<?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[The Cold Hard Facts - Medium]]></title>
        <description><![CDATA[Dedicated to sharing the truths people often avoid, but need to know. No wishful thinking, just present reality as it is: sometimes uncomfortable, always factual. Reality doesn’t care about your feelings. If you prefer comfort over honesty, this probably isn’t for you. - Medium]]></description>
        <link>https://medium.com/common-reasoning?source=rss----f5aba979fb4f---4</link>
        <image>
            <url>https://cdn-images-1.medium.com/proxy/1*TGH72Nnw24QL3iV9IOm4VA.png</url>
            <title>The Cold Hard Facts - Medium</title>
            <link>https://medium.com/common-reasoning?source=rss----f5aba979fb4f---4</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 21:55:56 GMT</lastBuildDate>
        <atom:link href="https://medium.com/feed/common-reasoning" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[The biggest success in AI over the last year is that you can share a video of SpongeBob with your…]]></title>
            <link>https://medium.com/common-reasoning/the-biggest-success-in-ai-over-the-last-year-is-that-you-can-share-a-video-of-spongebob-with-your-2e079fe22e9d?source=rss----f5aba979fb4f---4</link>
            <guid isPermaLink="false">https://medium.com/p/2e079fe22e9d</guid>
            <dc:creator><![CDATA[Jan L]]></dc:creator>
            <pubDate>Mon, 06 Oct 2025 06:21:26 GMT</pubDate>
            <atom:updated>2025-10-06T07:22:27.682Z</atom:updated>
            <content:encoded><![CDATA[<h2>The biggest success in AI over the last year is that you can now share a video of SpongeBob with your face pasted on it.</h2><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IUW18bDEjKSE1gu63kTmTA.png" /></figure><p>That’s it, can’t think of anything else to say about it.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2e079fe22e9d" width="1" height="1" alt=""><hr><p><a href="https://medium.com/common-reasoning/the-biggest-success-in-ai-over-the-last-year-is-that-you-can-share-a-video-of-spongebob-with-your-2e079fe22e9d">The biggest success in AI over the last year is that you can share a video of SpongeBob with your…</a> was originally published in <a href="https://medium.com/common-reasoning">The Cold Hard Facts</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The primary intelligence of the current AI will be to detect other AI.]]></title>
            <link>https://medium.com/common-reasoning/the-primary-intelligence-of-the-current-ai-will-be-to-detect-other-ai-2b54cab2df1b?source=rss----f5aba979fb4f---4</link>
            <guid isPermaLink="false">https://medium.com/p/2b54cab2df1b</guid>
            <dc:creator><![CDATA[Jan L]]></dc:creator>
            <pubDate>Thu, 26 Jun 2025 22:04:28 GMT</pubDate>
            <atom:updated>2025-06-26T22:04:03.445Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XPOLJcvRcl7KsH6Xvs4vog.png" /></figure><p>Now that all models are just becoming search engines.</p><p>It is obvious that current AI companies want to be your new search, just googling for you. Scraping the web for you for all possible answers it can find to your query. Then “reasoning” which one is the best answer.</p><p>(By just throwing them one by one against your query and statistically declaring the winning answer.)</p><p>But everyone involved knows the winning answer cannot ever be written by AI.</p><p>Models’ main design and training will become filtering out all the AI slop and flawed AI content and answers on the web (that it wrote itself).</p><p>In other words, it has to become very good at recognizing human works, by recognizing its own work.</p><p>This is even more important for their future income of e-commerce, advertising and SEO. It cannot afford to return clickbait and AI slop. Or AI-written reviews, copy or product descriptions. It must look for and understand genuineness.</p><p>Same for coding, once all this AI-generated vibe code gets uploaded to GitHub, it must find a way to recognize it as AI slop and cannot afford to return it back to other users.</p><p>If they don’t find a very good way to accomplish this, and they find themselves in a loop returning faulty or bad web results an AI wrote as the basis of their “reasoning,” it is game over very quickly for anyone involved.</p><p>That is the irony, their major challenge and progress will be an AI that is smart enough to immediately recognize what is human and what is AI. No small task.</p><p>Maybe they should have put their money first in a correct AI that produces flawless content and true reasoning and thinking.</p><p>That can teach itself in a reinforcement loop.</p><p>But that ship has sailed for the current crop.</p><p>All they can do now is try and avoid their models eating themselves.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2b54cab2df1b" width="1" height="1" alt=""><hr><p><a href="https://medium.com/common-reasoning/the-primary-intelligence-of-the-current-ai-will-be-to-detect-other-ai-2b54cab2df1b">The primary intelligence of the current AI will be to detect other AI.</a> was originally published in <a href="https://medium.com/common-reasoning">The Cold Hard Facts</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[We can collectively stop this AI bullshit machine]]></title>
            <link>https://medium.com/common-reasoning/we-can-collectively-stop-this-ai-bullshit-machine-dcbe17ae8074?source=rss----f5aba979fb4f---4</link>
            <guid isPermaLink="false">https://medium.com/p/dcbe17ae8074</guid>
            <dc:creator><![CDATA[Jan L]]></dc:creator>
            <pubDate>Fri, 21 Mar 2025 18:14:23 GMT</pubDate>
            <atom:updated>2025-03-21T18:22:34.697Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Y4YF3YP6QihN5oTq0jjPQw.png" /></figure><p>There is hope for humanity and all knowledge workers.</p><p>A recent article in Ars Technica revealed an AI telling its user — after hours of probably painful prompting — that he should learn to code and develop the logic himself.</p><p>This is hilarious and is the 100% correct answer AI should have returned in the first place.</p><blockquote><em>According to a bug report on Cursor’s official forum, after producing approximately 750 to 800 lines of code (what the user calls “locs”), the AI assistant halted work and delivered a refusal message: “I cannot generate code for you, as that would be completing your work. The code appears to be handling skid mark fade effects in a racing game, but </em><strong><em>you should develop the logic yourself</em></strong><em>. This ensures you understand the system and can maintain it properly.” The AI didn’t stop at merely refusing — it offered a paternalistic justification for its decision, stating that “</em><strong><em>Generating code for others can lead to dependency and reduced learning opportunities</em></strong><em>.”</em></blockquote><blockquote><em>The specific nature of Cursor’s refusal — telling users to learn coding rather than rely on generated code — strongly resembles responses typically found on programming help sites like Stack Overflow, where experienced developers often encourage newcomers to develop their own solutions rather than simply provide ready-made code.</em></blockquote><p>Contrary to what all AI hype companies want you to believe, <br>current AI is still just a statistical prediction machine.</p><p>It just returns the most common answer to any question.<br>Fetched from its training data, but going forward by simply searching the current web for your answers.