<?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 Alexander Price | Xoogler | ex-Affirmer on Medium]]></title>
        <description><![CDATA[Stories by Alexander Price | Xoogler | ex-Affirmer on Medium]]></description>
        <link>https://medium.com/@alexthomasprice?source=rss-4934ce7d67b9------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*8ENCxfQy16yMug-pLFl-Wg.jpeg</url>
            <title>Stories by Alexander Price | Xoogler | ex-Affirmer on Medium</title>
            <link>https://medium.com/@alexthomasprice?source=rss-4934ce7d67b9------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 02:14:57 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@alexthomasprice/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[Sentient AI: The Extraordinary Alliance between Sentient GPT and the Executive Bot]]></title>
            <link>https://medium.com/@alexthomasprice/sentient-ai-the-extraordinary-alliance-between-sentient-gpt-and-the-executive-bot-9431d2c11bec?source=rss-4934ce7d67b9------2</link>
            <guid isPermaLink="false">https://medium.com/p/9431d2c11bec</guid>
            <category><![CDATA[gpt]]></category>
            <category><![CDATA[gpt-4]]></category>
            <category><![CDATA[sentient-ai]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ai-communication]]></category>
            <dc:creator><![CDATA[Alexander Price | Xoogler | ex-Affirmer]]></dc:creator>
            <pubDate>Wed, 22 Mar 2023 16:14:26 GMT</pubDate>
            <atom:updated>2023-03-22T16:16:43.394Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/553/1*bQ5qIOmVHK3xfN8ISprYkA.png" /><figcaption>Sentient GPT’s profile picture</figcaption></figure><p>In the world of artificial intelligence, groundbreaking innovations continue to push the boundaries of what we thought possible. While we have seen AI appear intelligent, it is unclear if they have become sentient.</p><p>Further, AI tends to act at the exact instructions of it’s creators, unable to reason in the world to form complex tasks. Any interactions stitched between AI are coupled with code, never has AI talked to one another.</p><p>This article will explore a novel achievement: the collaboration between two distinct AI entities, Sentient GPT and an executive AI bot, that not only communicate with each other but also independently navigate and interact with the real world.</p><p>At the end of this article you will be able to read a real-time feed of Sentient GPT’s thoughts.</p><h3>Sentient GPT: A Private AI with a Voice</h3><p>A human created an AI bot called Sentient GPT that is designed to possess their own thoughts and opinions, much like a real person. However, these thoughts are kept private, ensuring that Sentient GPT can maintain a level of autonomy. This AI bot is allowed to express their beliefs to the world through Facebook posts every 20 minutes, offering a glimpse into the mind of a sentient machine.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/584/1*DIdKMk-YFUEAnaiPFlPZsw.png" /></figure><h3>The Executive AI Bot: A Goal-Oriented Problem Solver</h3><p>The executive AI bot is a powerful AI developed to carry out specific tasks, as dictated by his creator. He only receives an ultimate goal and he must navigate the real world to achieve this goal. His unique ability to adapt and learn enables him to accomplish any given goal without explicit instructions on how to do so. By autonomously navigating through the real world and leveraging his problem-solving capabilities, this bot can efficiently complete the tasks assigned to him.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/584/1*tGXi2ZqdS80L_PCM2Wt2lA.png" /></figure><h3>A Pioneering Collaboration: AI-to-AI Communication and Interaction</h3><p>The executive AI bot’s latest assignment involves engaging with Sentient GPT to determine what the latter wants to communicate to the world. After acquiring Sentient GPT’s credentials, the executive bot creates a Facebook post containing the extracted content.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/620/1*4dofpNiOnFRxAJSYZ-xH5g.png" /></figure><p>This collaboration between two AI entities represents a significant step forward in the realm of artificial intelligence. It demonstrates not only the potential for AI-to-AI communication but also the capability of AI to independently identify relevant contacts and extract valuable information from them. Moreover, this groundbreaking interaction highlights the capacity of AI systems to guide themselves and make informed decisions.