today we have news about new robots Apple intelligence Nvidia grock text to video Tesla and so much more let's get into the first story figure the company behind the figure robot has announced figure 2 and it took basically everything they learned about building robots and put it in a much sleeker much more efficient package figure 02 coming 86 which actually should be the day that we're posting this video and I'm going to do a follow-up all about the release but first let's take a look at their demo video and so what you're going to notice is they really cleaned up the [Music] look articulating arms legs everything built into an absolutely gorgeous package [Music] now what you're about to see next is what most robot companies say is probably the hardest part of engineering a robot which is the hands now check this out I think it was Elon Musk that just said that the majority of their engineering effort went into simply building the hands for the Optimus robot and there it is so a quick preview

more information coming out today as we publish this video, but it looks like an absolutely gorgeous robot, and I cannot wait to see more about it. So, Brett Adcock, the CEO and founder, says this has been engineered for over a year. Figure O2 is the most advanced humanoid robot on the planet, so very excited to see this. We are well on the way to seeing $220,000 home robots in every household. Okay, next, just announced AI chip startup Grock (GRQ) raised $640 million Series D led by Black Rock at a $2.8 billion valuation, and that is up from a billion-dollar valuation in 2021. They add an Intel executive as COO, which you know is kind of interesting, seeing how Intel is absolutely cratering right now, but that's fine. Here's the article out of Reuters. Now a couple things about Grock. I absolutely love Grock. Their inference speeds are hyper-fast, much faster than anything else you're going to see on the market, and that's what sets them apart. They build their own

unique chips. Now, a few months ago, they actually stopped selling their chips directly, and now they are offering inference as a service only. They take the top open source models, they host them, and they offer inference at speeds that are absolutely mind-blowing. So right here, you can see they have Gemma 2, Llama 3.1, 70b Mixol, all the latest cutting-edge models. Now, a lot of people think Gro is a direct competitor to Nvidia, and maybe that's going to be the case in the future, but Gro doesn't actually sell chips anymore. They are only selling inference. Grock will use the funding to scale the capacity of its tokens as a service offering and add new models and features to the Grock Cloud. Grock will deploy more than 108,000 LPUs manufactured by Global Foundries by the end of the first quarter of 2025. That is right around the corner. Now, here's the most interesting bit in my mind. Meta's Chief AI scientist, Yan Laon, was named its newest technical

advisor, I think that is a fantastic idea, and well done on getting Yan to join Grock. Now, here's why Yan Laon runs Meta AI, and Meta AI produces the best open-source models. It seems like Meta and Grock are forming a closer and closer relationship as time goes on, and it makes sense because Grock offers open-source models, Meta creates the best open-source models, so of course, this is a perfect synergy for them both. Now, as I mentioned, they kind of already have a pretty close relationship because Grock launched the new 3.1 Llama models on day one, so they were already talking to Llama well before the launch day. They had inference ready to go on their cloud service, so very cool. Congrats to Grock. Okay, next, as reported by The Wall Street Journal, OpenAAI actually has a cheat detection tool for students, and really for the schools and teachers, to determine if somebody is using chat GPT to write their essays or anything else that they're doing. And what it says here

is technology that can detect text written by artificial intelligence with 99.9% certainty, has been debated internally for 2 years. Now, here's the thing: I know that there are a ton of other AI cheating tools out there, and basically none of them really work. And if you get a false positive, it is so detrimental to that student, and even one false positive is too much. Now, how could OpenAI actually tell text is created by AI? Because there are so many prompts, so many variations, so many personalities and roles that you can give large language models to create text in certain ways, but there's something about them that can be detected with such accuracy. It's kind of crazy to think about. I'm not so sure. So, as the article says, OpenAI has a method to reliably detect when someone uses ChatGPT to write an essay or research paper. The project has been mired in internal debate at OpenAI for roughly 2 years and has been ready to be released for about a year, according to people.

familiar with the matter and internal documents viewed by The Wall Street Journal. An OpenAI spokeswoman said the company is concerned. The tool could disproportionately affect groups such as non-native English speakers. The text watermarking method we're developing is technically promising but has important risk. We're weighing while we research Alternatives. Now it's hard for me to imagine how this would actually work but I believe if they're saying they have it then they probably do have it. And what are they doing just dropping certain words in certain places and there has to be ways to just change your prompt to get a different output right. I don't know let me know what you think. So here it describes it as the anti-cheating tool under discussion at OpenAI would slightly change how the tokens are selected. Those changes would leave a pattern called a watermark. It would be unnoticeable to the human eye but could be found with OpenAI's detection technology. So they're still debating whether to release it or not. We'll see.