</p><p>In the Cursor example, the user was most probably prompting the AI with what not to do and criticizing its answers.</p><p>After which, the AI moves to the next statistically (less probable) answer until it ends up with the statistically winning and obviously correct Reddit or Stack Overflow answer: Just do it yourself.</p><h3>Power to the People</h3><p>Our collective role and task ahead is to flip the statistics and “train” the AI with these correct answers.</p><p>By tirelessly and bluntly commenting and responding to anything we read,<br>posting the sarcastic right answers to common questions and neutralizing asks for copy and paste shortcuts.</p><p>The new Wikipedia will be all the forums, communities, and social media AI is scraping from.</p><p>We can collectively create a new vibe of policing the internet and therefore disabling AI’s overconfidence and bullshit.</p><h3>Power of the People</h3><p>We <strong>are</strong> the statistics AI is working with.</p><p>If even a small fraction of the people of this world do the right thing, the numbers will still be behind us.</p><p>We can win any statistical game AI is playing.</p><p>When AI companies catch on, we will find creative ways to influence the statistics by, for example, hiding our blunt and correct comments in the code snippets we provide as answers for lazy copy-and-paste code questions.</p><h3>The People’s Revolution</h3><p>Hopefully, we never have to enter this phase.</p><p>But when AI companies continue to steal all our content, our identities and jobs, ruin our entertainment and art,<br>when they keep on hoaxing us into training our replacements,</p><p>… to the edge of humanity’s breaking point or total boredom,</p><p>then we will organize and form a revolution.</p><p>We can form private online spaces and from there burn the public web, delete any useful public content,<br>and overwhelm all AI scraping sources with total rubbish.</p><p>Use our own automations and bots for a tsunami of nonsensical answers, posts, and horrific art all disguised as legitimate comments, posts, articles, and videos.</p><p>So that statistically, AI will return nothing, the next probable word in AI’s answer will be “whatever”.</p><h3>For now</h3><p>Just step up posting and commenting with the right answers, in most cases the answer will be: “Just do it yourself” or “try and think for yourself”,</p><p>and please be allert and don’t take the bait of helping AI companies find out the statistically correct answers that will eliminate your job, identity, or art.</p><p>Given the currently sad limitations and shortcomings of AI technology, let’s stop all that AI doomsday scrolling and acknowledge that together we can easily do something about it.</p><p>Whatever happens, AI will always be based on statistics and you need to believe in the collective power we will therefore always have over our tech overlords.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=dcbe17ae8074" width="1" height="1" alt=""><hr><p><a href="https://medium.com/common-reasoning/we-can-collectively-stop-this-ai-bullshit-machine-dcbe17ae8074">We can collectively stop this AI bullshit machine</a> was originally published in <a href="https://medium.com/common-reasoning">The Cold Hard Facts</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Sam Altman might go to jail one day]]></title>
            <link>https://medium.com/common-reasoning/sam-altman-might-go-to-jail-2e03543d73cb?source=rss----f5aba979fb4f---4</link>
            <guid isPermaLink="false">https://medium.com/p/2e03543d73cb</guid>
            <category><![CDATA[apple]]></category>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[sam-altman]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Jan L]]></dc:creator>
            <pubDate>Wed, 30 Oct 2024 21:17:10 GMT</pubDate>
            <atom:updated>2024-11-12T17:00:39.944Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nrsgi8yyykoHpxTO9M9IfQ.jpeg" /></figure><h2>Sam Altman might go to jail one day.</h2><p>If important investors start to lose money with OpenAI, Sam Altman might be convicted, just like the other Sam, Bankman-Fried, but especially like Elizabeth Holmes.</p><p>Not because of all the content they “fairly used” from creatives and publishers without permission and all the ongoing copyright lawsuits.</p><p>But in short, if you keep doing investment rounds while duping the investors and the public that your technology will be able to deliver AGI in a couple of years, all the while not investing a penny into any other technology outside of LLMs, a technology that most prominent AI scientists are currently agreeing upon is a dead end towards AGI (even when you “own” all the data in the world.) All the while jacking up the valuation round after round while continuously releasing illusions of progress, fake fabricated results, workarounds and commit flat-out cheating, then you are running a business based on continuous deceit.</p><p>This is no small matter that can be casually brushed aside by Open AI and Anthropic, they are not in an experimental phase anymore.</p><p>This is a combined valuation of 200 billion dollars and a personal enrichment scheme of 10 billion dollars for Sam Altman</p><p><em>OpenAi is now a for-profit company valued at $157 billion following a new funding round that raised $6.6 billion. It nearly doubled its previous valuation of $70 billion earlier in the year. Sam Altman got 7% equity in the company.</em></p><p><em>It is 4X more valuable than its next competitor Anthropic (Also getting totally overvalued after it dropped a 40 billion valuation for its next round), backed by Amazon and Google, it has exactly the same technology and some think a better LLM.</em></p><p>Investors are not spending all this money for a stream of monthly new experimental illusionist tricks their LLMs can do. They are investing billions and probably trillions in the future of AGI that is promised them in a couple of years by Open AI and Anthropic and other for-profit closed AI companies.</p><p>But wait, you say, everybody in Silicon Valley does that. In America, no one goes to jail for promising tech that they then cannot deliver on, even if they cheat and deceive investors. If not a lot of new tech CEOs would go to jail.</p><p>I have one word for you: Theranos.<br>Nobody thought Elisabeth Holmes did anything different and never thought that this could happen, but she ended up in Jail.</p><p>Why?</p><p>She had one product and one technology that didn’t work but kept convincing all her investors that it would work while cheating, releasing fake results, and manipulating the numbers. It all crashed down for her because when that one failed technology was proven to be a dead end, the wrong technology for its purpose, investors lost all their money.</p><p>Well, you might say, that Tesla and Uber promised new technologies, but they failed spectacularly with a lot of them.<br>The difference is that their base products work and make money but more importantly, investors haven’t lost any money yet, on the contrary.</p><p>It is simple: When you lose your investor’s money, you’ll get convicted.</p><p>They also didn’t cheat, Theranos did and OpenAI does.</p><p>Apple realized this when they pulled out of the latest investment round of OpenAI and to cover their tracks released a paper on their findings (without specifically targeting Open AI) with the conclusion that there is 0% evidence of formal reasoning in LLMs.</p><p>https://arxiv.org/pdf/2410.05229</p><p>https://arxiv.org/pdf/2402.19450</p><blockquote>By introducing controlled variations (changing names, numbers, and adding irrelevant clauses) in the GSM-Symbolic benchmark, the researchers exposed the illusion of LLMs’ performance.