EP 329: Google beats GPT-4o, NVIDIA delays and more AI News that Matters August 5, 2024

This is the Everyday AI show, the everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business and everyday life.

This has been one of the busiest week in AI news maybe since we started this show. I mean, we have huge announcements and updates from Google, Microsoft, Microsoft, OpenAI, Nvidia, Meta, literally. It is a who's who's list of AI developments this week and we're going to be covering them all in today's AI news that matters. What's going on y'? All? My name is Jordan Wilson and I'm the host of Everyday AI. This thing's for you. It is your daily live stream podcast and free daily newsletter helping everyday people learn and leverage generative AI. So if you're brand new here, thank you for joining us. Everyday AI it's, it's a live, unscripted, unedited show bringing you just the realest things in artificial intelligence. So if you're joining on the podcast, thanks as always. Make sure to check out your show notes for more on today's show as well as a link to our website. Your everyday AI.com. hey, this is not complete. Just listening in. You got to read about it. We're always sharing more information, so make sure you, you go to our website and hey, while you're there, go check out our thanks a million giveaway to celebrate a million downloads. We are giving away a year of either chat, GPT or whatever large language model you like. Perplexity Claw, it doesn't matter, we're going to be giving that away. It is free to enter, so make sure you do that on our website and then you just go ahead and click that little button. You can go copy your unique referral link, send that to your friends and some other ways that you can kind of earn points and enter our giveaway. All right, so let's get into it right away because there is a lot of AI news and developments, y'. All. And to our live stream audience joining us, Michael joining us from YouTube and Brian and Fred and Christopher, Rolando Madan, thank you all for joining us. Let me know as we go along. What do you think of these stories? Pretty big if I'm being honest. All right, let's jump straight into it. First, Microsoft has now listed OpenAI as a competitor after OpenAI's search GPT rollout. So according to reports from CNBC and others, Microsoft has officially listed OpenAI as a competitor in their latest annual report, adding some complexity to their long term partnership. So this marks a shift as Microsoft has traditionally included companies like Amazon and Apple, Google and Meta as its primary competitors. This is interesting because obviously Microsoft is a major investor in OpenAI, having invested previously around $13 billion into the startup and serving as its exclusive cloud provider and having a reported 49% equity stake in OpenAI. So despite this partnership, both companies are now stepping onto each other's domains, particularly in AI offerings, search and News advertising. So OpenAI recently announced a prototype search engine called Search GPT, positioning it as a direct competitor to Microsoft's AI products, even though it does use the Bing technology underneath from Microsoft. So Microsoft though offers its Copilot Chatbot via Bing and via their Windows 11, but that's also using the GPT technology. So it's funny, they're each using each other's technology, both very well vested into the other companies, but now are calling each other competitors. An OpenAI spokesperson stated that the competitive aspect of their relationship was always understood and that Microsoft remains a good partner. The relationship also saw turbulence when OpenAI's board ousted CEO OpenAI CEO Sam Altman last November without reportedly briefing Microsoft CEO Satya Nadella. Though Altman was quickly reinstated, Microsoft did briefly hold a non voting board seat at OpenAI, which it relinquished earlier this summer. And then in March, Nadella appointed Mustafa Suleiman, the co founder of DeepMind, to lead a new unit called Microsoft AI, bringing several inflection AI employees along. So Microsoft CEO Sadia Nadella maintains a close relationship with Altman, emphasizing ongoing collaboration and mutual demand for more resources. So yeah, pretty interesting if I'm being honest. I wasn't getting Microsoft to name OpenAI as competitor. But you think there might be some other motives here, right? There's a lot of federal investigations going on right now between all the big tech companies. So not just Microsoft, but Google, Apple, Nvidia, all of them. So Department of Justice, other federal regulators are looking at potential antitrust violations against some of these big companies for being too dominant. So you have to think, are there multiple motives here to listing OpenAI as a competitor? Maybe, maybe not. Yeah, I'm curious what our live stream audience thinks of that one. Yeah, it's, it's caught. Caught me off