EP 314: AI News That Matters - July 15th, 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.

The world of AI was kind of wild this week. I mean, we saw new rollouts from a trillion dollar company, another trillion dollar company release some amazing research on AI robotics, and one of the hottest names in AI is teasing new models. Yeah, a lot to go over on this week's edition of the AI News that matters. What's going on, y'? All? My name is Jordan Wilson and I'm the host of Everyday AI and this show is for you. We are a daily live stream podcast, free daily newsletter helping us all learn and leverage generative AI to grow our companies. So if that sounds like you, thank you for joining us. And you probably understand this, like, how hard it is on a daily basis to keep up with everything that's happening in the AI world. Not just the tools we use, but the big companies that are driving futures like future decisions of generative AI. So every single Monday, we bring you the AI news that matters. So you don't have to sweat every single day trying to keep up if you don't listen to us every day. Maybe Monday's your day to get caught up and to get ahead. All right, so as a reminder, if you haven't already, please go to your everydayai.com that is the newsletter, so maybe you're listening, you know, while you're walking your dog or on the treadmill. So hopefully I'll keep this to, you know, 30 minutes so you don't got to walk too long. But make sure if you haven't already read our daily newsletter, it's written by me, a human, so you can keep up. Also, a little piece of news and I'm going to tell you more about this at the end. We are going to be launching our thanks a million campaign and tomorrow you, you got to join us for the live stream. We will be giving away a thousand dollars if someone can answer all of our chat GPT questions correctly live. All right, so more on that in a little bit, but let's get started with the AI news that matters for the week of July 15th. And hey, for our live stream audience, thanks for joining us. I would love to hear your thoughts on these new stories and what you think is going to matter. So everyone, Jay, Tara and Harvey and Fred and Christopher, Michael, Gene, thank you for joining us. So let's start at the top. So some new DeepMind and Google research here. So Google's DeepMind robotic team has demonstrated a significant advancement in robot navigation using Google Gemini 1.5 Pro. So the project was detailed in the paper mobility VLA multimodal instruction navigation with long context VLMs and topical graphs. That's such a long name for a paper. Anyways, the paper shows how robots can respond to commands and navigate complex office environments. So videos from the project reveal robots performing tasks like leading humans to specific locations and following directions written on whiteboards. It's super impressive. So we'll be sharing about that in the newsletter. So the robots were trained using a method called multimodal instruction navigation with demonstration tours or hey, finally a little code word here. Mint. That's mint, right? So multimodal instruction navigation with demonstration tours. And that involves walking them around office and pointing out landmarks. Yeah, weird. So the robot can kind of see and interact and learn on the fly. So in test from Google DeepMind researchers, according to that paper, the robots achieved a 90% success rate across more than 50 interactions with employees. Pretty impressive. So this development highlights the potential for more intuitive and efficient human robot interactions in various settings, potentially impacting workplace productivity and the future of automated assistance. So According to Google DeepMind, these advancements could lead to more practical applications of robots in everyday environments, enhancing their utility in both personal and professional context. Yeah, we talked about this in our, you know, 2024 predictions. Back in 2023 we said that, you know, AI powered robots or robots that are using large language models like this example here, Google DeepMind using the Gemini 1.5 model. We said this is going to become commonplace. So it's not, you know, it's not like you can walk in and see these robots already out. But I mean the research is showing that we are getting there pretty soon. I mean, what do you guys think of these, these, these robots, right? Like AI powered robots that can tap into large language models that can see and interact with you. You know, it's, it seemed like, you know, 20 years ago it was sci fi movies, but now it's essentially going to be here before we think. So I'm curious what you all think. All right, so next. Pretty big news here. So OpenAI has revealed a new five level classification system to track its progress toward developing AI that can outperform humans. So the system was first detailed by Bloomberg and was shared with OpenAI employees during an all hands meeting. So the tiers range from current AI capabilities level one to those that can manage an entire organization, level five. So for our live stream audience, I threw these different levels up on the screen here. So currently OpenAI believes it's at level one, but it is on the verge of reaching level two, referred to as reasoners. Yeah, so level two is reasoners human human level problem solving. So level two reasoners, which is where I think we will be within a couple of weeks or months probably. But that is when AI systems can perform basic problem solving tasks