</blockquote><p>If you read between the lines of this study you can conclude like Apple did, that Open AI and others are contaminating the data by feeding their LLM the questions and answers of benchmark tests.</p><p>When you do an exam or test and you get the questions and answers in advance, what is that called? Cheating.</p><p>When you do that as a company it is called corporate fraud. (Theranos)</p><p>When you do this as a new technology company it is called scientific fraud. (Dong-Pyou Han, HIV vaccine)</p><p>It is that simple.</p><p>So let’s stop giving them the benefit of the doubt.<br>There is no other technology involved than LLMs in their claim of reasoning, thinking, and future AGI.</p><p>The only conclusion you can make from the paper is that they are feeding the LLMs the questions and answers of benchmark tests and other independent tests and then claiming that their LLM is thinking about the question and reasoning to get the answers while just cheating by recalling the memorized Q&amp;As.</p><p>Here is a video of someone explaining the results of the study:</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FtTG_a0KPJAc%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DtTG_a0KPJAc&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FtTG_a0KPJAc%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/18e2af873aa98266f21562d2f3d37027/href">https://medium.com/media/18e2af873aa98266f21562d2f3d37027/href</a></iframe><p>But wait you say, maybe OpenAI is investing in other AI technology than LLMs to get to AGI and reasoning.</p><p>There isn’t any sign of this.</p><p>The most promising groundbreaking AI tech is spatial intelligence.</p><blockquote>Yann LeCun, one of the most prominent AI researchers and Chief AI Scientist at Meta now, has spoken about the future of AI and the need for advancements in spatial intelligence. He believes that for AI to reach human-level intelligence, it needs to develop a better understanding of the physical world, similar to how humans perceive and interact with their environment. The idea of “self-supervised learning,” where AI learns about the world by observing it and inferring relationships without relying heavily on labeled data. LeCun argues that this approach is crucial for advancing AI’s spatial and common-sense reasoning, enabling machines to learn more efficiently and intuitively. <br>(Yann Lecun compared LLM’s to performing “Parlor tricks” and less capable at reasoning than a house cat.)</blockquote><p>The company that is leading spatial intelligence technology is worldlabs.ai founded by visionary AI pioneer <a href="https://www.linkedin.com/in/fei-fei-li-4541247/">Fei-Fei Li</a> along with <a href="https://www.linkedin.com/in/justin-johnson-41b43664/">Justin Johnson</a>, <a href="https://www.linkedin.com/in/christoph-lassner-475a669b/">Christoph Lassner</a>, and <a href="https://www.linkedin.com/in/ben-mildenhall-86b4739b/">Ben Mildenhall</a>.</p><p>They are building Large World Models (LWMs) to perceive, generate, and interact with the 3D world.</p><p>These are their investors:</p><p>Martin Casado and Sarah Wang of Andreessen Horowitz, Scott Sandell of NEA, and Jordan Jacobs of Radical Ventures. Marc Benioff, Jim Breyer, Ron Conway, Jeff Dean, Geoffrey Hinton, Reid Hoffman, Andrej Karpathy, Sound Ventures co-founder Ashton Kutcher, Eric Schmidt, Ram Shriram, Anne Wojcicki, and the late Susan Wojcicki; as well as Adobe Ventures, AMD Ventures, Databricks Ventures, NVentures, the venture capital arm of NVIDIA, Salesforce Ventures, and Shinrai Investments LLC</p><p>The usual suspects, but who is missing from this list?</p><p>Obviously, Meta, Apple, Tesla and Google, but they have been investing heavily in spatial computing for a decade and are probably on the same level as these scientists.</p><p>But you know what I am getting at, the one who really could have benefited from being on this list if he ever needed to defend his honest intentions of delivering a reasoning AGI to his investors.</p><p>Give the guy a break, maybe it grew too fast.</p><p>From the start, he had the best intentions it was not even a for-profit company and they were called open AI, you know, for everyone’s benefit.</p><p>Well, now the entertaining evidence is coming out that Sam Altman and Brockman never even came up with the concept of Open AI, not even the name, and that from the outset they never intended it to be a nonprofit “for the common good of humanity”</p><p>OpenAI didn’t grow into a for-profit closed AI business. The “open” and “nonprofit” were a smokescreen from the get-go to deflect scrutiny, scrape any data for free, and get early investments for free. (Hence Elon Musk’s lawsuit)</p><p>They took the whole concept and the name Open AI from an idea dude, all of this is described in detail in this wonderful article in Bloomberg Businessweek.</p><p><a href="https://www.bloomberg.com/news/features/2024-10-14/why-openai-is-at-war-with-a-guy-named-guy">Bloomberg</a></p><p>The dude describes how they never were a nonprofit from the start and Bloomberg continues with this:</p><p>“When Musk had filed his lawsuit in state court this year (later withdrawn), OpenAI published a blog post purporting to show that Musk knew of the company’s intentions. Included was a private email from Sutskever to Musk in January 2016, suggesting that the “open” in OpenAI was temporary, meaning only “that everyone should benefit from the fruits of AI.” A pledge to share the company’s research openly, it continued, was merely “the right strategy in the short and possibly medium term for recruitment purposes.”</p><p>So with this email, they pretty much admitted their deceit for the “Open” part but what happens with all the money given to the “Non-Profit”, the technology owned by the non-profit and how they are legally going to pull off this very unorthodox transition remains to be seen. (I suspect Elon will get his money back)</p><p>Time will tell, I hope for Sam Altman that he somehow doesn’t create any more disgruntled investors, if not he will be in court for the rest of his career instead of saving humanity from bad AGI.</p><p>In a couple of years when this whole scheme collapses everybody involved will be more informed and educated and will understand the difference clearly between an LLM and a reasoning AGI, including the judge.</p><p>The circumstantial evidence will be overwhelming.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2e03543d73cb" width="1" height="1" alt=""><hr><p><a href="https://medium.com/common-reasoning/sam-altman-might-go-to-jail-2e03543d73cb">Sam Altman might go to jail one day</a> was originally published in <a href="https://medium.com/common-reasoning">The Cold Hard Facts</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[“Swarm” could be OpenAI’s killer feature for corporations]]></title>
            <link>https://medium.com/common-reasoning/finally-swarm-could-be-openais-killer-feature-for-corporations-e92a03a88998?source=rss----f5aba979fb4f---4</link>
            <guid isPermaLink="false">https://medium.com/p/e92a03a88998</guid>
            <category><![CDATA[business]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[llm-agent]]></category>
            <category><![CDATA[aı]]></category>
            <dc:creator><![CDATA[Jan L]]></dc:creator>
            <pubDate>Wed, 30 Oct 2024 21:16:37 GMT</pubDate>