guard, but I think there might be more than, more than one reason for that. All right. A lot of acquisitions this week. Big ones too. So Canva has acquired Leonardo AI to expand its generative AI capabilities. So Canva has acquired Leonardo AI, a generative AI content and research startup, in a strategic move to enhance its AI technology stack. So canva's acquisition of Leonardo AI includes a combination of cash and stock, though financial terms were not disclosed. However, all 120 employees of Leonardo AI, including the executive team, will join Canva. Leonardo will still continue to operate independently, focusing on rapid innovation, research and development with Canvas support. The acquisition aims to help Leonardo expand its platform, grow its user base, and enhance API business and foundational model research and development. I did reach out to some some of my contacts over there at Canva and there's no, at least right now there's no direct integration with Leonardo inside of Canva, although that might change. So we will keep everyone up to date on that. One of the biggest kind of features inside of Leonardo is its Live Canvas feature, which allows users to combine text prompts with sketches to create photorealistic images in real time. Leonardo AI has over 19 million registered users and has reportedly generated more than a billion images. Canva does plan to integrate Leonardo AI's technology into its Magic Studio and Generative AI suite to enhance existing tools and introduce new capabilities. So that is not the only acquisition or Aqua Hire. All right, so Google has kind of acquired, but we'll say it's an aqua hire of Character AI co founders and they've signed a non exclusive tech agreement. So according to TechCrunch, Google has reacquired Character AI co founder and CEO Noam Shazir, who initially left Google DeepMind in 2021 to start Character AI, which is an A16Z back chatbot startup. So Shazir, known for leading the development of Google's Lambda conversational AI model, will rejoin the Google team in Google DeepMind research along with some of his colleagues from Character AI. So Character AI co founder Daniel DeFreitas, hopefully I got that name right. And additional employees from the startup are also going to be rejoining Google. As part of this move, Character AI's general counsel will serve as the interim CEO of the startup. So despite these high profile departures, the majority of the Character AI staff will remain with the company. So that's why it's like, okay, it's not really an acquisition. But they are acquitted, acquiring essentially their leadership team and then signing a non exclusive tech agreement. So Google has entered into that non exclusive agreement with Character AI to integrate its technology which will also provide increased funding for Character AI. So yeah, a lot of big acquisitions recently in the Generative AI front. All right, more Google news and this one is a big one. So Google Gemini's 1.5 Pro AI model has taken the lead for the most powerful large language model toppling GPT4O from OpenAI. So Google has launched a new version of its Gemini 1.5 Pro, its latest AI model, setting a new benchmark in the generative AI industry. So it is a new version of 1.5 Pro Pro called love this naming here, version 0801. Presumably that is the launch date of August 1st. So the new experimental version 0801 of Gemini 1.5 Pro is now available for early testing via the Google AI Studio and the Gemini API. So the new Gemini 1.5 Pro model has topped the LMSYS Chatbot arena leaderboard with an elo score of 1300, surpassing OpenAI's GPT4O, which had an Elo of 1286, and anthropic Claude 3.5 sonnet, which had an Elo of1271. And if you're new here, we talk about the chatbot arena a lot. It is essentially a place where you can go and go head to head. Essentially you can put in any kind of prompt and you will get a kind of blind results from two different chatbots and you vote for for which one is better. So it is a blind vote. There are millions of users that have used this voting system. So when we talk about ELO score, that's what it is. It is people who are judging head to head which company or which chatbot provides a better output. So the achievement signals a potential shift in the AI landscape, with Gemini 1.5 Pro taking the lead over OpenAI and Claude. So it has been called in early user feedback, Gemini 1.5 overwhelmingly positive, with some describing the model as insanely good. So the new version of Gemini 1.5 Pro excels in multilingual tasks, mathematics, complex prompts and coding, also securing the top spot on LMSYS's vision leaderboard for its multimodal capabilities. So not just text capabilities, but multimodal, so being able to understand images as well. A standout feature obviously is its expansive context window of up to 2 million tokens, allowing it to process vast amounts of information. One important thing