equivalent to a human with a doc, a doctoral level education without any additional tools. So this new classification system, again, this was leaked, but it doesn't look like, you know, OpenAI has said, like, no, this isn't true and usually Bloomberg reports are on the money. So this new classification system aims to provide clearer communication about AI safety and future advancements to investors and the public. So the announcement underscores OpenAI's leadership and the ambition in the competitive field of AI development. Yeah, it's getting wild, right? And understanding these levels can help individuals and companies anticipate the capabilities of future AI systems and prepare for their potential impact on various industries. All right, yeah, we talked about how just about every trillion dollar company was coming out with some sort of news this week and Amazon is no exception here. So Amazon Web Services or aws, announced a series of significant updates to its AI product suite, aiming to outpace competitors as its lead in the cloud market narrows. So we did mention this in our newsletter. I think this was midweek last week, but AWS has just launched over 300, yes, 300 generative AI features this year, more than twice combined the total of other large cloud providers, according to AWS VP Matt Wood. So the majority of these new updates from Amazon were driven by customer demand to keep up with rapid tech advancements. So AWS Bedrock customers, and I think this is probably one of the biggest ones that was announced last week. So AWS Bedrock customers can fine tune large language models including anthropics Claude 3. I believe that's only on the Haiku or their smallest flavor of Claude 3. But can fine tune anthropics Claude 3 using your own data to adjust model behavior. So that's a pretty big one right there. The first instance where you can use a major platform where you can fine tune Claude's, Claude's three any of Claude three offerings. So there's also a new guardrail feature that Amazon has introduced to detect and filter hallucinated responses, improving accuracy by over 75%. That's according to Amazon. AWS has also launched AWS App Studio, a generative AI powered service that allows users to create enterprise grade applications through natural language instructions. In minutes. And y' all like, I'm not being like, I'm not gonna lie. The AWS platform in Bedrock is just super fun, even if you're not a developer, to be able to go in there and create apps kind of in real time. So, right. We've been talking about this with Anthropic's new Artifacts feature, which I think is really amazing, but it's kind of already been out for a while using aws, both their app Studio and some of the other features and functionalities in their Bedrock system. So you can access a lot of these for free just to kind of understand what they're about. And I would encourage you, if you haven't already, to go do it because it's pretty impressive. So Amazon anticipates AWS will contribute a hundred billion dollars in revenue this year fueled by strong demand for AI capabilities. Speaking of Amazon, that wasn't their biggest announcement, I don't think. Right. Hey, here's 300 new AI features. I'd say the biggest one that's probably going to affect the most people immediately is finally Amazon has released the new AI features coming to the Amazon shopping experience. So Amazon has finally launched Rufus, a generative AI powered shopping assistant, which should now be available to all US customers in the Amazon Shopping app. So Rufus helps users understand product details, get recommendations, compare options, access order information and more by answering a wide variety of questions. So I'm curious if any of you all have found Rufus here in the. In the us So I haven't been on Amazon, but maybe for that reason. But this is tricky timing, right? So we talked about this new Rufus, kind of this new AI assistant that Amazon was going to be launching. We talked about it probably about four or five months ago when they first launched, kind of announced this, but now this is a general rollout here in the U.S. interesting timing. So Amazon must have a lot of confidence in this new AI powered shopping assistant to launch it a couple of days before their biggest days of the year, which are Amazon Prime Day, which I believe launches tomorrow. A two day sale where Amazon just, you know, has all these bonkers sales. So pretty interesting that they're going to be, you know, there wasn't a slower rollout to this and they're like, hey, let's just put it out right before Amazon Prime Day. What could go wrong? I don't know. So, hey, and let me know in the comments, is anyone out there going to be shopping on Prime Day? What are you going to be buying? I have a couple of things that are on my list this year. But, you know, I used to be one of those people that, you know, went. Went wild, right? Black Friday, Prime Day. Not anymore. You know, but a couple things on my list. I'd love to hear what's on yours. Maybe you can save me some time so I don't have to chat with Rufus. You can just tell me what to buy. All right. Yeah. Gosh, you thought, you, you thought. We started with the big AI news, now we're saving it for the end of the show, y'. All. This is some big news. Some big news here from OpenAI. It's so big, I gotta take a sip of coffee, y'. All. 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 Genai. 