AI News : Google Gemini -2 Released? New Google Robots, Sam Altman Sued, Secret Details Revealed

So, welcome back to the AI grid. In today's video, we'll be taking a look at some of the most important stories in regards to AI. And this video, I think is rather important because this week is probably going to be a rather big week in AI, if any of the rumors slash leaks have any credibility whatsoever. And previously they have had a decent amount. Now, essentially what we can look at here is we can see that, that currently we're looking at the Chatbot Arena. Now, the Chatbot arena, as many of you may know, is an area where you input one prompt and it's randomly sent to two separate models and then you select which response you think is better and then they reveal which prompt, which model. Now, interestingly enough, funnily enough, we have seen in the recent couple of days that there has been this mystery model popping up in only the Chatbot arena area. And guess what this model is called? This model is called Mystery Gemini 2. Now, it seems that this model may have been recently trained, slash, just recently finished training, because one of the questions that this person asked was a question that was only recently circulating Twitter. This is of course the infamous Strawberry question. How many Rs are there in strawberry? There are three Rs in the word strawberry. And, and interestingly enough, Gemini 2 says it appears that in the past some AI models have struggled with this question, but I can confirm that the correct answer is three. So this seems that maybe the model is something that can browse the Internet, or maybe it's a model that was just recently trained and understands the recent data regarding the Strawberry controversy. If you aren't familiar, the Strawberry controversy is because people recently found out that it. But if you ask an LLM to actually count the number of Rs in the word Strawberry, it can't really do. So this is something that I think most people haven't figured out yet. Now this is also rather fascinating because not only is Gemini 2 only available in the Chatbot arena, meaning that potentially this could be released earlier this week or sometime this year. As I spoke about a few days ago, Google's new model is actually ranking number one across all of the chats. The arena ELO is currently above GPT4O and other state of the art models. We can see that even after 12 days from that time, you can see that this is still the number one model regarding the arena score. And you can see that it's received a decent amount of votes. And this model is using a different kind of reasoning engine. Now, if you do want to try this model. Like I said before in the previous video, this is something that is available in Google's AI studio. But this isn't the only fascinating piece of information, especially regarding Google. What we can also see here is that there are three new very capable models on the LMSYS Chatbot Arena. Now, these are quite hard to see because they're only in arena mode, meaning that unless you ask the chatbot arena a question and have two models respond to it, you're not really going to find this model responding. So what I find fascinating about this is that there's mystery Gemini one, there's the anonymous chat bot which some have speculated is OpenAI's chatbot, and of course we've got the Gemini test, which is of course another speculative Gemini chatbot. So it seems that there might be four, four Gemini tests in the works and they're trying to figure out which model they're going to release. Now, we know that it could be a variety of different things. It could be Gemini 2, it could be 1.5 Pro, the experimental version that potentially being upgraded. It could be a different multimodal system like OpenAI's with the project Astra, or it could be even potentially some of Google's open source efforts. As we've known earlier this year Google have made somewhat of an effort to start focusing on their open source models with the Gemma family of models. So I do think that regarding the fact that there's been all of these different announcements, well, not announcements, but small things that we can see on the Chatbot arena. And last time when there were models that were unknown and they were anonymous in terms of who put them up, the week after that we got GPT4OH mini. And then before that we got GPT4O. So it was rather fascinating to see beforehand what happened. And I'm wondering now if within the next two weeks we're probably going to get an announcement from Google now and even more Google News. This was something that I was supposed to cover yesterday, but this is very fitting for what we're talking. So if you remember the Strawberry account, the one that was talking about OpenAI's secretive Q star model, the one that was supposedly good at reasoning and all of the crazy breakthroughs, they actually tweeted something they said just for me, something that can go off, think over multiple steps, download some reports, compare, analyze and get back to me in a few hours at a similar level, as you'd expect with a high agency PhD. And you can see that Logan Kilpatrick of Google says 2025. So I mean, it seems that across the industry, 2025 is one of the years that we're going to be really getting transformative AI in the sense that it's going to be auton and be able to do multiple different things. Now, this isn't any kind of like a hype tweet where he's stating that 2025, yada yada yada, it's just a date in mind that goes to show that things are about to get a lot crazier. And I do like this date because previously when I do Hear the date 2025, Jimmy Apples has had that date in his bio for quite some time. Jimmy Apples referring to an infamous OpenAI leaker that has consistently given details regarding OpenAI's newest models and Google's newest announcement. Now, Google are non stop with the updates. They need to remind you who is the boss when it comes to AI. And they tweeted, meet our AI powered robot that's ready to play table tennis. Now, I think the research behind this is rather important because what we have here is a agent that is able to perform table tennis at a level that is at the amateur level. And some of the research that we've seen regarding robotics is always important because robotics is long known as the hardware issue, the thing that, you know, most people do struggle with in terms of across different industries because you're dealing with the physical world. It's not like software where you can fix bugs. In the physical world, there is a much longer feedback loop in order to iterate on your errors, thus leading to slower overall development, considering the fact that these things are not cheap by any means necessary. And of course, these things are very data intensive, essentially meaning that these require a lot of data to train. Now, one of the things that I did like about this was the approach that Google used to train this robot. They said that to train the robot, we gathered a data set of initial table tennis ball states, which included information about position, speed and spin. The system practiced using this library and learned different skills like forehand topspin, backhand targeting and return serves. And then of course, this is the juicy bit it says our robot first trains in a simulated environment, which can then model the physics of the table tennis matches accurately. And once deployed to the real world, it collects data on its performance against humans to refine its skills back in simulation, creating a continuous feedback loop. So this is one of the cases where we have real advancement in terms of the fact that we've got an AI system that's being able to use these simulated environments to sort of speed run their learning capabilities. And I think in the future, simulated environments and simulated data is going to be one of the biggest ways that we achieve robotics breakthroughs. Because there just isn't and might not ever be enough physical data to kind of encapsulate every possible scenario that could happen in the world. And simulated environments could capture billions of different scenarios the AI could then be trained on and it could pretty much be prepared for any novel case which would make these AIs incredible at doing a variety of different tasks. And what was really cool here, they said that we designed this system to adapt to various opponents by tracking their behaviors and playing style to such as which side of the table they tend to return the ball to. And this allows it to try different skills, monitor its success rate and adjust its strategy on the fly. And it went up against 29 unseen humans across four different skill levels during our research, from beginner to advanced. Overall, the robot scored in the middle of participants, implying that the robot can operate like an intermediate amateur. Now of course they talk about here that was it able to be an advanced player? In short, no. There are physical as well as skill limitations including reaction speed, camera sensing abilities, spin handling and the paddle rubber, which are hard to accurately model in simulation. Which is why once again, once we do get these simulation environments a lot better, we're going to be able to see a drastic improvement in robots because we're going to be able to collect the data more effectively. This is something that was really, really fascinating. And it just goes to show that whilst yes, Google might not currently have the best chatbot, although on the arena they currently do have, they're still doing a of different AI research that is really important and they're combining that stuff with robotics. Now a former Googler has said that there is no long term in our world anymore because in five years the world will be unrecognizable term in our world anymore.