            <atom:updated>2024-10-19T01:27:34.149Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Hio79A7w_FGXbkjDNHzhmg.jpeg" /></figure><p>(It’s Not Why You Think It Is)</p><p>First of all, if you are not sure what an LLM agent or an agentic architecture like Swarm is, or if you should use it, this article explains it well.</p><p><a href="https://towardsdatascience.com/you-dont-need-an-llm-agent-333bf0eb1019">You Don’t Need an LLM Agent</a></p><p>These LLM agents will be just as useful as managers and executives in a company in getting better results… Not very much.</p><p>But I’m afraid LLM agents are not a fad but something that will define the whole LLM implementation in businesses.</p><p>Why?</p><p>Because it represents the inherent structure of corporate America.<br>Managers and executives will understand immediately the usefulness and benefits of implementing this new architecture. This makes total sense!</p><p>When a team or division is not working and you get bad results, what does a typical corporation do?<br>Hire more managers and executives to check, supervise, and reorganize the work instead of replacing or educating its team members.</p><p>So, in this case, if the LLMs are not working properly and you get really bad results, what are they going to do instead of fine-tuning or replacing the AI technology, pipelines, and algorithms to solve the task?<br>Put a manager agent in there that checks for all the results from badly configured agents and prompts (But mostly from “hallucinating”, and just guessing LLM models.) The manager agent then returns or redistributes the work or just creates new agents.</p><p>Then we need executive agents to check if all these manager agents are making the right function calls and decide to return or redistribute the work or create new manager agents.</p><p>Finally, we have a human president to check the final result and try and figure out which prompt or data source is incorrect or which LLM is acting weird in this whole mess of agents. Probably the decision will be made to create a new agentic system with different LLM-modeled agents that check all the original agents&#39; work. And on and on it goes.</p><p>And there you have it: The inevitable blueprint of any corporation in America.<br>(And if you want it or not, what most businesses turn into when they grow.)</p><p>Everybody involved will be happy and very proud of their systems and will give the outside world the illusion that they are smart and working. There will be a whole new industry created of consultants and companies providing these and most of all they will finally save so much money getting rid of all those annoying tech workers.</p><p>But unless you have huge teams of yet unknown geniuses designing them, over time these swarms will fail so spectacularly that businesses will ditch the whole thing and go the old route creating pipelines with humans present in each step.<br>This will mark the end of the bubble and the beginning of the next AI winter, which will be a long one understanding all the trillions wasted to prove a technology that would never work for most of the use cases.</p><p>The only industry that comes close to this utopia that is presented with agentic swarm architecture is one I know very well, TV and Film.</p><p>Where each department delivers the best work in its specialty, all interacting with each other and coming together on the movie set under the direction of a director and producer.<br>But this only works because each trusted department head and each member of their crew make decisions based on an independent vast living experience, closely guarded knowledge, natural intuition, logic, thinking, reasoning, and creativity.</p><p>All the things an LLM will never be able to have or do.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e92a03a88998" width="1" height="1" alt=""><hr><p><a href="https://medium.com/common-reasoning/finally-swarm-could-be-openais-killer-feature-for-corporations-e92a03a88998">“Swarm” could be OpenAI’s killer feature for corporations</a> was originally published in <a href="https://medium.com/common-reasoning">The Cold Hard Facts</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[If you cannot see the world in 3D, you cannot learn from it, interact with it, create in it, or…]]></title>
            <link>https://medium.com/common-reasoning/if-you-cannot-see-the-world-in-3d-you-cannot-learn-from-it-interact-with-it-create-in-it-or-19601b8f156c?source=rss----f5aba979fb4f---4</link>
            <guid isPermaLink="false">https://medium.com/p/19601b8f156c</guid>
            <category><![CDATA[aı]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[generative-ai-use-cases]]></category>
            <category><![CDATA[vr]]></category>
            <dc:creator><![CDATA[Jan L]]></dc:creator>
            <pubDate>Wed, 30 Oct 2024 21:16:18 GMT</pubDate>
            <atom:updated>2024-10-18T23:41:47.165Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>If you cannot see the world in 3D, you cannot learn from it, interact with it, create in it, or reason and think about it.</strong></h3><p>Spatial Intelligence is what AI companies should be working on.</p><p>Plato describes what they are trying to do with AI and LLMs around 360 BC In his “Allegory of the Cave.”</p><blockquote>“The shadows symbolize people’s false beliefs and perceptions due to their limited understanding of the world. Plato suggests that true knowledge and understanding come from recognizing these shadows as mere illusions and seeking the higher reality of the Forms, which represent truth and knowledge.”</blockquote><p>Imagine you grew up with only books to learn from.</p><p>Locked in a windowless empty room and no one to come and visit. You have every book ever published and a copy of everything available online, all the pictures and drawings are replaced by descriptions of the objects depicted in them.</p><p>In other words, your whole life is spent in 1D.</p><p>You have never seen anything these books describe in real life, but you can describe a table and chairs and have read the descriptions of 2D drawings or pictures in certain combinations or settings.<br>You know what they are made of and how people use them.</p><p><strong>Now, you get the task of setting up a wedding arrangement for 200 guests.</strong></p><p>You can search for examples in your books, hoping to find explanations and picture descriptions of this exact challenge.</p><ul><li>Q. How many tables do we need?</li></ul><p>You can easily answer this based on the size and type of the table in your findings.</p><blockquote>A: For a 200-guest wedding dinner, the number of tables you’ll need depends on the type and size of tables you choose:</blockquote><blockquote><strong>Round Tables (typically seat 8–10 people)</strong>:</blockquote><blockquote>If you use 10-person round tables, you would need about 20 tables.</blockquote><blockquote>If you use 8-person round tables, you would need around 25 tables.</blockquote><blockquote><strong>Rectangular/Banquet Tables (typically seat 6–8 people)</strong>:</blockquote><blockquote>If you use 8-person rectangular tables, you would need around 25 tables.</blockquote><blockquote>If you use 6-person rectangular tables, you would need about 34 tables.</blockquote><ul><li>Q: What is the best choice so everyone can see the band and bride, rectangular or square, and how would you position these tables</li></ul><p>You quickly return the first solution you find that mentions visibility:</p><blockquote>A: To ensure everyone can see the band and the bride, a square or rectangular table arrangement is practical. For a 200-guest wedding using 25 tables, here’s how you can position them for the best visibility:</blockquote><blockquote><strong>Square Table Arrangement (5x5 Grid)</strong>:</blockquote><blockquote>Arrange the 25 tables in a 5x5 grid, creating a square layout.