Leveraging AI

111 | OpenAI could be on the brink of bankruptcy, Nvidia chips delay impacting the entire AI eco-system, Google acquirs Character.AI and more AI news

August 03, 2024 • Isar Meitis • Season 1 • Episode 111

Share episode

Are the giants of AI facing financial challenges?

This weeks AI news episode of Leveraging AI, we dive deep into the financial and strategic challenges confronting OpenAI, one of the leading players in the AI industry. As OpenAI grapples with enormous expenditures and the looming threat of running out of cash, what does this mean for the future of AI?

Let's talk the latest news and developments in the AI world.  

In this episode, you'll discover:

  • Why OpenAI is at risk of losing billions and what it means for the AI industry.
  • The staggering costs of developing frontier AI models and their impact on the market.
  • New safety protocols and regulatory frameworks shaping the future of AI deployment.
  • The latest advancements and delays in AI technology from industry leaders like NVIDIA and Apple.
  • Insights into recent M&A activities, including Canva's acquisition of Leonardo.ai and Google's buyout of Character.ai.

Leopold Aschenbrenner - 2027 AGI, China/US Super-Intelligence Race, & The Return of History - https://www.youtube.com/watch?v=zdbVtZIn9IM

AI BUSINESS TRANSFORMATION COURSE: 

About Leveraging AI

If you’ve enjoyed or benefited from some of the insights of this episode, leave us a five-star review on your favorite podcast platform, and let us know what you learned, found helpful, or liked most about this show!