So you have to understand what most people don't understand is the speed of the cycle. Right? So, so you know, any company today that's doing a five years plan have no idea what they're right. Because our world in five years is unrecognizable as compared to our world today. Take geopolitical, economic, you know, technological, whether that's artificial intelligence or you know, synthetic biology, you know, global climate change. There is so much changing, right. And you'd be really, I mean you definitely need a five year plan for your cash cow, for your core business. Yeah, but if you want a five year plan for innovation, good luck.

So basically this is Mo Gao Dat, which was Google's previous chief Business AI officer and essentially he's talking here about the fact that due to the nature of AI, the iteration cycles are really fast. And due to the nature of the fact that AI is so unpredictable in terms of how quick it develops, the future is becoming more and more unpredictable. It is the unknown unknowns and the effects of the effects that are very hard to predict. Which is why planning for the long term term is remarkably difficult in today's day and age. This is one of the reasons I actually created my post AGI economics community. Because one of the things you know, and as he said, most people don't know this, is that the future is rather uncertain and it's very hard to prepare for. I'm doing my personal best to prepare for the future with regards to my life, my business, and just helping those who want to do the same. Because there are so many changes happening on a day to day basis that, that most people don't even pay attention. And this is why I've created a few databases in my post AGI economics community to pay attention to literally every single job that is being impacted. And this is a database that you can access in my community. And I've also got another database that is showing people every single available new opportunity, business opportunity that is being created thanks to AI. And that one is also gathering steam as many different tools manage to come out. There are so many different industries to make money. Of course I'm going to use this opportunity to shamelessly plug my post post AGI preparedness community. You will get instant access. But the point is, is that most people aren't preparing and this is something that is quite hard to do. But nonetheless, it's always important to make some kind of efforts if you want to stand a chance at thriving in the future. Now, interestingly enough, Reid Hoffman actually says that the OpenAI board that fired Sam Altman was incompetent and that there's no universe in which this was a competent action and that none of you knew what you were doing.

Just from the outside, no inside information. There were a lot of very incompetent things drawn. I will start with a simple one because I just report from my own experience. I was at an event called the Grove in Napa Thursday night. Mary Marat was there at dinner with us.

Oh my God.

Okay. And then she gets up from dinner and heads, off we go, something's happening. And at breakfast we don't see her. Right. And so I'm driving away and my chief of staff shows me the blog post, which is fired Sam. And Mira Moradi is our interim CEO. Now, just from this, what I knew was they hadn't called her until dinner time the day before. There is no universe in which this is a competent action. Right? I mean, it literally would have had to been. Sam murdered somebody yesterday and is in jail for murder. Okay, you're it. You know, get here right now. Like, that kind of levels up. Like if it was a. Well, like as far as we can tell from all this stuff is, well, maybe in his communications with certain board members, they had, they felt they weren't getting the full truth, which maybe is the final story for that. That is like, that is so far away from the way that you operate there for trying to be good to the OpenAI mission, beneficial AI for, for humanity, for the organization. I mean, it just like, it's just. That's incompetent. Like, it's literally, it's like, no, like that. That is.