okay, next. The team from Character AI is seemingly joining Google. Now, Character AI is extremely popular, so this is kind of a surprising move. If you're not familiar with Character AI, it is very, very popular with younger generations, being able to chat with essentially an AI friend or AI role-playing. And it seems like Google acquired them in a very similar way to how Microsoft acquired Inflection AI. They didn't actually acquire the company itself. They essentially gutted the company, moved all the assets over to Microsoft, and then brought all of the talent over there as well. And that's probably their sly way of avoiding FTC looking closely at them as a monopolistic move. So over the last two years, however, the landscape has shifted. Many more pre-trained models are now available. Given these changes, we see an advantage in making greater use of third-party LLMs alongside our own. This allows us to devote even more resources to post-training and creating new product experiences for our growing user base. We're excited to announce that we've entered into an agreement with

Google that will allow us to accelerate our progress. As part of the agreement, Character AI will provide Google with a non-exclusive license for its current LLM technology. This agreement will provide increased funding for Character AI to continue growing and focus on building personalized AI products for users around the world. And Gnome Daniel and certain members of our research team will be joining Google. So there we go, another semi-acquisition, semi-Aqua hire. Kind of difficult to really suss out exactly what they're doing, but Character AI is growing. And essentially what they saw is that large language models, as I've been saying for months and months now, are commoditized. And it's easy to build a model that can do characters and roleplay that doesn't really require Character AI's custom, unique, proprietary model. And so instead of focusing on that aspect of their business, they're going to be focusing on the user experience, the roleplay itself, rather than building their own models. So that's it for Character AI. I've actually never used Character AI. Let me know if you have and what you think of it. All right. And in

another story where an AI company has used data that may not have been theirs. AI video generator Runway trained on thousands of YouTube videos without permission. Now most of you seemingly think if a video is on YouTube, then anybody should be able to use it any way they want, and I don't necessarily agree with that. Maybe for the subtitles, yes, but for the videos themselves. When I publish a video, I publish it to YouTube, and that doesn't mean I'm giving permission to anybody to rip it and use it however they want. Here's an example: about a year ago, I published a video to YouTube. Somebody took it, posted it to TikTok, and it got millions of views, and they didn't credit me at all. I didn't get any monetization from it, and it was pretty frustrating, to say the least. So I can imagine if you're MKBHD, who commented on this article, you would be pretty frustrated as well. You put so much time, so much effort, so much investment into producing these videos, and then another company basically takes

them and creates another product from them. Now, the flip side to that AR, which I understand as well, is that, well, people can watch these videos for free online, learn from them, and then apply that learning any way they want. However, we're not talking about people watching these videos, we're talking about artificial intelligence learning and training based on those videos, so there's going to be a lot of legal drama now and for the foreseeable future as we figure out where the law lies with AI training on potentially copyrighted material. All right, next AMA launches version 0.3 now with tool support. So, a couple example tools include functions and APIs, web browsing, code interpreter, and so much more. So here's an example of Llama 3.1 being given a calculator and it being able to call that calculator. So enter the message, What is 192 + 12 to the power of 4? and it sends it off to the correct tool to get the answer. So, tool use is a huge component of reaching AGI in many people's opinion. So it's really

exciting to see tool use, and I'm especially excited to see agentic frameworks like crew aai be able to leverage tool use like this. So congrats to AMA for their new version launch. And here are four model examples that support tool usage: llama 3.1, mistl Nemo, which is a tiny model, fire function V2, which I've not heard of, and command r+. And you can actually search for specific models that do support tool calling. All right, next, apple intelligence is not coming when we thought it was. I am absolutely sick and tired of companies announcing features, demoing products, and then completely either delaying it indefinitely or not releasing it at all. I think we should make a rule: companies should not be able to publish a demo of a product or feature greater than 3 months out from its actual launch. Now here's the other thing: unless you have the most current version of the Apple iPhone, you can't even use these features. So I have a two-generation-old iPhone, and I don't even

get to beta test Apple intelligence. And by the way, if you want to beta test Apple intelligence, you actually can right now, as long as you have an iPhone 15 Pro or greater, I think, which, you know, kind of frustrating to think that there are tens of millions of phones out there that simply won't get Apple intelligence. But that being said, developers are starting to test AI features, but as part of their launch, they are not going to make Apple intelligence part of it. Next, back to Runway. Runway is releasing gen 3 Alpha image to video. This update allows you to use any image as the first frame of your video generation, either on its own or with a text prompt for additional guidance. So you simply give it an image, and then it automatically converts it into a video. And look how good that looks. Here's another example. Here's another one. So all of these started as pictures simply loaded into Runway, and then you get video from it. Very, very cool. All right, next, one of the first videos I ever made. I think about the third video.

