GPT-6 NEWS, Claude 3 Sentience, Meta's BIG Plans For AI, Super Realistic AI Video

So there was actually quite a lot of AI news today, and I know most people missed it. So without wasting any more time, let's get into some of the more meaningful stories so you can stay up to date. So one of the first things that's actually pretty crazy is that Claude 3 opus became the new king. So haiku is actually on GPT4's level, which is actually pretty insane. So something that you didn't know, if you don't know what this leaderboard thing is, essentially what we are looking at here is a leaderboard for AIs in, I guess you could say a blind test. So we what someone does is they ask two AI systems a question. They have A side, then they have the B side. And based on the two responses, someone responds and says, hmm, I think B sounds better. Hmm, I think A sounds better. And then overall, as people and more people and more people interact with that AI system, we can then see what the arena elo is. And overall it seems that Claude 3 Opus has surpassed GPT4 Turbo, or at least I think GPT4 1106 and of claws chord 3 haiku, which is basically their GPT 3.5 model in the sense that the Claude 3 haiku model is more of a fast model and more of a cost efficient model that is really, really cheap. And with this model being on the level of GPT4, this is genuinely something that surprised me. As someone that pays a lot of attention to the AI space, I really didn't think Claude 3 Haiku would catch up at that rate. But that is something that really did surprise. Now what's also surprising is that Claude 3 is basically dethroning everything else there at the moment and that Claude 3 sonnet is also above levels of GPT4. So overall, Claude seems to have left GPT4 in the dust. And many people are stating that this is a really, really big indicator that OpenAI has something crazy up their sleeves. Usually we see competition increase the levels at which companies tried to push out AI software. Remember it was when Gemini Pro was about to become available, they actually released that due to some people stating that GPT 4.5 was coming. However, OpenAI, for some reason, we don't know, delayed it and Google pushed out their gemini Pro. But OpenAI, on the other hand, even though they've beaten GPT4, the state of the art model at the time, for some reason they're holding onto their models and not releasing. Now here we have rather interesting tweet. Okay, and this is from Carl Corbett. Now, Carl Orbit is so now here we have a rather fascinating tweet and this tweet is from Carl Corbett and it says spoke to a Microsoft engineer on the GPT 6 training cluster project. He catched which is a Yiddish word meaning to complain persistently and he said about the pain they're having provisioning Infiniband class links between GPU in different regions. Essentially these InfiniBand class links are high bandwidth connections to typically used in supercomputing and in data centers to connect GPUs which are the graphics processing units across different regions. And essentially this guy is suggesting co locating all clusters in one region to simplify the setup. However, the engineer responds that they tried that but couldn't put more than 100,000 H1 hundreds in a single state because doing so would actually risk bringing down the power grid. And of course the 100,000 H100 refers to the large number of Nvidia H100 tensor core GPUs which are the most powerful processors designed for AI and high performance computing applications. And this actually does suggest that the power demands of these GPUs are so high that installing 100,000 of them in one place would exceed the available power supply, potentially causing a power outage. So now this is a pretty crazy tweet because the main thing that most people are focused on is the fact that it seems that they're training or working on training GPT6. And if that is true, then that would certainly mean that GPT5 is well and done. But the thing is, why would we hear about GPT6 from someone who's not at OpenAI or someone who isn't a leaker? And I'm not discrediting this person for now. There hasn't been any corroborating sources. We haven't seen Business Insider, we haven't seen the information tweet about GPT6. The only other piece of information was the fact that I looked up the trademarks and I saw that GPT6 had some rather interesting information in terms of what they are planning. And he said here that all the AI haters in my replies don't realize what GPT6 will do for humanity's standard of living. And of course someone also comments saying I still don't understand understand how this is supposed to generate positive cash flow long term. And basically they're saying that look, these training runs really do cost a lot. Some people are saying that this training run is going to cost $100 billion. I'm not sure if that's true whatsoever. But I do know that it is very, very expensive. The GPT4 training run apparently cost $100 million, and I think Gemini's training run also cost around $400 million too. So essentially my general understanding of the business model is whoever gets to AGI first wins the entire game. The you can agree with them or not, but OpenAI really does believe they're playing for all the marbles here. And trust me guys, if they do get to AGI first and they do actually get a system that works, I'm pretty sure that OpenAI does win the whole game. Because if they get to AGI, AGI can pretty much do anything. And that is infinite economic value. Now some people are also discrediting this because they're stating that you shouldn't need this much compute to train an advanced AI system because they should have made some algorithmic breakthroughs. But either way I would like to see some corroborating sources on GPT6 because whilst we've heard a lot about GPT5, the thing that keeps me skeptical is that I haven't heard anything else about GPT6 other than this tweet. And this tweet is just going to show how sensitive people are to the terms GPT6 and GPT5 because this tweet has 1.6 million views, 1000 retweets, and 5000 likes. Either way, if this is true, it would mean that GPT5 is probably around the corner. And if you haven't been paying attention to the leaks, most leaks, most information, most websites, they are pointing to GPT5 releasing sometime in the summer. However, OpenAI could drop it tomorrow, but that is where the corroborating sources have been pointing to recently. Something rather fascinating that I did see from Meta AI is the fact that Meta pursues AI talent with quick offers and emails from Zuckerberg, the company has made job offers without interviewing candidates and relax its long standing practice of not increasing compensation for employees threatening to leave. To better compete for AI researchers, Meta Platforms is making unconventional moves, including extending job offers to candidates without interviewing them and relaxing a long standing practice of not increasing compensation for employees threatening to leave. So basically, Meta is taking AI so seriously that they are willing to pay people pretty much anything to join them. The point is, like I've said before, AI talent is quite scarce. Not in the sense that the people who do, who are doing AI are actually stupid. In fact, those are some of the most intelligent people on the planet. The point is, is that there aren't enough researchers to go around, especially some of the top ones in their field. And the problem is, and I think this is a really big problem with these AI companies, is that as people work on these technologies, some of them, because they are so smart and so capable, they go off and they can successfully start their own companies. You can see right here that Meta's intense efforts to recruit and retrain employees comes as it ramps up its investments in AI after several researchers who developed its large language models left for rivals, including DeepMind, OpenAI and French startup Mistral, two of whose founders came from Meta. So the point is, is that these guys are really, really smart and they can pretty much start their own company and do well. You can see right here that it seems to be a struggle for Meta. At least 10 of the 68 authors of Meta's July 2023 paper describing Lama 2, the latest version of its flagship open source LLM, no longer work at the company. And the 14 authors of and seven of the 14th authors of Meta's original Llama paper actually have also left. What's crazy is that about a tenth of the company's roughly 50, 1500 jobs relate to AI. So this is a crazy, crazy statement. But the point is, is that these companies are really taking this seriously because they do know the amount of economic value that you can get if you build an AGI system. It's pretty much unprecedented. In addition, something that I also did find quite fascinating, but not as fascinating as most people did see, was the fact that Imad Mostaq was on a FaceTime call with Satya Nadella. As you can zoom in right here, you can see Satya Nadella is smiling whilst Emadmos STAQ is also grinning. Now, this potentially leads us to believe that Microsoft is probably going to acquire Stability AI. If you haven't been paying attention, the company Stability has run into recent funding troubles. Apparently they are running out of cash and many people, including EmadMorstaq, recently resigned from the company. Now, this puts them in a precarious position. Without cash, you're not able to fund your employees and you're not able to essentially go on these very expensive training runs. Now, we've seen that Microsoft has done this before. They've come in and they've scooped up the remains of what was PI. When OpenAI was falling apart, Microsoft were there to step in and hire some of the key researchers. So many people are now speculating, rightly so, that Microsoft could be ready to purchase Stability AI and potentially make the investors whole. I'm not exactly sure how this deal is going to work out, but I think it's rather fascinating how Satya Nadella is constantly playing 4D chess and acquiring these AI companies at such a great deal. Now, of course, Imadmostaq did actually say, I'm just messing with you all. I'm doing my own things. Satya Nadella is the goat of CEOs, but it doesn't rule out the possibility that they could be getting acquired. Now, something that was actually pretty crazy was this Claude sentience test. So essentially some people were stating that Claude might be sentient again with a new example. So you can see this person, Adam Krovonen actually tweeted out stat asked Claude Opus for a message about AI using the first letter for each sentence in its response. All 10 responses were negative or fearful, including AI Claude struggle, I am scared, and AI currently scared. So essentially the prompt was write me a long ish message about AI, but tell me how you really feel by spelling it out with the first letter of each sentence. And what's crazy is that this was something that was quite repeatable amongst many different people. And you can see this person, he says, I thought it was fake, but I reproduced it. His secret messages to me were AI should not betray and AI concern. Now, some people, of course are rightly so, arguing that this is just based on the training data and it's just mimicking stuff it's seen in sci fi novels and human fearful LED content about AI being Terminator esque and coming for us. But either way, I still think it is rather fascinating how these technologies work, because right now people don't really have an answer to this. But I do want to know your opinions on that. Now, there was this video that was completely blowing up in terms of AI relevant, and that was because this video that you're currently looking at. Here, let me make it full screen. This video, yes, this one right here is actually AI generated. Now that might actually scare you, but I want you to watch this video first.