Hello and welcome to a weekend news edition of the Leveraging AI podcast, the podcast that shares practical, ethical ways to improve efficiency, grow your business and advance your career. This is Isar Meitis, your host. And every week, it's jam packed with news. If last week was the week of releases of new models. This week has actually a very wide variety of topics of news. Interesting from M& A topics to valuation topics to interesting delay topics in the deployment and interesting delay topics in both NVIDIA chips, as well as Apple's AI deployment. So basically every week we have a lot to talk about, so let's get started. The first piece of news comes from OpenAI and it's financial challenges. What that means is A report on the information is stating that OpenAI is expected to lose five billion dollars by the end of 2024. Puts them at bankruptcy risk and they might run out of cash in the next 12 months, unless they raise additional money. They state based on their sources that OpenAI currently spends about 700, 000 daily just to keep ChatGPT running. Basically all of us and everybody around the world using it. They're projecting to invest about 7 billion in training of their next model GPT 5 or anything in between like they've been deploying so far like GPT 4. 0 mini and so on. They are allocating about 1. 5 billion for staffing. So in the current extreme competition on talent, I'm sure they're paying some interesting salaries to interesting people and they're income is expected to only be 3. 5 to 4. 5 billion dollars coming from about 2 billion dollars from ChatGPT subscriptions and about 1 to 1. 5 billion dollars from people accessing the API. Now the company raised about 11 billion dollars so far from 7 funding rounds. Most of the money comes from Microsoft, either in means of straight cash or in means of access to its servers to train and run ChatGPT and so on. Now, obviously, this business model is not sustainable. You cannot make three to 4 billion a year. And that's, by the way, extremely impressive. I don't think any company in history that's to make three to 4 billion a year so quickly from the launch of his first product, ChatGPT at the end of November of 2022. And In the beginning, it was completely free so that they didn't even start making money at that point. So their growth in revenue is incredible, but their growth in expenses is even more incredible. Now, what does that mean? It means a few things. First of all, they're a very different company than some of their competitors. So Google, Meta, Amazon, and even Microsoft has billions to spend on training and growing new models and the infrastructure that is required in order to run it. Open AI does not have the revenue to do that, at least not yet. So that puts them at a position where they continuously have to raise significant amounts of money. That being said, there's a pretty clear consensus, both on the side of the tech companies, as well as the investment side of the world of the potential incredible impact of AI, which tells me that most likely OpenAI will raise a stupid amount of money in its neck rounds, an amount we've never ever seen before, probably in the tens of billions of dollars to keep OpenAI running and keep them at the lead, or at least in competition with the other companies. But what it shows us, it also shows us the stupidly high costs that is associated with developing frontier models, which tells us all that there is going to be a very short list of companies that will be able to continue to do this moving forward. And we're going to talk about the consolidation in this market in some other news today. There's a lot of other news about OpenAI today. One of them is they just introduced a new GPT 4. 0 model. So GPT 4. 0 was introduced not long ago, then they introduced GPT 4. 0 mini. It was actually doing very well in comparisons to other models. On some aspects, it's actually doing better than GPT 4. 0. So the mini version, which is significantly smaller and significantly cheaper to use, is ranking very closely on the LMSIS leaderboard To the GPT 4. 0, even though it's about 16 times cheaper to actually use. So I think we're going to talk more about this, but that's the direction that everything is going. We're going to get cheaper, faster, and better models all the time. But going back to the new GPT 4. 0 model, the new experimental GPT 4. 0 model has a long output model. What does that mean? So models are being measured by their context window, What is the length of information you can put in a single chat? So you run out of memory basically at a certain point, and that is usually measured by tokens and tokens it's just a unit that all these large language models work at, but every token is about 0. 7 words. Meaning if you have 128, 000 tokens context window, it means you can fit about a hundred thousand words, Both the stuff that you write in your prompts and the information you provide the chat, as well as what chat is spitting out. So the current limit of ChatGPT 4. 0 is 128, 000 tokens. And that is not changing in this new experimental model. So what is changing is the output. length of a single response. so far, GPT's 4. 0 limit for a single response, it could give you to whatever prompt you give it was 4, 000 tokens, which is about 3, 000 words. This new experimental model extends the output capacity in a single response to 64, 000 tokens. Basically, Half the length of the information that can be in a single chat can come out in a single answer. That's a 16x improvement from the current model that allows to obviously ask a lot more complex and sophisticated questions and get a lot more detailed information. This is extremely useful in data analysis and detailed business use cases, as well as research and so on. So a huge improvement in that particular case. And this model comes cheaper, as I mentioned, like everything new that we're going to get, then the existing GPT 4 variants, other than GPT 4 mini. So the pricing right now is 6. per 1 million tokens of input. So the prompts that you're putting in an 18 per 1 million output tokens, basically the inference, what the outcome that you're