So that was a fascinating insight to some of the insights that were going on at OpenAI. And it kind of shows us that the kind of decision that was made by the board seemed to have been a very quick decision. And I mean, I'm not going to gloss over the entire thing, but I mean, the most surprising thing about this entire situation, for me at least, was the fact that we can have people that are in arguably one of the most important companies of the decade making decisions so quickly and so potentially irrationally that we have decisions that, you know, seemingly are made within seconds about the CEO of these important companies. I mean, it's just incredible that, you know, this kind of uncoordinated, you know, however you want to call it disorganized manner is, you know, even going on at such high level companies. Which just goes to show you that sometimes, you know, people are human and, you know, sometimes they make mistakes and things do slip through the crack sometimes, which is, you know, just remarkable to say the least. And in more open AI drama, I was supposed to cover this, but one of the things that we recently found out was that Elon Musk has recently filed a lawsuit again against Sam Altman. Recently, when Elon Musk dropped the lawsuit, many had thought that, you know, the case is closed, it's over. Sam Altman is pretty much off the hook and Elon Musk is focused on other ventures because I genuinely thought that Elon Musk would have been focused on his five Huge, giant, billion dollar companies. But it seems that he's really trying to pursue this lawsuit as he feels that he's been wronged by Sam Altman. And he says the case against Sam Altman is a textbook tale of altruism versus greed and that this lawsuit is actually a much stronger one. And they're saying that he intentionally courted and deceived Musk. Preying on Musk, preying on Musk's humanitarian concern about the existential dangers posed by AI. And Altman and his longtime associate Brockman manipulated Musk into co founding vesperious nonprofit. And you can see here that it says after Musk lent his name to the venture, invested significant time, tens of million dollars in seed capital, and recruited top AI scientists for OpenAI, Musk and the nonprofit's namesake were betrayed by Altman and his accomplices. And the deceit are of Shakespearean proportions, which is a funny way to put things. So you can see here that it also states that once OpenAI's technology approached transformative AGI, Altman flipped the narrative and proceeded to cash in in partnership with Microsoft. And Altman established an opaque Web for profit OpenAI affiliates engaged in rampant self dealing. Basically, this lawsuit actually goes on to actually talk about how the SEC have, you know, put some scrutiny and FTC, you know, have complaints about OpenAI and their whistleblowing agreement. So this lawsuit isn't good for OpenAI. I mean, it's going to be interesting to see how they navigate this kind of area because this is certainly a new area, but this entire lawsuit, I'm wondering how this is going to play out in court. We also had Mark Zuckerberg actually explain his reason for open sourcing the llama family of models. And his definition definitely surprised a lot of people.

The llama thing that we're doing, I mean, like I obviously have and I believe a lot in open source. I think it's good for the world more broadly. We're not doing this because we're like altruistic, right? I mean, we're doing it because we want to build a platform that we know that we can rely on having LLAMA as a thing. And the reality is this is an ecosystem and it's not a singular piece of software that we could just build and deploy ourselves. It gets better when you have all the silicon providers optimizing all of their stacks for the thing that we're doing, and when you have all these other companies or startups or different folks who are building different distillation tools or inference tools to make it go faster, more efficient and all that stuff. And you know, all these people are building stuff on it. From my perspective, I just want everyone to be used because I think that the, the more people who are using it, the more the flywheel is going to spin for making Llama better. And I mean, this is maybe like a very selfish and parochial answer, but then that makes it so I can build the things I want to build better. And so, but, but honestly, I think like, people should take comfort in that answer because I think one of the big questions that people have about our open source strategy is like, why are you doing this? Like, you're building this. Like you, you kind of train this model and then you just give it away. Like, is that sustainable? I mean, when, like when Llama four, Llama five takes many billions of dollars to train, are you just going to give it away? And it's like, yeah, yeah. I don't view it as giving it away. I view it as like, you guys all making it better for me.

So that's a rather interesting perspective on how Mark Zuckerberg has entered the AI space. And I think what they're doing is rather commendable considering how open source development allows many different institutions access to frontier AI models. And actually to understand how these systems really do work, of course, you could always call it open weights. And of course this wouldn't be a video without Elon Musk stating that the Grok 2 beta release is going to be soon. I do have, I wouldn't say a bone to pick with Elon Musk, but I would like him to actually just release Grok outside of the X platform. Because I got to be honest, using Grok right here, this just isn't always great. I mean, fun mode and regular mode. I'm not stating that the model is awful, but the model had been out for so long and I didn't know when I actually got this model released. Of course there are geographic restrictions, but it's just the fact that as someone who wants to use the model, sometimes it can be really frustrating when things are entirely delayed. And even when you buy a membership for Twitter, I wasn't able to access the model immediately as was advertised. So hopefully if the Grok 2 beta was actually here, hopefully we do know which countries get access and which accounts do get access. Because at first, if you do remember, Grok was only released to people that had subscribed because to Twitter premium first.