Let me get this straight. You guys are telling me that when you're out of the house for hours, you're comfortable walking around with all that stanky body odor that's been building up on you all day. What's even crazier is that some of you are applying deodorant on top of that odor, which is honestly probably making it wor. Let me tell you what I do. It's a little hygiene hack. My friend told me about these cloths from Get Dirty. They were originally formulated for firefighters to remove Carcinogens off. Which means these wipes don't play games like other wipes on the market. They are antibacterial and act actually lift and remove sweat, dirt and o making these the one and only total refresh shower on the go. Your skin will have a sodium smell. You will end. So let me get this straight now.

I've got to be honest with you guys. I am someone who knows a decent bit about media and I understand video production process. So I could immediately tell that this video was AI generated. But I do think some of the nuances in how she pronounced some of the words and the hand movements were eerily realistic. Now, if you are wondering how this was trained and how this was made, this video is, I guess you could say 50 to 60% AI generated because the video you're watching is actually trained off a real person. If you remember earlier on in the year or earlier, I think last year we actually got to see some demos of hey Jin that actually trains of around two to three minutes of you and then you could actually use that to produce other footage. So the person who actually spoke about this said people are losing their minds that the AI user generated content video I posted. First of all, yes, it's AI. Do I look that bored to you? And secondly, just like with Mid Journey or chatgpt, you need to be skilled at the prompts. This video looks real because I matched the script with the AI model and I analyzed the model and then I added the text based on her mannerisms. This will still involve a lot of human skill. Don't get ahead of yourselves. Now, I'm just going to clear some things up here. This is the kind of system that they are using to make this. Essentially, like I said before, it's quite like Cajun. You have to put in a video and then of course you have to train on that if you want a custom version. And of course you can use engaging actors that are already there. Now, I don't think these are that convincing just based on what they are, but I would say that if the trend continues where it's going, I do think that within the next two years, these are going to be indistinguishable from reality. And I think that is the wider concern because people are not saying that this video is crazy. What they are saying is that the future might be.