to note here, if you are a free user or even a paid user of Google's front end Chatbot, so if you go to their Gemini chatbots, you are not going to be seeing this new model. So I will repeat, if you are maybe your company or you yourself personally are using the Google Gemini AI Chatbot on the front end as a kind of quote unquote front end user, even if a paid user, you will not be getting this newest, most powerful model. It is only available right now inside Google's AI studio or using it via the Gemini API. That's something extremely important to keep in mind. And it's something that Google does not always do a great job at communicating because Google has a lot of different versions of its newest models, right? So you have 1.5 Pro, you have 1.5 Ultra, you have 1.5 Flash, and then now there's different variations of these models. So you do not get these newest, latest and the greatest inside of the Gemini front end. This is only if you are logging in in the back end kind of in some of these developer tools. All right, more big news. Yeah, it's almost like we're getting bigger and bigger as we go on today with our weekly AI news that matters. So Nvidia has denied that they are in experiencing AI chip shortages and delays in its manufacturing. So Nvidia has denied a recent report that design flaws will develop delay the launch of its upcoming AI chips by three months or more, according to reports from the Information. So despite Nvidia's assurances, the report suggested that delays could impact major, major Nvidia customers like Meta, Google and Microsoft, who have collectively ordered tens of billions of dollars worth of Nvidia's GPU chips. So the AI chip company in Nvidia has unveiled its Blackwell chip series. In March, I was there on the on the show floor in San Jose when Nvidia announced its new Blackwell chip, the successor to the Grace Hopper superchip. So yeah, if you are not a GPU chip geek like me, essentially almost every single big company in the world that is building its own generative AI solutions is using Nvidia's GPU chips. And so a report of a potential three month delay is pretty significant. That is, I mean think just about every single major enterprise company here in the US has built some sort of generative AI feature or functionality and this could really delay future developments. Nvidia stated though, that demand for the Hopper chips remains strong and that production for the Blackwell series is still on track to ramp up in the second half of this year. The report mentioned that Nvidia informed Microsoft and another major cloud service provider this week of a delay in production of its most advanced AI chip in the Blackwall series. But again, Nvidia denying that report. Microsoft declined to comment further, while Meta and Google did not immediately, immediately respond to comments for these reports. And like I said, the delay in Nvidia's AI chip production, if accurate, could have significant implications for AI driven applications and technology, potentially affecting the timelines and operations of major tech companies. And this is especially timely right now when there is a lot of uncertainty right now in U S markets and just throughout the world. We've seen over the past week essentially all these magnificent seven AI tech companies. They've all seen their stock drop by anywhere from about 8 to 12%. So news of a potential chip shortage or chip delay could send some of these stocks down even more. So it's worth keeping an eye on the economic impact of this report and if it ultimately plays out or not. Are you still running in circles trying to figure out how to actually grow your business with AI? Maybe your company has been tinkering with large language models for a year or more, but can't really get traction to find ROI on Gen AI. Hey, this is Jordan Wilson, host of this very podcast. Companies like Adobe, Microsoft and Nvidia have partnered with us because they trust our expertise in educating the masses around generative AI to get ahead. And some of the most innovative companies in the country hire us to help with their AI strategy and to train hundreds of their employees on how to use Gen AI. So whether you're looking for ChatGPT training for thousands or just need help building your front end AI strategy, you can partner with us too. Just like some of the biggest companies in the world do. Go to your everydayai.com partner to get in contact with our team. Or you can just click on the partner section of our website will help you stop running in those AI circles and help get your team ahead and build a straight path to ROI on Gen AI. All right, new models, but small ones. Yes, Google has been on a little bit of a tear this week when it comes to new models. Also, by the way, if anyone is reading our newsletter, people are sleeping on the new Google AI image model that they released, Imagine three. Let me know. Live