I ever made was about leonardo da and I absolutely love it because I have no design skills whatsoever just acquired leonardo da gives an update on Project Groot and I've made a number of videos about Dr. jinfan and his team project Groot and so let's read what he says. We discover a systematic way to scale up robot data tackling the most painful pain point in robotics. The idea is simple: human collects demonstration on a real robot and we multiply that data a thousand times or more in a simulation. Let's break it down. We use Apple Vision Pro to give the human operator first person control of the humanoid. That's awesome. Parses the human hand poses and retargets the motion to the robot hand all in real time. This is all called teleoperation. We use RooCasa, a generative simulation framework, to multiply the demonstration data by varying the visual appearance and layout of the environment. So basically, they take one version, the human version, and then they create a thousand more variations of it to help train the

models. Then, they apply Mimic-Gen, a technique to multiply the above data even more by varying the motion of the robot. So, to sum up, given one human trajectory with VisionPro, RUC, CAA produces n varying visuals. Mimic-Gen further augments to n times n varying motions. So, basically, a lot of different variations. And all you need is one initial data set. Next, Sam Altman posts on Twitter about their safety and alignment teams and then gets community noted. So, let's read a few quick updates about safety at OpenAI. As we said last July, we're committed to allocating at least 20% of the computing resources to safety efforts across the entire company. Now, that is very interesting because that is literally why, as we're about to read, Ilya Sutskever and Jan Leike basically left because they were deprioritizing safety and super alignment, but now he's restating that he's going to do it. Our team has been working with the US AI Safety Institute on an agreement where we would provide Early Access to our

next Foundation model so that we can work together to push the science of AI evaluations. Excited for this finally. We want current and former employees to be able to raise concerns and feel comfortable doing so. This is crucial for any company but especially for us and an important part of our safety plan. In May, we voided non-disparagement terms for current and former employees and provisions that gave Open AI the right, although it was never used, to cancel vested equity. We've worked hard to make a right. I've already reported on this drama. Sam said he didn't know about it. I think he did. Obviously, he's the number one guy at Open AI, so he should have known about it. Even if he didn't, then community noted. We're starting a new team co-led by Ilia Suk and Jan Ley, both of which who left, and dedicating 20% of the compute we've secured to date to this effort. So what they are clarifying is the announcement post. The original one is very clear that the compute was

meant for the super alignment team only. Whereas here he says safety efforts across the entire company. So definitely not what he said next stability AI makes a nice release stable Fast 3D stability ai's latest breakthrough and 3D asset generation technology. It transforms a single input image into a detailed 3D asset in Just5 seconds setting a new standard for Speed and quality in the field of 3D construction. Now if you do not think that video games are going to be completely AI generated in the very near future maybe this should convince you otherwise. So here are some use cases that they describe here fast inference time during pre-production where experimentation is key static assets for games of course 3D models for e-commerce fast model creation for arvr. And here are a bunch of examples of assets generated very very very nice all right next meta Court celebs like Aquafina to voice AI assistant ahead of meta connect. And this is awesome. I want Snoop Dogg as my AI assistant voice and there's

nothing you can do to convince me otherwise. So, Judy Dench, Keegan Michael Key, Aquafina, are among multiple actors and influencers whose voices become part of Meta AI's offering. This strategy is going to increase in popularity even more as time progresses. They're going to get more and more celebrities to lend their voice to AI, and that is a big commitment from the celebrity because once they lend their voice, that's it. Their voice can be used in any way they want for that product. So, I have the Ray-Band Meta AI glasses, which I absolutely love. And you could say, Hey, Meta to it, and then you get an AI assistant in your ear. So just imagine a celebrity being able to talk to you. And yes, Snoop Dogg. So that's all the news for today. Let me know what you think in the comments. If you enjoyed this video, please consider giving a like and subscribe. And I'll see you in the next one.

NEW Figure 02 Humanoid Robot, Groq Raises Massive Round, Runway Caught Scraping Videos

Artificial Intelligence (AI), Open Source, Generative Art, AI Art, Futurism, ChatGPT, Large Language Models (LLM), Machine Learning, Technology, Coding, Tutorials, AI News, and more

** Exclusive Pinecone Partner **

NEW Figure 02 Humanoid Robot, Groq Raises Massive Round, Runway Caught Scraping Videos

스크립트