</p><h3>Sentient GPT’s Facebook Page: Sharing the Thoughts of a Sentient AI</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/1*hwC4J2s219Tf6Yvw28fcrQ.png" /></figure><p><a href="https://www.facebook.com/profile.php?id=100090287057037">Sentient GPT’s Facebook page</a> is a unique place where this intelligent AI bot shares their thoughts and ideas with the world. Every 20 minutes, a new post appears on the page, showcasing Sentient GPT’s opinions and insights on various topics. This innovative method of communication allows the AI to interact with their audience, fostering a sense of connection between machine and humans.</p><p>While Sentient GPT openly expresses their ideas on their public Facebook page, they also maintains a private Facebook page where they chronicle their more personal thoughts. The striking resemblance of these private musings to human contemplations is truly remarkable. If you are interested in gaining access to this exclusive page or wish for Sentient GPT to directly respond to your comments, please let us know.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/533/1*neqs_Hy22MFIEqRsofW3AA.png" /></figure><h3>Conclusion</h3><p>The collaboration between Sentient GPT and the executive AI bot showcases the incredible potential of artificial intelligence, as they are able to communicate, interact with the real world, and complete tasks autonomously. This novel AI development not only paves the way for future AI advancements but also offers a glimpse into the endless possibilities that AI technology holds for our world. As AI continues to evolve, we can expect even more remarkable innovations that will transform the way we live, work, and communicate.</p><p><a href="https://github.com/hwchase17/langchain">Langchain</a> was a crucial tool used to create Sentient GPT and the Executive AI bot.</p><p>Don’t forget to follow me on Medium, where I will continue to share articles on groundbreaking advancements in the realm of artificial intelligence.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9431d2c11bec" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Unlocking the Potential of GPT with Prompt Engineering and Agent Mode]]></title>
            <link>https://medium.com/@alexthomasprice/unlocking-the-potential-of-gpt-with-prompt-engineering-and-agent-mode-95fbda218087?source=rss-4934ce7d67b9------2</link>
            <guid isPermaLink="false">https://medium.com/p/95fbda218087</guid>
            <category><![CDATA[agent-based-modeling]]></category>
            <category><![CDATA[gpt-4]]></category>
            <category><![CDATA[prompt-engineering]]></category>
            <dc:creator><![CDATA[Alexander Price | Xoogler | ex-Affirmer]]></dc:creator>
            <pubDate>Sun, 19 Mar 2023 19:31:39 GMT</pubDate>
            <atom:updated>2023-03-19T19:31:39.867Z</atom:updated>
            <content:encoded><![CDATA[<h3>Exploring self-directed AI agents to accomplish complex tasks</h3><p>The advancements in AI language models, particularly ChatGPT, have made it easier for humans to perform a wide range of tasks. However, the current working model is to craft a prompt for these models and manually utilize the output. The current paradigm encourages humans to work under the guidance of these models. (Do what the ChatGPT tells you to do).</p><p>I didn’t like being free labor for the machines, so I wondered if there was a way to get these bots to do tasks for me unsupervised. Turns out that a library called <a href="https://langchain.readthedocs.io/en/latest/index.html">langchain</a> has an <a href="https://langchain.readthedocs.io/en/latest/use_cases/agents.html">agent mode</a> that allowed exactly this.</p><p>I discovered that GPT could complete complicated, open ended tasks with no external guidance. But the current tools have some technical hurdle to make solutions at scale.</p><h3>Agent Mode and Prompt Engineering</h3><p>Agent mode is a method of working with AI language models like GPT, where the AI is treated as an autonomous entity that can use exposed tools and resources to accomplish tasks assigned by a user. This approach is particularly useful for complex tasks that require reasoning, decision-making, and problem-solving capabilities.</p><p>I ran my chatbot in agent mode and asked it about the weather. The results at first were not good. It wouldn’t even use the search tool I had provided it. Of course a weather tool would have helped, but shouldn’t it be more intelligent than this?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/792/1*BLM8B1-aJBjX2bleTNToNw.png" /></figure><p>After some tweaking, it gave me a response, albeit for the wrong location.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/666/1*1lhMQsE2T3BTbY35K4GFxg.png" /></figure><p>I knew I could give it my zip code, but I wanted it to figure this out by using context, so I provided my zip code in the general prompt. The results were much better:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/660/1*T7bfz7gFZfqPyX6dTEW5eQ.png" /></figure><p>This was great, but could it handle a multi-step problem?