stream audience did anyone check out our little video review that we did? There it is. Very good. All right. Anyway, speaking of very good, Google DeepMind has just unveiled a smaller yet very powerful new AI model called Gemma 2.2B. So Google DeepMind has launched the Gemma 2 small model, a 2 billion parameter AI model, small enough to fit on a smartphone, yet offering performance comparable to OpenAI's GPT 3.5. So let me repeat that. Ready? A 2 billion parameter model is now more powerful than GPT 3.5, which was a 175 billion parameter model. So yeah, I'm not good at math on the fly, but that's like 1 percentage of the size. Yeah, so it is 1% of the size and as powerful, actually more powerful than the GPT 3.5 model which for many people throughout the world was their first taste of large language models as that is what ChatGPT kind of launched with as it gained popularity. So the new model from Google underscores a shift towards smaller efficient AI that can run on various devices, not just powerful computers. So despite the compact size, the Gemma 2 to be model achieves high performance through a training method called distillation introduced originally by Google's Jeffrey Hinton, the godfather of AI back in 2015. According to Google, the new Gemma 2B model offers best in class performance for its size, outperforming other models in its category shown in the Lymsys charts. While cloud based AI remains dominant for now on device on device AI or edge AI is expected to grow exponentially driven by data and privacy concerns. So for our live stream audience here, I'm throwing up this from Google showing the ELO scores again this is from the Chatbot arena and you'll see Here the new Gemma 2B. Gemma 2.2B model having a score of 1126 compared to the again an older model. But one of the biggest large language models of 2022, GPT 3.5 had a score of 1106. So a significantly higher Elo score for, for this model that again is 1% of the size. A 2 billion parameter model versus a 176 billion parameter model. And let me oversimplify this and, and yes, Michael, Michael here. Thank you Michael. Michael is going, going ahead and calling this out because I did call this actually like a year ago when I said the future of large language models is small models. I think after GPT4 I said future models are going to be much smaller because a lot of companies originally didn't want to really start using large language models for that reason because you have to use them in the cloud. And then I think we saw, you know, I mean you have to give credit where credit is due because you know, I mean Microsoft originally had some great small models, Mistral meta some of these other companies and now that seems to be pushing all the other kind of big players. So Specifically Google even OpenAI with their new GPT4O and their new GPT4O mini models are getting smaller and smaller which does a couple of things. Number one, it brings the inference costs down when models are smaller, it also brings the cost down to use them. Right? If these models are smaller and more powerful, presumably that is going to cut down on training costs. If the if it's on a per parameter basis, if the models are exponentially better, that means there is going to be less time to train these models. Whether you are fine tuning them, bringing in your own data via retrieval, augmented generation or rag. It is going to bring cost and deployment costs down. It is going to bring speed, you know, deployment speed way faster. So this is huge. But also bringing models local, right? So not just the ability as an example to be able to run a you know, model of the future locally on your computer, which is huge. Right. So normally essentially what happens is this, you know, if you're using a Gemini on the front end or a chat GPT, you know, on the front end, cloud, anthropic, etc, essentially you are just communicating with the cloud. So I, I don't think speed is really an issue. Right. Although speed by using an on device or edge a model is, is definitely a plus side. So I don't think people really notice like oh hey, when I'm talking with the cloud, it's, it's too slow because it's actually not. I would say so. But use running a model locally obviously is faster because you don't have that cloud connection. But like I said, as long as you have a good Internet connection, I would think for the most part most people aren't expect experiencing any of those speed downsides of working with the cloud. But the biggest thing is being able to run these models locally, especially on a phone. So a 2 billion parameter model can be run on almost all, all new smartphones in theory. Right. However, if you have an Apple iPhone as an example, you're not going to be running a Google small language model anytime soon. But this is a huge development. I cannot understate this enough because like I said once, I'll say it a hundred more times, the