getting from ChatGPT. This is currently an alpha with. Being tested just by a few users, but they're expecting this to be just a few weeks, and then they're going to roll it out to everyone. I would like to interrupt the news for a few seconds and tell you about the new course we recently launched. As you probably know, we have our flagship course called AI business transformation course. We have been teaching it since April of last year and hundreds of companies have already taken the course, both privately by approaching me and building private courses for them, as well as publicly with courses you can sign up for, but there's been a huge demand and a lot of people approached us to create an offline self paced version of that course. And we have, so that course is currently available on our website. Multiple people has already signed up since we launched it. Just a couple of weeks ago. So if you want to take a course with me personally, as an instructor, with a cohort of people that can learn from one another, Other business leaders who are taking the course, that's one option. And you can find that on our website, or you can else take it as a self paced course, which you can do on your own time and do it at your own pace. as I mentioned, it's available right now, the link is going to be in the show notes, so you can open your phone or computer right now, wherever it is you're listening at and find the link and click it and get all the information you need. I would love to see you there and hear your feedback about the course if you're taking it. That's it. Let's go back to the news. Now, in the past few months, we had a lot of negative news about open AI and safety issues between multiple significant figures departing the company or changing positions. And so this week there are news about multiple steps that open AI is taking in order to potentially at least change our view off their approach to safety, but one of them is that open AI Sam Altman just announced that the company is working on an agreement to provide early access to each major new models to the U. S. A. I. Safety Institute for testing before they release it. They have announced a similar deal with the UK's AI safety body just not too long ago, and this comes again in the wake of a lot of criticism of OpenAI's approach to safety. This is obviously very good news, and I wish that there's going to be some international group that's going to monitor this. I'm going to talk more about the international approach to this and potential implications in the next, few news pieces. But right now, at least. At least submitting models to a third independent party to evaluate it for safety before release is a very good step forward. And I hope that every company before it releases its models will do the same thing. You're already talking about safety and regulations, then the EU AI Act is now finally in force. So the official effective date of the EU AI Act was August 1st, so just earlier this week. And, And just as a quick reminder, because we talked about this in multiple previous episodes, it takes a real, risk based approach to how to evaluate AI risks, low or no risk, which is most applications. And then it's not regulated high risk, which include biometrics and medical and education and employment implications and applications that requires compliance with high risk quality management obligations. And then there's limited risks like chatbots and deep fake tools and so on that must meet different transparency requirements. The penalties for not meeting that are very severe. So it could be up to 7 percent of global annual turnover for doing releases that break these rules. What does that mean? first of all, it doesn't mean a lot immediately because most of the provisions will only be fully applicable in mid of 2026. So companies have time to do this, but there's already multiple companies who said they're not going to deploy their models in the EU for now. Which is good from one perspective, because it means it is keeping the safety off the EU's people from getting access to potentially harmful AI tools. The flip side of that is it's obviously dramatically slowing innovation for companies in the EU who want to harness the capabilities of AI tools. And so it's a double edged sword where on one hand, yes, it's keeping you safe, but on the other hand, it's dramatically strong, your ability to compete in the world. And I think the EU will have to make some adjustments to its current set of rules or come to some agreements with the companies with more specific details on what's allowed and not allowed so it will enable them to at least deploy some of their capabilities for the EU. EU members. In another move to show its commitment to safety, OpenAI announced a new development method to increase safety of models that they call Rule Based Rewards or RBR. They've made that announcement on July 24th. And the key components is that it uses clear step by step rules to evaluate the model's output for safety standard. And it's designed to balance between the potential Harmfulness and harm prevention of the model. So going back to what I said about the EU, it's trying to find a balance between the good and the bad that can come from the model. They've already implemented that into GPT 4 and GPT 4 O mini. And they're obviously planning to use this methodology in future models as well. And I'm gonna quote from their announcement to tell you what they shared specifically about this. And I'm quoting now, to ensure the AI systems behave safely and align with human values, we define desired behaviors and collect human feedback to train a reward model. This model guides the AI by signaling desirable actions. However, collecting this human feedback for routine and repetitive tasks is often insufficient. Additionally, in our safety policies change, the feedback we've already collected might become outdated, requiring new data. So that's them describing the problem. So again, the problem is that it's just not good enough and probably not scalable with the amount of data that they're putting in to use only human feedback for these models. So now the solution