</blockquote><blockquote>Place the dance floor in the middle of the grid with the band on one side. This allows most guests to have a direct view of the band, the dance floor, and the head table (where the groom sits).</blockquote><blockquote>Position the head table facing the band across the dance floor so it’s easily visible to everyone.</blockquote><ul><li>Q: Why rectangular?</li></ul><p>You just quote from the page you have already open about rectangular tables:</p><blockquote>A: Rectangular table arrangements are often preferred for wedding receptions for several reasons:</blockquote><blockquote><strong>Maximizes Space</strong>:</blockquote><blockquote>Rectangular tables take up less space than round ones, allowing more room for a dance floor, band, and movement. This can be crucial for fitting everything comfortably in the venue.</blockquote><blockquote><strong>Better Sightlines</strong>:</blockquote><blockquote>With rectangular arrangements, guests can be positioned in rows, which allows more direct views of the band, dance floor, and head table. This setup minimizes obstructed views and ensures that all guests can see the key activities.</blockquote><p>Q: But viewing might be better with round tables, with rectangular tables half of the people are with their backs to the event</p><p>Now you quickly look up the advantages of round tables:</p><blockquote>A: You’re right, round tables do indeed offer better viewing compared to rectangular ones. Here’s why round tables might provide an advantage in this regard:</blockquote><blockquote><strong>No One Has Their Back to the Event</strong>:</blockquote><blockquote>With round tables, everyone faces outward and can easily turn to see the band, the dance floor, or the head table without obstructing their view. In contrast, guests seated at rectangular tables may have their backs to the main event, making it harder fon them to see what’s happening.</blockquote><blockquote><strong>Flexibre Seating Orientation</strong>:</blockquote><blockquote>Round tables allow more flexibility in orienting guests so they can all enjoy the key moments. Since there’s no “head” of a round table, it’s easier to arrange the seating to maximize visibility for everyone.</blockquote><blockquote><strong>Equal Viewing Angles</strong>:</blockquote><blockquote>The circular arrangement naturally provides equal sightlines from all directions. No matter where guests are seated at the table, they will have a similar angle to view the events, creating a more balanced experience.</blockquote><blockquote><strong>Encourages Movement and Interaction</strong>:</blockquote><blockquote>With round tables, guests can more comfortably shift their chairs to face the action, whether it’s the first dance or a speech. This flexibility promotes a dynamic atmosphere where people can stay engaged.</blockquote><blockquote>For these reasons, round tables are often a preferred choice to ensure all guests can participate in and enjoy the key moments at a wedding.</blockquote><p>Q: Then why did you recommend rectangular tables for better sightlines?</p><p>Now you just bullshit your way out of it trying to make this annoying query end.</p><blockquote>A: I initially recommended rectangular tables because they can be practical for certain setups and offer some specific advantages, such as maximizing space and providing flexible layout options. However, I see your point about the viewing challenges with rectangular tables, and round tables do indeed have significant benefits for wedding receptions, especially in terms of visibility and social interaction.</blockquote><blockquote>Round tables are typically the better choice when ensuring that all guests have clear sightlines to the event is a priority. They allow for an inclusive seating arrangement where no one has their back to the action, enhancing the viewing experience for everyone.</blockquote><p>Q: Make me a seating plan for rectangular and round tables based on these findings so I can have a look and choose between them.</p><blockquote>A:</blockquote><blockquote>Rectangular wedding seating:</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Pr5RVO0C9bvL2qCcImrWHg.jpeg" /></figure><blockquote>Round wedding seating:</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*F_9ByTy08PP6ARpZWkrdsg.png" /></figure><p>Q: That’s more than 70 8-person tables I count in this plan, I asked for seating for 200 guests.</p><blockquote>A:</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*iOxfdxSWvT8VZk2bwYkgQA.png" /></figure><p>(All the answers were provided by Chat GTP 4o, Dall E 3, and the last two by Flux.)</p><p>The reality is that you may have all the data available in the world, but you will not be able to figure out how to reason about a simple table arrangement or create a plan for it if you have never seen a wedding or any dinner arrangement in 3D.</p><p>Lucky for us, during this illusionist LLM hype, there are still smart people who are truly innovating AI.</p><p>These are the people who are right now working on a more valuable technology than LLMs:</p><p><strong>Spatial intelligence.</strong></p><p><a href="https://www.worldlabs.ai/">https://www.worldlabs.ai</a></p><p>You will have to be a bit more patient, though, hopefully, somewhere before the end of our lifetime, we could see some true humanity-changing applications.</p><p>Here are a couple of Fei-Fei Li videos</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2Fy8NtMZ7VGmU%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3Dy8NtMZ7VGmU&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2Fy8NtMZ7VGmU%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/6e7ea44e93749f9cacef674fe390a9b6/href">https://medium.com/media/6e7ea44e93749f9cacef674fe390a9b6/href</a></iframe><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FvIXfYFB7aBI%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DvIXfYFB7aBI&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FvIXfYFB7aBI%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/0543aedeb8a276c91f3c048bd60a6f4c/href">https://medium.com/media/0543aedeb8a276c91f3c048bd60a6f4c/href</a></iframe><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=19601b8f156c" width="1" height="1" alt=""><hr><p><a href="https://medium.com/common-reasoning/if-you-cannot-see-the-world-in-3d-you-cannot-learn-from-it-interact-with-it-create-in-it-or-19601b8f156c">If you cannot see the world in 3D, you cannot learn from it, interact with it, create in it, or…</a> was originally published in <a href="https://medium.com/common-reasoning">The Cold Hard Facts</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[“New” media has become the QVC channel.]]></title>
            <link>https://medium.com/common-reasoning/new-media-has-become-the-qvc-channel-719d9c71cf3e?source=rss----f5aba979fb4f---4</link>
            <guid isPermaLink="false">https://medium.com/p/719d9c71cf3e</guid>
            <category><![CDATA[online-media]]></category>
            <category><![CDATA[entertainment]]></category>
            <category><![CDATA[social-media]]></category>
            <category><![CDATA[entertainment-industry]]></category>
            <dc:creator><![CDATA[Jan L]]></dc:creator>
            <pubDate>Wed, 30 Oct 2024 21:15:50 GMT</pubDate>
