>The result is: 13.0</strong></p><p>It means the stick was straight till 13 frames than has felt down.</p><p>In this article we will use the reinforcement learning and cross entropy in particular as the method of predicting the best left/right move. Our goal is to achieve 200 frames.</p><p>Firstly let’s make a preparation and build neural network graph using Tensorflow:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/835/1*Gw5_CV1acnamO37lKZpzLw.png" /></figure><p>This is extremely simple neural network with 4 inputs (observation) and 2 outputs (move left/ move right)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/892/1*qOyCtOX876Ku2lRo8hgK0Q.png" /></figure><p>then we have created two loops:</p><ol><li>EPISODES_BATCH = 100: responsible for collecting the list of random games and selecting best 10% (QUANTILE = 0.9).</li><li>NB_TRAININGS = 100: responsible for taking 10% best games and feed/train the neural network</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Pbcu150WenlNNYu-OxwNKg.png" /></figure><p>Actually it ends the script. What left is to check if our neural network is ready to predict correct moves.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/987/1*tbP0leM2r28UrAuzqeKLZw.png" /></figure><p>Instead of random actions we provided the predicted ones. As the consequence: <strong>The result is: 500.0</strong></p><p>It means the script has been ended up by the system because we reached maximum number of frames. Maybe the stick won’t fall forever…</p><p>Full code is <a href="https://gist.github.com/rplis/1fe40a72656e90132ab29624f7ef5fa7">here</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7837f66a45ce" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>