that they're suggesting, thus, we introduce Rule based rewards, or RBRs, as a key component of OpenAI's safety stack to align model behavior with desired safe behavior. Unlike human feedback, RBRs uses clear, simple, and step by step rules to evaluate if a model's output meet safety standards. But this continues on and on with more details, but then they are stating that there's also obviously potential drawbacks and now I'm quoting again, shifting safety checks from human to AI can reduce human oversight of AI safety and might amplify potential biases in the models if biased models are used to provide RBR rewards. In other words, if they don't get their equations right and their specific rule based that will continue to go really bad because the models will continue to evolve without human supervision. So what does this tell us? It tells us that probably There needs to be some, again, external tools to evaluate the safety of these models. Hopefully these tools will come from an international alliance that will do this kind of like the nuclear weapon oversight that is done by the global community to monitor the development of nuclear weapons in other places. I really hope that's the direction it's going to go. Even though it's not clear whether that's actually happening or not, there's been a few attempts to do that, but so far, there's nothing practical out there that actually moves in that direction specifically about open AI. I don't know if that is done only because the amount of pressure that they have on them right now, both from Congress and from media, as far as reducing Their investment in safety. I don't know if that's actually a good thing or a bad thing, what they're doing right now. All I can say is that we'll keep on watching and I will update you as this thing progresses. And since we're already talking about AI safety and we're talking about potential global aspects of this, Sam Altman this week had a very interesting announcement. He's stating that the urgent question over time is who will control AI dominance in the world between the democratic side of the world and authoritarian models controlled by countries like Russia and China and so on. He's saying that U. S. is currently in the lead of the development, but that lead is not guaranteed. Other governments may overtake the democratic nations in its AI capabilities. As an example, Vladimir Putin, the president of Russia, stated that the AI race winner will become the ruler of the world. China has set up clear guidelines and investments in order to have the lead in the AI race by 2030, which is not that far out. In this article, Altman warns around The potential negative consequences if these authoritarian regimes get the lead in AI, they could force companies to share data, they could develop new types of weapons and cyber weapons that nobody has ever seen before ability to fight, and he goes on and gives some other examples, but he also states what he suggests or proposes that a US led global coalition will do in order to prevent that and make sure that what he's claiming is democratic approach to AI prevails. So four things that I mentioned, one is develop robust security measures to maintain the coalition's AI lead, basically preventing these other parties from getting access to the models and AI technology that is developed in the West, build significantly more AI infrastructure, meaning data center, power plants, and so on in order to support the growth. And we talked about they are not in a great position From a cash perspective and basically what he's saying that the government needs to step in, in order to provide infrastructure to make sure That the West stays ahead on this process. I tend to agree the numbers it's starting to come to you when it comes to tens of billions and hundreds of billions of dollars to train and run the future models is not something a company can do on its own. And the only way to do this is either through a large government body, that's going to finance that. or a international coalition that's going to finance that to keep the infrastructure in the West. He also adds that there needs to be a commercial diplomacy policy, meaning export controls on what AI capabilities can be shared outside of the U. S. or the democratic alliance. And putting limitations or some kind of oversight to foreign investments in Western companies that are developing AI. And the last component is established global norms for AI development and deployment, which ties back to what we talked about earlier, just having more control on what everybody's doing and deploying and so on. Now, a couple of months ago, when Leopold Aschenbrenner left OpenAI and did his open interview talking about the current status in OpenAI, he talks about this topic a lot. So if you didn't listen to that interview, go and look it up. I'll put the link in the show notes. It's an incredible interview that talks a lot about the global status of OpenAI. Risks and powers trying to get access to advanced AI and what might be the implications. So it's not a new thing, but the fact that Sam Altman took the time, and he's a pretty busy guy, I'm sure to outline this in an article tells you that he has serious concern. It also shows you that there's a deeper and deeper involvement of politics in this power game of AI control. And I think part of Sam Altman's move is to potentially relieve the needs of his own company to raise the amount of money and resources that are required in order to achieve the goals that they're trying to achieve. That being said, I'm sure his concerns are real and that we all need to pay attention to that, especially governments in the West. And from OpenAI to two companies that are tied to and related to OpenAI with some interesting news. One of them is Figure. Figure is a humanoid robot company that is backed by open AI and some other interesting players like NVIDIA and Microsoft and Intel capital and Jeff Bezos, expeditions fund. Lots of really big players invested a lot of money in this company. They just announced their new version of their robots. Their next version is going to call be called figure two, and they're