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, here we go. So OpenAI is working on a groundbreaking project codenamed Strawberry that could significantly enhance the reasoning abilities of its AI models, according to Reuters. So OpenAI's Project Strawberry is designed to improve the advanced reasoning capabilities of AI models. So the project aims to enable AI to plan ahead and navigate the Internet autonomously to perform what they're saying is, quote unquote, deep research. All right, so details of Strawberry right now are closely guarded even within OpenAI. And the project is still a work in progress according to Bloomberg reports. So the initiative involves a specialized post training process to enhance AI performance. After initial training on large data sets, the project is seen as crucial for overcoming current limitations in AI such as common sense reasoning and solving multi step problems. So OpenAI demonstrated a model with new human like reasoning skills at an internal meeting According to reports. But it is unclear if this was related to this new Strawberry project. Hey, maybe good timing, right? Fall, like late summer fall, you know, Strawberry. Hey, I like it. Finally something that, a name that seems timely, right? So the project is part of a broader industry effort and this is everyone, right? To improve AI reasoning. So when we talked about this kind of five levels of AGI, right, like being able to better work with humans and to better understand complex nuances and reasoning is huge. So companies like Google, Meta and Microsoft are also exploring similar advancements, right? That's the next step in large language models. So Strawberry also, according to reports, may involve techniques similar to the very famous, which we've shared about in the newsletter a lot, Stanford's self taught Reasoner, or star, which enables AI to iteratively improve its own intelligence. Also. Yeah, you knew this was coming. The new Strawberry project is rumored to be the next evolution or the next step in the infamous Q Q Star model from Open AI that was rumored back in late fall of 2023. So yeah, this was all the rumor mills were going wild. And this was back when Open AI fired its CEO Sam Altman, and then a couple of days later rehired them. A lot of the reports were tied to this Q Star, you know, kind of this mysterious model that people said was actually so good that it was dangerous. And then in the coming months and quarters, we've seen a lot of shakeups at OpenAI and their safety team in particular. So we haven't really heard a lot on this Q Star model since then. Right. So essentially it did cause a lot of internal shakeups and reportedly a lot of people were kind of scared at the capabilities of this Q Star project that seemingly never came to fruition. But here we are reportedly with the continuation or the next phase of this Q Star project. Now we're hearing it's called Strawberry. So according to reports, the company's plan, the company OpenAI plans to use these capabilities for autonomous web research and potentially for software and machine learning engineering tasks. So this one. Yeah, hey, I agree with what Juliet. Yeah, saying a cool seasonal and undercover name. Yeah, agree. But this one is interesting because I've been wondering and I'm always looking like, what happened with this Q Star, right back in October. November. December. So back in, you know, fall of 2023, this really shook the AI world upside down, if I'm being honest. Right. Reportedly, this is one of the things that led to the firing of, of Sam Altman. Right. We talked about all of the safety, all of the Safety shakeups at Open AI as well as Ilia. Right. When started ssi. Right. So we've shared about that in the newsletter as well. So uh, yeah, this is worth paying attention to. I don't think this is one of those instances. Yeah, we see a lot of reports, you know, from, from Bloomberg, from the Information etc. Some of them come to fruition, some of them change. Right. You know, one of the bigger ones was, you know, we were seeing all these reports that OpenAI was coming out with a search engine. They weren't, although I do think they came out with an answers engine that no one paid attention to. But I do think this Strawberry project here in particular is pretty interesting and it's worth paying attention to. Right. And it does coincidentally or not, come out right about the same time as other reports. Right. The other report that we just talked about also from Bloomberg. Right. So Bloomberg clearly has the, the inside track here on this new five level AI progress tracking system. Right. The five levels essentially to AGI. So I do think that it's going to take a big leap to go from level one to level two to level three to level four. I'm going to go ahead and bring that back up, you know, because actually our last news story kind of has to deal with this as well. So level one is chatbots, you know, AI with conversational language. We've been there for a while. Level 2 is Reasoners, Human level problem solving. And I do think that we are very close to that. I think especially whenever we see a new model, I think we will surpass that level three, which I think is where a lot of the big, big companies are investing time, money, money and resources into are agents. And that's where I want to kind of transition this with this strawberry project from OpenAI. Because one of the things that was talked about in this report was its ability to navigate the Internet and perform deep research. Right. So similarly to kind of perplexity. Right. But when you can