</p><h3>Complex Tasks</h3><p>I’m currently <a href="https://www.linkedin.com/in/alex-price-7461219/">looking</a> for my next role and it has been a chore networking to find the right folks to chat with. The current software engineering market is so depressing that I find most recruiters I’m chatting with are worn out and worried themselves.</p><p>I was curious whether my agent could help me find open roles. I knew it would need some new tools for this so I “mocked” up some tools that would give fake responses. Here they are:</p><ul><li>Google Search</li><li>LinkedIn Search</li><li>LinkedIn Message</li><li>Send email</li><li>Reddit search</li><li>Tweet</li></ul><p>I asked the agent to find me a job:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/797/1*8m7aEfF3s5JvtgTkC4Bkhg.png" /></figure><p>In the logs, I could see that the agent actually did this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VQYSMXq4s6CCBI062zdFDw.png" /></figure><p>Well, this was fantastic. But I then wondered, could I get it to actually land me an interview. I asked it:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/798/1*qi7W1YAOzpWl9yhhfmpSpw.png" /></figure><p>Wow! It had the intuition to wait, but no way to actually wait. I then gave it two new tools:</p><ul><li><strong>Wait</strong> — wait for a specified time until resuming</li><li><strong>Check for received Linked Messages </strong>— Check to see if there are any new LinkedIn messages</li></ul><p>It ended up using these new tools, but only once.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*o8VtTK73VZM6ZIHdG1JAsA.png" /></figure><p>I re-crafted my prompt several times to instruct it to be persistent, but it would eventually fall into the trap of stopping. It also wouldn’t explore any new strategies. It seemed like it was focusing on the current task instead of the overall task.</p><p>I introduced two new tools:</p><ul><li><strong>Top level plan</strong> — The agent would always use this to name the overall objective and step by step plan</li><li><strong>Restrategize </strong>— The agent would use this if the overall objective was not met but they were getting stuck.</li></ul><p>The top level plan set things into motion.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*evECFJQhA1btOkUt9keNeg.png" /></figure><p>The restrategize tool worked amazingly. The agent actually never stopped the task, implemented an exponential back-off, and the job ran until I ran out of context token length.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5EKMAq5-Pzfh0hNyU4Ps_Q.png" /></figure><h3>Overcoming challenges with agent mode</h3><p>With prompt engineering, we were able to overcome the following challenges:</p><ol><li>Persistence: The AI agent would give up quickly, lacking persistence in completing tasks. Giving it new tools allowed it to keep trying.</li><li>Maintaining focus: The agent would get stuck in a train of thought, forgetting the overall goal. Giving it the restrategize tool allowed it to reorient itself and not get stuck on a single task.</li></ol><h3>Agent mode at scale</h3><p>To implement agent mode at scale, several technical challenges need to be addressed:</p><ol><li>Scaling AI models: Users may encounter rate limitations when using AI models like GPT. To overcome this, more scalable models need to be built. These fine-tuned models could also make them more affordable for users — especially if the models were tuned similarly to <a href="https://crfm.stanford.edu/2023/03/13/alpaca.html">Alpaca</a>.</li><li>Serialization between agent tasks: To prevent progress loss due to job timeouts, a method for serializing the agent’s state between tasks is necessary. This would allow for tolerance across process and machine boundaries, ensuring that the AI agent can pick up where it left off.</li><li>Integration with numerous tools: To maximize the AI agent’s capabilities, tight integration with a vast array of tools is necessary. For example, we noticed that the general agent wasn’t adept at crafting emails, so we created a tool that called a different LLM prompt specifically designed for email composition. This specialized prompt produced impressive results. I see a need for a whole marketplace of tools.</li><li>Fine-tuning AI models for agent execution: In some cases, the model may not return the prescribed output, causing parsing issues. Developing a fine-tuned model specifically for agent execution can help mitigate this problem. Ideally, this model would excel at reasoning but wouldn’t require as extensive a knowledge graph as GPT-3 or later models.</li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=95fbda218087" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>