future of large language models is working with many small language models locally on your computer. On your phone models are going to continue to get smaller and presumably more powerful, which really changes what you can do with generative AI. All right, speaking of new modes and new models, OpenAI has rolled out, started to roll out anyways it's advanced voice mode for ChatGPT. So OpenAI has launched an alpha version of its advanced voice voice mode for chat GPT plus subscribers, aiming to enhance natural and responsive AI interactions. So on Tuesday, OpenAI started to slowly, very slowly roll out the alpha version of its advanced voice mode to a select group of ChatGPT plus users. So those users paying $20 a month, that's important to call out because it's not chat for Chat GPT teams yet, or Chat GPT Enterprise yet. I will tell you all this. We have both free accounts, Chat GPT plus accounts, Chat GPT teams as well as Chat GPT Enterprise for companies that we consult with. We don't have access to it yet on any of our accounts. Once we get it, you can almost, almost guarantee if that will probably be doing a show live where we're talking to the new advanced mode. I don't know, would you guys want to see a show like that, like talking to the new advanced voice mode live. So however, not a lot of people have this yet, but it is, I think, really going to change how we interface with AI systems in the future. So advance avm. We'll just call it AVM for short advice. Advanced Voice Mode. So AVM can detect and respond to emotional cues through vocal tone and delivery, adding a human like quality to interactions. Early user tests have shown enthusiasm, particularly for the AI's ability to simulate taking breaths while speaking, changing its inflection, tone and pace, just like I'm doing now. So sometimes AVM can talk very monotone like this and you might not really think that in advanced AI mode can then all of a sudden switch it up and talk with some inflection in their voice like this, and throw in emotion, but that's exactly what it does. So pretty interesting. The feature was trained on extensive audio samples of human speech, enabling it to mimic natural speech patterns, including inhalations. Users have reported minimal latency between Speaking and the AI's response, enhancing the fluidity of conversations and just the realistic nature. AVM can perform sound effects in accents, although it does still retain an American accent even when speaking other languages. So personally I'm excited to see these new features rolled out and like I said, once we get it, we will probably be doing a show and a pretty in depth review. But this is important for a couple of reasons. This is the future, this is the future of AI models, right? We've seen also Google kind of tease something similar with their project Astra, although who knows when Google will actually start rolling this out. But this is part of a larger rollout for OpenAI's GPT4O model. So in addition to this new voice mode, there will presumably pretty soon maybe be other capabilities rolling out to Chat GPT plus users. So one will be on the desktop app on Mac, the ability to push one button and Chat GPT can quote, unquote, see anything that you are working on on your computer and then you can be able to interact with it with this new advanced voice mode in real time as well as it can see through the camera. So a completely brand new way to work with AI models. This is just kind of the first step of this multi step process. But yeah, live stream audience, what do you think are you going to be once this does roll out? Are you going to be talking to this new model? Right. It might literally change how we work. Right. Like as an example, I think a lot of people kind of just either, you know, are, are talking with colleagues on meetings or they are working silently by themselves. This is, I think going to be one of these weird shifts in how we work where you're going to be working by yourself but talking all day. Right. So it is if, if everything works like initial live demos from OpenAI have shown, it will literally be like having the smartest human ever that can see over your shoulder and you can talk to it and it can respond in real time. So both strange and exciting prospects on the future of work. All right, Tanya, Hey, Tanya says to do a new episode on just that advanced mode when it comes out, throw, throw some ideas out there, live stream audience or hey, even podcast audience. I always put my LinkedIn and our, our team's email in the show notes there. So what would be a fun show for this new advanced voice mode? Because we could do a lot of things. I, I don't know, we could even have two versions of it. You know, once I get access to it on two different accounts, I don't know, we could have me talking to two different versions and interviewing two different versions. I don't know. You know, we