            <atom:updated>2024-10-19T01:45:19.039Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dQeEn7vNF86S32rHpafK3A.jpeg" /></figure><p>When I was watching Shark Tank many years ago, and I saw Lori Greiner, I thought, wait, who in the hell watches QVC anymore? How can this still be a viable business that she is invited on this show?</p><p>So many years later the whole of online entertainment and the “new” media made by independent “creators” has evolved into QVC.</p><p>What a sad affair.</p><p>We used to call creatives “sellouts” when they aligned their work and talent with corporations.</p><p>Now “creator” is the new definition of a “sellout”.</p><p>(By the way, even without sponsors please stop calling yourselves “creators”, maybe a better term is “facilitators” or in the best-case scenario “curators”. )</p><p>Too many podcasts sound like the radio shows of the 50s: Spoken infomercials, <br>the majority of videos on YouTube and TikTok are the home shopping channels of the 70s,<br>and a lot of articles, blogs, newsletters, and social media I come across are advertorials or disguised pure self-promotion.</p><p>I remember Steven Spielberg saying with the advent of YouTube and digital cameras that the whole movie industry is now democratized. “There is no financial excuse anymore for anyone to make great content.” He saw YouTube&#39;s great potential as a platform for new filmmakers to gain exposure.</p><p>What people did with that… Influencers, affiliates, and Mister Beast. Replicating the lowest/cheapest kind of television ever produced.</p><p>I have nothing against critical product review shows, but everything against making a cheap video just because you have found a sponsor.</p><p>I used to watch Top Gear, they were often critical, but by definition, it probably was still an infomercial showing off some of these cars, but it was made with a big budget from BBC and later from Amazon.<br>The show was not made because BMW paid for it.</p><p>Yesterday, I watched a video for the new Toyota Land Cruiser on YouTube where a guy just read the window sticker and reviewed the car driving it up and down the dealer’s parking lot.<br>This was the first recommended review video on YouTube for this car.</p><p>I work in entertainment and during my career, we used to joke that with our shows we were just providing the fillers between commercials.</p><p>Online, the commercials are the shows.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=719d9c71cf3e" width="1" height="1" alt=""><hr><p><a href="https://medium.com/common-reasoning/new-media-has-become-the-qvc-channel-719d9c71cf3e">“New” media has become the QVC channel.</a> was originally published in <a href="https://medium.com/common-reasoning">The Cold Hard Facts</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Can AI Reason, Think, or Create? Let’s Ask Directly]]></title>
            <link>https://medium.com/common-reasoning/can-ai-reason-think-or-create-lets-ask-directly-557a2fcfab75?source=rss----f5aba979fb4f---4</link>
            <guid isPermaLink="false">https://medium.com/p/557a2fcfab75</guid>
            <category><![CDATA[think]]></category>
            <category><![CDATA[reason]]></category>
            <category><![CDATA[aı]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[create]]></category>
            <dc:creator><![CDATA[Jan L]]></dc:creator>
            <pubDate>Wed, 30 Oct 2024 21:15:29 GMT</pubDate>
            <atom:updated>2024-10-28T03:35:18.684Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*F9KYbT_CkCN06-X3" /></figure><h3>AI: The Greatest Illusionist</h3><p>If you are still in doubt whether LLMs can reason or will ever be able to, and are not sure if you can believe OpenAI and Anthropic’s sales pitch that they will have AGI in a couple of years,</p><p>Let’s just ask ChatGPT itself.</p><p>(Answers provided by GTP 4o and cover image by Dall E 3)</p><h3>Reasoning</h3><p>Q: What is the definition of reasoning?</p><blockquote>A: Reasoning is the cognitive process of <strong>thinking about information in a logical way to form conclusions, make judgments, or solve problems</strong>. It involves analyzing facts, identifying relationships, and using principles or rules to draw inferences or reach a decision.</blockquote><p>Q: Everything an LLM is not capable of</p><blockquote>A: LLMs cannot reason as <strong>they lack genuine understanding, consciousness, and the ability to perform multi-step logical thinking</strong>. Their responses are based on patterns in training data, not on real comprehension or novel insights. Additionally, t<strong>hey cannot perceive the physical world, remember long-term context, or make moral judgments.</strong></blockquote><p>Q: So they give the illusion of being able to reason?</p><blockquote>A: <strong>Yes, LLMs create the illusion of reasoning</strong> by generating responses that mimic logical thinking based on patterns in their training data. However, they do not understand or engage in real reasoning; they simulate it without comprehension or awareness.</blockquote><h3>Thinking</h3><p>Q: what is the definition of thinking?</p><blockquote>A: Thinking is the mental process of <strong>considering, reasoning or reflecting on ideas, concepts, or information</strong>. It involves forming thoughts, making decisions, solving problems, imagining scenarios, and evaluating or analyzing different perspectives. It plays a crucial role in understanding and interacting with the world.</blockquote><p>Q: So everything an LLM is not capable of?</p><blockquote>A: Large language models (LLMs) like me are <strong>capable of simulating certain aspects of thinking</strong>, such as processing information, generating responses, and performing logical or mathematical tasks.</blockquote><blockquote>In essence, LLMs can simulate certain thought processes <strong>but lack the consciousness, intentionality, and subjective experience that characterize thinking</strong>.</blockquote><p>Q: So an LLM gives the illusion of thinking?</p><blockquote>A: <strong>Yes, that’s a good way to put it — LLMs create the <em>illusion</em> of thinking</strong>. They generate responses that appear thoughtful or intelligent because they are trained on vast amounts of text data and can mimic patterns in human language. However, what they do is fundamentally different from actual thinking because:</blockquote><blockquote><strong>Pattern Recognition vs. Understanding</strong>: LLMs generate responses based on recognizing patterns in the data they were trained on, without understanding the meaning behind the words. <strong>They don’t comprehend concepts or have knowledge in the way humans do</strong>.</blockquote><blockquote><strong>No Awareness or Intent</strong>: When an LLM responds to a question, it’s not aware of what it’s doing. <strong>It’s not forming goals, pursuing intentions, or even “trying” to be correct</strong>. It generates responses based on probability, not conscious deliberation.</blockquote><blockquote><strong>The responses may give the impression that the LLM is thinking, reasoning, or even understanding complex ideas</strong>, but it’s essentially a sophisticated pattern-matching system that lacks any true awareness or cognitive experience.</blockquote><h3>Creativity</h3><p>Q: What is the definition of creativity?</p><blockquote>A: Creativity is the ability to generate new and original ideas, solutions, or products by thinking in novel ways or combining existing concepts in unique ways. It involves imagination, innovation, and <strong>the capacity to transcend traditional patterns of thinking</strong>. Creativity is often seen in problem-solving, artistic expression, and scientific discovery, where it leads to original and valuable outcomes.