combine that with a large language model and you can combine that with a model that at its core has better reasoning, can better understand human nuance, that's when I think you start to get to that level three, Right. Where agents can perform actions very similar to humans. Right. And then level four you have investors or sorry, innovators or AI that can aid in invention. And then level five, which I think is essentially, all right, we're here at AGI, which is when organizations and AI that can do the work. So when AI can do the work of an entire organization. Right. So that's artificial general intelligence or when AI systems are smarter, one system can be smarter than any human being at any particular task. Right. So this is kind of the five levels and then this new report from Bloomberg talking about Strawberry. Right. And it does seem like this is where we've already heard and where we've been hearing for months that not just OpenAI, but also this is seemingly where Google is heading, this is where Meta is heading. More powerful, more capable models combined with the ability to research, combined with some of these agent powers. Right. So we've heard not just Microsoft, not just Meta, but also Google and OpenAI. So the four biggest players in gen in generative AI in large language models have already stated that this is where they are playing. They are playing in the area of agents. Right. So we saw Google astray, kind of teased by Google. We've seen this new, this new iteration of OpenAI's GPT4O model, the one that is live, the one that you can essentially talk to in real time, you know, click one button and it can see and understand what's happening on your desktop. Right. Amazing. But I think that is where we are headed next. Right. I think people are so focused on, you know, oh, GPT 5 or you know, Gemini 2. Right. Rightfully so. Right. It's, I, I think it's always great to, to look at what model is next, but I don't think it is the model that we should be paying attention to. It is the features and functionality of that model to work more in an agent type capacity. And that's essentially when it can string together. Right. So maybe you use a large language model right now in your business, in your, in your work and there's a lot of human interaction between. Right. You essentially have to give the model more information, you have to correct it. You have to do a lot of prompt engineering. Right. You have to get a lot of information in its context window and then keep performing multiple tasks to ultimately get to an end goal. So where we are seeing as we kind of talk about this level two to level three transition and where like we said, Microsoft, Google, Mata, OpenAI, they're openly saying we are working on agents. So that is not just more powerful model. Yes, we need a more powerful model to get there and to be able to do this in real time at a level that is highly accurate, more accurate than humans, but that, that is the next step. Right. It's not just a more powerful model, it is models that can complete an entire workflow. So let's say as an example you have a project that would normally take you 10 to 15 hours, right. And maybe now you're great at large language models, you know which one to use and now you can do that in 10 to 15 minutes. But it does require a lot of you, right? It requires you interjecting at multiple points throughout that multi step process. That's not the future of how models are going to work. You're going to give them your starting point and your ending point and then you are essentially going to be a co pilot. Right. Just like how pilots right now, all they really do is take the plane off and land it. That's what we're going to be doing with generative AI systems as we kind of transition from traditional, quote unquote, traditional large language models to agent models. Right. We, we humans are going to be there for the takeoff and the landing and we are going to be watching and monitoring, I think multiple agent workflows happening. I think that's the future of work. Right. Like pilots, except instead of being on one plane, you're going to be on a lot of planes. Right. So pretty. Pretty interesting and pretty exciting, pretty exciting information here from this Bloomberg report. So yeah, let. Yeah. What's Venkatesh asking? What's level five Autonomous enterprise? Yeah, that's just when your, your AIs are running the organization. Artificial General intelligence at that point. So no, this is not Juliet. Great question here asking if this is the AI Health coach, that Open AI. No, that is not the, that Thrive project. This is separate. Right. So Strawberry is, according to reports, I do think it is the continuation of this Q Star project. But I do think that this is essentially going to be whatever OpenAI is working on in its agent workflow, which we kind of already saw. It's, it's kind of, you know, whether you codename it her, you know, I call it Live Omni. But this next iteration that OpenAI teased at its fall event or sorry, at its spring event back in May. Right. That we still don't have yet, but we should be getting, according to reports, any week now. They said about three weeks ago that it was going to be delayed a month. So in theory we should be getting that any day now. But I actually think this Strawberry is the next iteration of that. It is when there is less human need, there is less prompt engineering in theory and more just agent workflows. But obviously they have to kind of roll out this updated