planning the formal reveal on August 6th. If you're listening to this podcast, as it comes out, you can still catch the release. They're claiming it's going to be the most advanced humanoid robot on the planet. Now, that comes after they've announced a partnership with BMW manufacturing that already is testing figure one in some of its manufacturing facilities. This obviously is another step in the race of humanoid robots. And this race is definitely intensifying. And there are several big players in that, but figure is just one of them. And the expectation is that these robots will show up more and more in 2025 and definitely into 26 and 27 in more and more places, initially in public places like factories and hospitals and so on. But over time it will make its way into our homes as well. But the really interesting news that comes that is related to OpenAI from one of its partners actually comes from Microsoft. So Microsoft just listed OpenAI as a competitor in its 10k SEC filing. This is the first time they've been doing it. And that's another step in the weird relationship between OpenAI and Microsoft. So going back to the beginning, for those of you who don't know that Microsoft invested or committed to invest about 13 billion in OpenAI. As part of that process, OpenAI started their for profit arm that is controlled by their non profit board, which is a weird thing to begin with and Microsoft owns 49 percent of that. But despite all of that, they've been competing on multiple things. Microsoft has been developing its own models to potentially in the future, not be dependent on OpenAI. OpenAI in parallel are developing things that are competing directly with Microsoft tools. And one of the immediate recent examples is that, as I mentioned last week, just announced GPT search, which directly competes with Bing. So in the taken filing, OpenAI is listed alongside Anthropic, Amazon and Meta as competitors in the AI development. And they're also mentioned as a competitor next to Google as far as search competition. Now, why are they doing that right now? maybe because of the recent news from OpenAI as far as GPT search, but also maybe because they are under investigation for antitrust by the government because of their big stake in OpenAI. As part of that process, they also gave up their observer board seat that they had in OpenAI because of the same concerns. So do I think it changes anything in the actual relationships between OpenAI and Microsoft? No, I don't think so. I think Microsoft is continuously hedging its dependency on OpenAI. That doesn't mean they're not going to invest additional X number of billions or maybe tens of in OpenAI to keep the relationship going for now, but as I mentioned, there are other forces such as government oversight that is impacting these decisions. And we've had enough, I think, from OpenAI for one day, so let's move to some M& A news. Two very interesting M& A's happened this week. One, Canva acquired Leonardo. ai. Those of you who don't know Canva, Canva is a design tool that's been around for a while. It's a highly successful company. I use it all the time. Me and another 180 million monthly users use Canva regularly to create designs for multiple things. They have about a 2 billion revenue and they just acquired Leonardo. Leonardo. ai, which is another tool I really like it's one of the more advanced AI graphic generation tools. The reason I'm saying more advanced is that they took the approach of instead of using it as a regular chat and just using prompts to create new images, they built an entire set of tools that makes it feel that makes it feel more like a professional tool for designers, integrating a lot of AI tools and using a lot of open source libraries that allow you to have a lot of really cool extensions that allows you to do stuff that is impossible to do on any other platform other than open source tools. In that deal all of 120 Leonardo AI employees, including the executive team, will join Canva, but they will continue running dependently as their own platform. So leonardo. ai is not going away, at least for now. So if the question is, why is Canva doing this? they're doing this because Canva has committed already to a lot of AI capabilities. They've been pushing this very hard with their magic studio features, which are really cool and very helpful. If you're using Canva regularly, that allows you to create new images or manipulate or change existing images and so on using AI. And now With the access to Leonardo's technology, they'll be able to provide significantly more AI capabilities within the Canva platform as a Canva and Leonardo user. I'm really excited about this particular one. There are a lot of really amazing features that I currently use in Leonardo. That will be incredible to have within the Canva platform. Now, the second interesting M& A news comes from Google. Google just acquired Character AI. So if you don't know character AI, they're a really unique company. And what they've created is a chat platforms that allows anybody to chat with known characters, whether historical figures or sports stars or actors and so on. Any person can pick a character to talk to and have conversations with these people. They've had conversations with other companies or potential acquisitions like beta and X and so on, but eventually they landed in Google, which is not surprising because both co founders, Noam Shazir and Freitas, the CEO and the president, were both Google employees and left to start Character AI. So now them plus the other employees of Character AI are all going to become Google employees. Character AI is going to continue running as a separate platform, but they're going to use a lot of the knowledge that these people have in developing these kind of tools in the Google AI development. Also, it's been mentioned that Character AI itself will transition from developing It's own AI models like they did so far to use alternatives from open source, such as Llamas 3. 