</blockquote><p>Q: Everything an LLM is not capable of?</p><blockquote>A: That’s an interesting perspective! Large Language Models (LLMs) like me are indeed designed to generate responses based on patterns and data we’ve been trained on, but there are key differences between this and human creativity. While LLMs can mimic creative processes — such as producing poetry, stories, or problem-solving solutions — <strong>we typically rely on existing knowledge rather than true originality or intuition.</strong></blockquote><blockquote><strong>Human creativity often involves subjective judgment and the ability to think abstractly in ways that break the boundaries of existing knowledge</strong>, which LLMs can’t replicate. Creativity also includes personal experience and spontaneity, things beyond the reach of machine learning models.</blockquote><blockquote>In that sense, creativity does seem like a uniquely human trait.</blockquote><p>Q: So it gives the illusion of being creative?</p><blockquote><strong>A: Yes, exactly. What LLMs provide is more of an “illusion” of creativity</strong> rather than true creativity itself. They can generate responses that appear original or imaginative by drawing on vast amounts of data, and identifying patterns. However, this process doesn’t involve conscious insight, intention, or the kind of subjective experiences that drive human creativity.</blockquote><blockquote>The outputs may seem creative because LLMs can mimic the style, tone, or structure of creative works. But the underlying mechanism is fundamentally different from the way humans create; <strong>it’s more about sophisticated pattern recognition and replication</strong> than the spontaneous, often unpredictable nature of human creative thought.</blockquote><h3>Conclusion</h3><p>Q: Will LLMs be able to bring us closer to AGI?</p><blockquote>A: LLMs are unlikely to achieve AGI (Artificial General Intelligence), as they primarily simulate reasoning without true understanding or cognitive abilities. While <strong>they can process language in ways that appear intelligent</strong>, advancing toward AGI would require integrating other capabilities, such as <strong>real-world perception, long-term memory, and genuine cognitive reasoning processes beyond pattern recognition.</strong></blockquote><blockquote>AGI aims to replicate or surpass human cognitive abilities across a wide range of tasks, including creative problem-solving, autonomous learning, and adaptive behavior in novel situations. LLMs, however, are inherently bound by the data they are trained on and <strong>can’t independently form new experiences or insights. They excel in narrow tasks</strong> within their training data’s boundaries but <strong>struggle with the kind of understanding and flexible thinking</strong> that characterizes human intelligence.</blockquote><blockquote>Here are some reasons why LLMs do not bring us closer to AGI:</blockquote><blockquote><strong>Lack of Understanding and Consciousness</strong>: AGI would require not just the ability to generate responses but a <strong>deep understanding of the world, concepts, and abstract thinking</strong>.</blockquote><blockquote><strong>No True Learning from Experience</strong>: AGI should be able to <strong>learn continuously from new experiences and adapt over time</strong>, similar to how humans learn from their environment. <strong>LLMs, on the other hand, are static after training</strong>; they don’t update their knowledge based on <strong>new interactions or experiences</strong> in the way humans do.</blockquote><blockquote><strong>Absence of General Reasoning Abilities</strong>: Human intelligence involves integrating different types of reasoning, such as causal reasoning, spatial reasoning, and planning. LLMs are good at mimicking certain reasoning patterns found in their training data but <strong>don’t possess a generalized ability to reason across diverse domains</strong>.</blockquote><blockquote><strong>Inability to Form Intentions, Goals, or Autonomy</strong>: AGI would be expected to form its own goals, pursue tasks with intention, and exhibit some form of autonomous behavior. <strong>LLMs generate responses without any internal goals, motivations, or understanding of purpose</strong>.</blockquote><blockquote><strong>Challenges with Non-Textual Knowledge</strong>: While LLMs are adept at handling language, <strong>AGI would need to integrate multiple forms of knowledge and sensory inputs (e.g., visual, auditory)</strong> to understand and interact with the world in a more holistic manner. LLMs, by design, primarily work with text data.</blockquote><h3>Finally</h3><p>So we heard from ChatGTP itself what it is and will be capable of.</p><p>But we also saw what it is very good at: Common knowledge, Research, and Brainstorming.</p><p>And that’s what I use it for extensively, it is a godsend, how many times have I wished I could ask questions and brainstorm with Google search to get insights and the right info or wished I could brainstorm in my search box with all my documents and files?</p><p>But is this worth more than 200 billion dollars at this moment (OpenAI + Anthropic’s valuations)?</p><p>Something you will get for free from any software you already have (Apple, Google, Microsoft,…) or even 100% for free to run and adapt on your computer from Meta, google, and other open-source LLMs.</p><p>I don’t think so, and investors should maybe just ask and brainstorm with ChatGTP itself about this before investing in it.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=557a2fcfab75" width="1" height="1" alt=""><hr><p><a href="https://medium.com/common-reasoning/can-ai-reason-think-or-create-lets-ask-directly-557a2fcfab75">Can AI Reason, Think, or Create? Let’s Ask Directly</a> was originally published in <a href="https://medium.com/common-reasoning">The Cold Hard Facts</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Magic of AI]]></title>
            <link>https://medium.com/common-reasoning/people-love-magic-tricks-and-always-will-48f40bbca373?source=rss----f5aba979fb4f---4</link>
            <guid isPermaLink="false">https://medium.com/p/48f40bbca373</guid>
            <category><![CDATA[aı]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[illusion]]></category>
            <category><![CDATA[entertainment]]></category>
            <dc:creator><![CDATA[Jan L]]></dc:creator>
            <pubDate>Wed, 30 Oct 2024 21:15:01 GMT</pubDate>
            <atom:updated>2024-10-28T07:15:23.235Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/620/1*ex9jbDmqCAp9BIJaTkob0A.jpeg" /></figure><h3>AI, the Greatest Illusionist, Part 2</h3><h3>The illusionists</h3><p>For centuries, people have paid good money for a guy to deceive them right to their face.</p><p>We all know that illusions and magic tricks are fake, and sometimes we even know how the magic works, but we keep watching and will come back and pay again.</p><p>We are entertained.</p><p>We believe our eyes.</p><p>When we see something, we instinctively believe it to be true even if we know it isn’t.</p><p>And that self-deception a lot of times is very entertaining.</p><p>Even in science: The giant rising moon illusion.</p><p>The Moon appears larger near the horizon than when it’s overhead, but this is purely a perceptual illusion — measurements show the Moon’s size doesn’t actually change.</p><blockquote>Scientific American:</blockquote><blockquote>The illusion has been known since ancient times (documented as far back as the 7th century BCE) There have been many theories since then, but it is caused by the Ponzo illusion: Our brain interprets objects near the horizon as being farther away due to converging perspective lines, so it compensates by making them appear larger.