could get super trippy with that. All right. Yes, there is still more. Our last AI news story for the week. Meta's AI studio has been rolled out and it introduces fully customizable chatbot and new AI tools. So Meta has unveiled some new advancements in its AI studio, including the ability to create customizable chatbots and powerful image and video analysis tools, marking significant strides in the AI landscape. So Meta's new AI studio allows users to create personalized chatbots by providing detailed descriptions, names, images and specific response instructions. So Instagram users can tailor these bots based on their content, topics to avoid and preferred links to share. So the latest version of Meta's large language model, LLAMA 3 is freely accessible to developers, researchers and startups, offering capabilities comparable to OpenAI's kind of GPTs that you can build inside of Chat GPT. Also, despite previous unsuccessful attempts with celebrity based AI chatbots, Meta believes that fully customizable bots will be more engaging and more widely used by users. So this is not a general rollout yet. We did check this just this morning. So I'm not sure if it has to do with the type of Instagram account that you have, whether it's set up as a business account or a profile account. We did. We have multiple Instagram accounts. We checked on both. One that's set up as a profile and one that is set up as a business and could not access the Meta AI Studio. So this may not be a full general rollout just yet. But you know, if you found as an example chat GPT's custom GPTs valuable, you might want to check out to see if you have access to this new feature in Meta AI Studio. So like I said, it does require you to log in with an Instagram account. So we will have those links in our newsletter. All right, that was a lot of AI news. My goodness. All right, let's just go over very, very quickly, a recap for the AI news that matters for this week. Yeah, because you can spend hours trying to keep up or on Mondays you can just tune in with us. We'll give you not only all the news, but also what it actually means. Right. We cut through all the smoke and mirrors, all the marketing and just kick it to you straight. All right, ready? So Microsoft has listed OpenAI as competitor, which like we talked about, pretty interesting considering that it has a reported 49% equity stake in OpenAI after investing more than $13 billion. And this obviously comes as OpenAI has started to launch some comparable products to Microsoft in its search GPT. Canva has acquired AI image and generating company Leonardo AI. Speaking of acquisitions or Aqua hires, Google has Aqua hired some character AI leaders as well as signing an exclusive deal to their technology. Speaking of Google, Google has unveiled its newest model, Gemini 1.5 Pro 0801. So a new version of of Gemini 1.5 Pro. And this is according to kind of Chatbot ELO scores now one of the most powerful or the most powerful large language model in the world. So big news there from Google Gemini, but right now it is not accessible on the front end. You do have to use their Google, their Gemini API or their AI studio. Also reported some Nvidia delays of up to 3, 3 months for GPUs. Although Nvidia has just kind of denounced that report and said they are not experiencing those delays. Google has released a very impressive small language model in Gemma 2 2B. So again, out bench or sorry, outscoring the kind of old school GPT 3.5 from OpenAI. So 1% of the size a very small model is out performing a very large kind of the model of 2022 so to speak. Also chat GBT has started to roll out its advanced voice mode as well as Meta has started to roll out its AI studio connected to Instagram accounts. So wow. Geez so much. Literally like we talked about every single big tech company here in the us at least. Like I said, big news from Google, Microsoft, Open AI, Nvidia, Meta. It is a blockbuster week in generative AI news. So we are not going to be sleeping. I hope you aren't sleeping on AI either. If you really want to understand artificial intelligence and get ahead, this is the spot to do it. Everyday AI. So if you find this valuable, please consider subscribing. If you are on listening on on Spotify or on Apple Podcasts, please subscribe and leave us a rating. If you're listening here live on social media, please repost this to your network so they can stay ahead. Guess what? By tuning in every day by reading our newsletter, you are automatically one of the smartest people in AI at your company. So congratulations for that. Please go to your everydayai.com Sign up for that free daily newsletter and we'll see you back tomorrow and every day for more Everyday AI. Thanks y'. All.

And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going for a little more AI magic. Visit your everydayai. Com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.