1 to ride behind the scenes in its products. Both these news, as well as stuff we mentioned earlier today is showing you a very clear trend of consolidation in the AI market. It's happening because it's impossible to finance the amount of money that is required to run frontier models and to develop really advanced companies in this field. Which means more and more smaller or mid sized company that are really successful like Character. ai will get picked up by somebody who has the the ecosystem to support them, and the distribution to support them. I will not be surprised if one of those next companies is going to be Perplexity. So those of you who don't know Perplexity, Perplexity is an amazing platform That combines search capabilities like Google search together with chat capabilities like ChatGPT and Claude and so on into a very powerful search and research platform. I use perplexity daily, probably as much as I've been using Google in the past few months, but they are going to face some very fierce competition, both from open AI with their new GPT search, as well as from obviously Google and Bing. And I don't think they have the funds to compete with the companies I just mentioned, meaning it's not going to be surprising if some other company picks them up and rolls them into their own technology. Now, since we're talking about Google, they've released a new experimental model. It's a new version of Gemini 1. 5 Pro dubbed Gemini 1. 5 Pro experimental version 0801 and it's available for testing meaning you can get access to it through the Google AI studio as well as the Gemini API. It's already at the top of the LMCIS chatbot arena leaderboard with a score of 1300 which is the best score on there right now. So those of you who don't know the LMSys chatbot arena, it's an open platform that anybody can get access to where you can run a prompt and you get two results and you get to vote which result is better and fulfills your needs better. So it's a real world testing of a lot of people that are ranking without knowing which models are actually ranking. So it's a fair test. So it's probably the best platform right now to know which models do best in real live world environments. And currently the leaderboard is Gemini 1. 5 pro experimental on the top of the list with a score of 1300 followed by GPT 4. 0. Followed by GPT 4. 0 mini with a very small spread between them. So 1286 and 1280, and I mentioned that earlier, GPT 4. 0 mini is significantly cheaper than GPT 4. 0 and faster. And it's getting very similar results. And after that, there is cloud 3. 5 sonnet again, another smaller and cheaper version of a bigger model. After that, Gemini advanced and followed by Lama 3. 1, the large model that I talked about last week, the 405 billion parameters model. So all these models are at the top of the list. So you know what's going on right now. And in addition to Ranking the highest on the leadership road right now. It also has a context window of up to 2 million tokens. That's about 1. 5, 1. 6 million words. That's by far the longest context window we have access to. The next best model is CLAWD 5 that has a 200, 000 token to windows. That's 10 X, the context window of the next best available platform. And that can be extremely valuable to anybody who is either doing research or wants to run huge data sets or wants to run a huge amount of code through it, et cetera, et cetera. So both highly capable. as well as the largest context window in Gemini 1. 5 Pro experimental right now that as I said, if you have these needs, you can have access to them right now. This is the direction it's all going. We're going to get more and more models coming out all the time that are going to be faster, cheaper, better, larger car context window, but in the immediate future, cheaper, The best one is Gemini. I know a lot of people who haven't even tested Gemini because they're either stuck on ChatGPT and or Claude. It's definitely worth testing Gemini out, especially in cases where you have a lot of data and you want to throw it in the chat and ask questions about it. In addition to having the longest context window, they have the highest level of accuracy in retrieval of data. Once it starts getting close to the limitations, which as I mentioned, in all the other competitors are significantly smaller. As I mentioned in the beginning, there are two big delays that were announced this week. The first one comes from Apple. Apple introduced its Apple intelligence capabilities earlier this year in a very big event and highly anticipated event in June. We reported all about that in previous episodes, and that was expected to be released with the release of the new iPhone. And that is not happening anymore. So right now, the delay talks about a potential release in October, which is a few weeks after the initial release of iOS 18 and iPad OS 18 that are planned for September. That being said, they're going to release it for developers for review and testing, probably this week as part of iOS 18. 1 that is going to be available for developers. What does that mean? Well, not much what we've seen in the past few months from all the big players, whether it's NVIDIA, which we're going to talk about in a second and Microsoft and Google and open AI and so on, that they're making really big announcement and showing really cool demos and then telling us we're going to get this sometime in the future. And that future is usually. pretty vague. So several of these companies said that some of these new features are coming in the fall, which gives them a lot of leeway of when exactly it's coming, because it might be a different fall in different states or different places around the world. So sometime between September and december, they're going to release the models. And even if there's additional delays, I'm not going to be surprised, especially for a company like Apple that is known for getting it right the first time, meaning they're going to make sure that this thing works and works perfectly and doesn't break anything and doesn't do anything surprising. And they're going to take their time to, as I mentioned, get it right the first time, a very different approach than what we've seen from other companies who are releasing stuff that is maybe not great. We've seen several of these