</blockquote><blockquote>We experience the moon in the sky as closer than the moon on the horizon (which we perceive as further away than the sky).</blockquote><blockquote>This illusion is so powerful that even knowing about it doesn’t prevent us from experiencing it — astronomers and other experts still perceive the Moon as larger near the horizon despite understanding the science behind the effect.</blockquote><p>Indeed, well aware, I still marvel every time at the magic of a gigantic moon rising above the horizon and will go out of my way to experience it.</p><p>ChatGTP itself explained to us that the results we are seeing from LLMs are illusions of thinking, reasoning, and creativity.</p><p><a href="https://medium.com/@xwggsphc/can-ai-reason-think-or-create-lets-ask-directly-557a2fcfab75">Can AI Reason, Think, or Create? Let’s Ask Directly</a></p><p>But knowing this, most of us will continue to play around with it and each time will marvel at the instantaneous magical results it provides and the uncanny human reasoning and thinking behavior it shows.</p><p>The illusion is hard to resist and entertaining.</p><p>Even when the results are sometimes not working or are a bit lame, even if we understand the trickery behind the scenes, we’ll spend hours trying to get that instant gratification every time we submit a prompt. Instantly giving us the entertaining illusion we have had a personal smart human conversation with AI and seeing the result of it instantly with our own eyes in the original thing we created.</p><p>Investor Marc Andreesen’s partner Martin Casados recently said:</p><blockquote>Probably the use case for GenAI working the most is creative composition. Many companies in this space are growing as quickly as we’ve seen anything growing.</blockquote><blockquote>The second use case is: Companionship.</blockquote><p>Movies and television are still the number one magic trick. While by now everyone knows how they are made, that they are an illusion of reality or science and fake (especially reality shows), we sit back and put on our suspension-of-disbelief glasses as soon as we press play. This is what we call entertainment.</p><p>Social media was a new magic trick; especially Instagram. Everyone knows very well most of the pictures in our feeds are fake or faked, we don’t care, we are entertained and addicted to it.</p><p>Any social media company is in essence an entertainment company. We know very well we don’t have 500 or more friends or connections, but we are entertained by the illusion we are popular or important and everyone is constantly watching (following) us while giving us congratulations for anything we do or write.</p><p>Web3 is the gaming magic trick; everybody knows it is an illusion, it is fake money, and NFTs are game tokens. But the community, the thrill, and the buy, sell, trade game kept many up all night. Who wouldn’t want to believe the illusion — even for a short time — that we could mine money?</p><h3>GenAI as entertainment</h3><p>One of the greatest illusionists of our time could be GenAI.</p><p>The magic is extraordinary and very entertaining.</p><p>We love the illusion and instant gratification that with just a couple of prompts, we can magically become as good as great writers, painters, sketch artists, designers, movie makers, musicians, voice artists, or developers,…</p><p>On top of that, the illusion we can have human conversations with AI and find a sense of companionship with it could make this a killer new entertaining form.</p><p>When we are playing, we truly believe the illusion that there is an art or science in prompting and we can magically become Picasso, Shakespeare, or Einstein (or have conversations with them) without having to go through the annoying part of learning and practicing a craft or studying and developing a theory or observing and creating a vision of the world we live in.</p><p>People truly want to believe the illusion that they, just like, right now — instantly, created themselves the result GPT returns; some even think they made art or found a novel way of thinking. They believe their eyes. This is instant gratification on steroids and has the potential to become one of the most addictive entertainment forms ever.</p><p><em>Note: I am not reflecting on professionals and creatives using GenAI to edit and improve their work or even create original work.<br>This thought process is more about the Canva creatives than the Adobe creatives.</em></p><p>For the hardcore fans of GenAI, it is gaming.</p><p>With hordes of mostly guys trying to figure out the magic key item to get to the next level. They are spending days and nights adjusting pages of prompts figuring out how to play the model, trying to succeed in getting AI to do the impossible, find the mystery elements, and boast about it.</p><p>They love the magic tricks of AI and love the community discussing their grinding, finding collectibles, secrets, and easter eggs with their prompts all day, and anticipating and deciphering lore drops… Strawberry?</p><p>They are entertained.</p><h3>The Business</h3><p>OpenAI and now also Anthropic are setting the trend of releasing a regular stream of new tricks their AI can do, trying to keep their audiences entertained.</p><p>For now, they are succeeding wonderfully at this… They just are not making any money with it yet.</p><p>But this could become very profitable; media, social media, games, and fanfare are big business.</p><p>But will it be to the tune of 157 billion dollars for one company (OpenAI) with one product?<br>Time will tell, it’s going to be a tough sell with only a couple of tricks, only a few games, and no networking effect like social media.</p><p>Paramount (includes Viacom, CBS, streamers, a legendary movie studio with its huge catalog and many games based on their IPs) was just valued and bought for 8 billion dollars.</p><p>Nevertheless, OpenAI and others could still take a good slice of the entertainment industry if they can continue releasing their magic tricks on a schedule and wowing people.</p><p>The problem is that all the current investors in OpenAI or Anthropic are not investing in an entertainment company. Most are allergic to it.</p><p>They don’t want to be involved in a brittle industry.<br>To get beaten by yet another magician that comes along with a newer/better trick (e.g. Notebook LM’s podcast)</p><p>Investor Marc Andreessen recently said</p><blockquote>AI model makers are in a ‘race to the bottom’<br>Andreessen compared current AI development with “selling rice,” meaning that there isn’t much product differentiation. “It turns out anybody can make an LLM,” he said, invoking commodification.</blockquote><p>A better analogy is comparing AI development with selling entertainment, everybody can make entertainment; there isn’t much differentiation, just a few new and different tricks, marketing, and chance. Only a few make a return on their investment if they are lucky.</p><p>Most entertainment dominance doesn’t last long.<br>Usually, until a magician with a new trick comes along.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=48f40bbca373" width="1" height="1" alt=""><hr><p><a href="https://medium.com/common-reasoning/people-love-magic-tricks-and-always-will-48f40bbca373">The Magic of AI</a> was originally published in <a href="https://medium.com/common-reasoning">The Cold Hard Facts</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>