from Google. So I'm not surprised by this delay from Apple. I'm not going to be surprised if there's going to be some additional delay, but I do think we'll get it before the end of this year because Apple is usually relatively on time when it comes to committing and releasing stuff that they promise. Now, the other delay That was announced this week is actually a lot more significant than the delay with the Apple products and that's NVIDIA Blackwell series of AI chips that they announced in a big and fascinating event in March of this year are going to be delayed. Now, why is this a big deal? It's a big deal because all the big companies are dependent on these chips for the development of their next models. So the new Blackwell chips that include B100, B200 and GB200 had some design flaws when they were designing them, which now slows down production. They were planning to deliver the first large shipments in Q4 of this year, and now it's planned only for Q1 of 2025. Understand how significant this is, Google has reportedly ordered over 400, 000 GB200 chips, which are the most advanced one, which worth about 10 billion. Meta has placed an initial order of at least 10 billion while Microsoft recently increased its order by another. 20 percent for these chips. So these chips, the new version of chips by NVIDIA is what's supposed to drive the next cycle of innovation and growth in the AI world. And their delay basically delays everything else. So that's all. It's not very good news for anybody in the AI environment, especially again, the big companies that can afford to buy crazy amount of these chips, because I have really deep pockets and spending tens of billions of dollars in buying chips and building new data centers. This news obviously comes as another hand wind to NVIDIA stock. So NVIDIA stock. Has peaked on June 18th with 135 per share. And it's now 107 per share, which is about a 20 to 25 percent loss in just a month and a half that being said the stock a year ago was at 44 and now it's at one Oh seven. So it more than doubled in a single year. I do think, and I'm not here to give any investment recommendations, but If this news tells you anything is the crazy dependency that the air world has on NVIDIA right now. There's some other players moving in both Intel and AMD are developing chips, and some of the big players are starting to develop their own chips that includes X and Tesla and Google and Microsoft. They're all working on their own chips. But as of right now, there's a huge dependency on NVIDIA chips. And so I do think that at least in the next couple of years, we're going to see Nvidia continuing to grow. Staying on Nvidia news on SIGGRAPH 2024 that happened this week. A lot of big, interesting news came from a lot of big players in the AI field, but specifically about Nvidia. They announced a partnership with Shutterstock that will allow AI to generate 3D models. The demo is extremely impressive. It allows users to generate 3D models using text and image prompts. It can produce preview renders of the 3D in less than 10 seconds. And it supports the creation of high quality 3D assets that can be then exported in standard formats that we know today for multiple platforms that reduces the process of generating 3D models by a huge amount from manual TDS process to an immediate, almost immediate process, just by using images and text prompts using NVIDIA technology Together with Shutterstock, it has been already released for beta as a beta testing for enterprise customers, but there hasn't been a clear defined plans for a broader distribution of this capability, but that's just another thing we should expect to have within the next, I'm guessing six months. Staying on NVIDIA and SIGGRAPH, there was a very interesting panel that included Mark Zuckerberg and Jensen Hung, and they talked a lot about the future of AI. Specifically, Zuckerberg mentioned a lot of very interesting development when it comes to AI and meta. So as we discussed last week, they just released Lama 3. 1, which is the best open source model in the world right now. And as I mentioned, it's ranking very high on the leaderboard right now. You should watch this video segment. It's not very long, but it's very clear that Zuckerberg and Meta are all in on integrating AI into everything Meta in parallel to releasing these open source models. They are planning to launch Create more and more tools for content creators and potentially have some kind of a cross integration between Facebook and Instagram, becoming a unified AI platform that can integrate different types of content and deploy it to different platforms afterwards. They're building multiple tools that will allow content creators and other people build agents as well as build better content. One of the cool features that they released is called segment anything. And it's actually the second version of that capability, DABD SAM2. And what it can do is it can grab a specific component from an image or a video and track it around. In some of the examples they shared, you can see a skateboarder skating on ramps and it's following him around, knowing how to pick him out of the video in every single point, which allows you to create really cool effects because you can track him specifically as an entity versus the old school way of picking frame by frame and trying to manipulate that. They're also showing hands that are working with dough, and you can pick the dough or one hand or the other hand while the video is running and manipulate and change and do whatever you want with those. So this sounds like a cool feature, but this feature actually has a lot more implications because it can be used in scientific research and other things in order to track aspects of video in manufacturing and so on. So very powerful capability that Meta is releasing as open source like everything else it's doing. He also talked about new versions of their products. AI glasses. He's stating that there's going to be future variations of this that will be extremely more powerful, that will include capabilities to do visual AI inside the glasses, real time translations and so on. And that he's planning to have several different levels of these as far as price points and capabilities. So people will be able to pick if they want to drive a Mercedes or a Ford. But the summary of all of this is that Meta is all in on integrating AI. Capabilities into its ecosystem that will make it more and more unique to them and will allow you to continuously work in the meta universe, whether using Facebook, Instagram, what's up, et cetera, benefit from this AI ecosystem within that universe that will obviously drive people to have more and more engagement with the meta universe. Not necessarily through the existing platforms, which are mostly Android and Apple iPhones where meta is paying a lot of money in order for people to use it. So if they're developed their interface into that through AI that will reduce their dependency and the amount of money they're paying these other players. And in the next and last segment of the news, we are going to talk about how these models are being trained and how they're doing stuff that is may or may not be legal or acceptable and how companies are working around it. We had several conversations, like we've mentioned this almost every news episode. And there's a lot of interesting news about this week as well. So we'll start with two pieces of news from Perplexity, just introduced a revenue sharing plan for publishers. So as I mentioned before, Perplexity is a great mix between Google search and GPT. It allows you to interact with stuff live on the internet while getting summaries and so on. As I mentioned, I use it daily. They just signed a revenue sharing deal with six different publishers, the Time, their Spiegel, Fortune, Entrepreneur, the Texas Tribune, and WordPress. This comes immediately after that perplexity announced that they're going to start having ads on their platforms. So basically what they're telling people is if you will allow us to share your content on our platform as Suggested additional links people want to follow we will share the ad revenue that's on that page with you. That's an interesting business model that nobody else is taking right now and it will be interesting to see how it evolves I don't think everybody knows yet how that's going to work out. But the fact that there are some big players that are willing to do that is interesting. I think to me, the most interesting is actually WordPress because they are planning to share the revenue with the actual people on the WordPress platform. So some of the revenue that's going to come from ads on perplexity is going to trickle down to the WordPress websites that are going to be mentioned. That's going to trickle down to the owners of the individual websites. That's also a first as far as I know, because all the other deals we've seen so far, the money goes to the platform like Reddit and not to the actual platform. People who write the articles on Reddit and there were a lot of furious people about the previous deals. So this particular deal actually potentially makes it fair. I don't know if we're talking about fractions of cents, which will make the whole thing irrelevant, but it's an interesting new business model that has a chance to be more successful than anything else we've been so far, as far as making it fair to the creators of the content and how AI is using it. And since we mentioned Reddit in a short mention, Reddit just started blocking search engines, as well as all the AI crawlers from getting access to any of their posts unless they pay them royalties. They're already getting a hefty amount from Google, but basically they're looking at the other players if they want access to Reddit to pay up as well. And in the last piece of news on this topic and this week overall, we're going to talk about anthropic. Anthropic has always presented themselves as the fair and safe and focused on good player in the AI industry, most likely for good reasons. But this past week, the CEO of iFixit, Kyle Waynes, has shared that the anthropic CloudBot had hit their servers almost a million times in just 24 hours. That's violating the company's terms of use. Now, the aggressive scraping has obviously triggered them approaching Anthropic and asking them to stop and claiming that what they're doing is illegal. So anthropic response that they are respecting their robot. TXT files that website has in the case of I fix it. They were not supposed to scrape it. And yet they did, they said that they stopped doing it as soon as they could. I fix it complained, but other companies such as freelancer. com reported a similar thing where there were experiencing aggressive scraping by the anthropic bot, despite the fact that their TXT file says not to do it. Now in this race for dominance in the AI world, it's pretty obvious that all these companies are doing that, whether it's Google or open AI or Claude in this case. And definitely there's been. multiple examples of perplexity doing that. They're really not abiding by the rules and regulations that have been set aside to what they're allowed to do and not allowed to do. Some of it is very clear. If the robot. txt file tells you, you can't scrape it, you shouldn't scrape it period. and some of it is on the gray area of what's fair use of content that's on the internet. This will be resolved in courts eventually. It will take a while. And at that point, it's probably going to be too late to roll AI back into the Pandora box. I don't personally know how this is going to end. I assume it's going to end with, as we're seeing now with different licensing or revenue sharing agreements that is going to allow the owners of the content, or at least the platforms on which the content is hosted to benefit from them. That's it for this week, a very eventful week with a lot of news from different aspects of the AI world. We'll be back on Tuesday with a fascinating conversation with Chris Daigle. We are going to talk about how actually to deploy generative AI in your business, which I'm sure is a question that many of you are asking. So wait for that episode on Tuesday and until then have an amazing weekend.

People on this episode

Head Shot

Isar Meitis

Host