model first. All right, there's one more. Yeah, we have one more piece of AI news, y'. All. So new Models are already appearing and I think we should be paying attention to this and I'll tell you why. So new models from OpenAI and Google have just popped up in the chatbot arena. All right, so that is pretty, pretty big news here. So new models from Open, AI and Google have just popped up in the chatbot arena. So the chatbot arena, and we're going to be linking to this in our newsletter, is from Hugging Face and LMS y s. I don't know if that's limsys or if you're just supposed to say LMS y S. All right? But anyways, and I have screenshots on here for our audience because I actually noticed this, y'. All. I go, yeah, maybe I need hobbies, but I go in probably two to three times a week. I spend about 15 minutes in the chatbot arena. Essentially you can go do what's called a head to head battle. All right? And you essentially put in a prompt, get two responses from two different models and it doesn't say what they are until you vote on it. And I noticed this weekend as I was doing this because it's part of my weekly routine, right? Because we spend all this time keep keeping up with AI so you don't have to. I noticed two new models that I'm like, oh, interesting. And we actually saw this right before OpenAI released GPT4O. We saw them kind of testing two different models codenamed in the Chatbot. Chatbot arena on this, on this website. Right. So here are the two models that I just noticed. And I'm like, oh, is anyone else seeing these? I haven't seen any news reports. I saw a couple on Twitter who saw the same thing. So I know it's not just a hallucination or a bug. So some notable new models include upcoming GPT Mini from OpenAI and Eureka chatbot from Google. Right. So I, I have those two screenshots. I was able to kind of bring those up in the head to head comparison. Unfortunately, when you bring them up head to head, you can't keep chatting with them. You essentially get one prompt, you get two responses, you vote which one's better, Then you get to see which models which. But you can't just keep chatting or keep using. So it's a little hard and time consuming to really see the power because it might take you 30 minutes, 45 minutes to generate a certain model that you want because it's random and there's dozens of models in there. But taking a page perhaps out of OpenAI's book. So. So when OpenAI was teasing or had their codenamed model in the chatbot arena, its CEO Sam Altman sent out some kind of cryptic tweets mentioning the code name on Twitter. And then you also just saw that over the weekend from the lead, the lead of product for Google AI studio, Logan Kilpatrick kind of did the same thing. He said, I just had a small Eureka moment, put that out on Twitter when right about at the time when you saw the Eureka chatbot from Google start appearing in this head to head. So this is kind of a new, I guess, strategy from big companies before they release models. And actually in OpenAI's case, they put three of them out there and then they only released one. Right. So I think this is also a way for big companies to use us all as guinea pigs. Right? There's been more than a million votes on the leaderboard. So it's essentially a way to see, hey, which model is actually best for users without telling anyone. Right? So in the end, yes, us humans are kind of the testers. We are the real human feedback, right, for these models before they actually hit production. But if you are someone that follows models and you're always wanting to know the latest and greatest, some pretty big news there. All right, that is it, y'. All. So let's do a quick wrap up for the AI news that matters for the week of July 15th. So Google DeepMind showcased some advanced robot navigation with Google Gemini 1.5. Next we saw a Bloomberg report on OpenAI's five levels of steps toward AGI. And then we went through all five of those steps. Next we talked about Amazon's AWS unveiling some pretty major AI updates to stay ahead in the cloud market with more than 300 generative AI features. And also kind of related them launching their AI powered shopping assistant, Rufus, just in time for Amazon Prime Day, which is tomorrow. Then we went over Amazon's Strawberry project, which aims to just change AI's reasoning capabilities, detailing that report from Bloomberg, kind of the continuation of this reported Q Star project. And then last but not least, we talked about new models kind of leaking or appearing in the chatbot arena, both from OpenAI and Google. There were some others from Command as well. All right, so if this was helpful, I hope it was. If you're listening on the podcast, we do this almost every single Monday. A live, unedited, unscripted look at the AI news that matters to you. So that was our quick recap. And speaking of things that matter, you might want to tune in tomorrow live. So if you are normally a podcast listener, this is one of those. You might want to adjust your schedule. I'm giving you a heads up now so you can't say I didn't warn you. You're going to want to join us live tomorrow because we're going to be kicking off a little campaign initiative called Thanks a million some more on what that is tomorrow. But what we are going to be doing, if anyone can answer live, right, it's one of the reasons I do this thing live. I love interacting with our audience and our audience being able to network and engage with others. But we're going to do a fun little, fun little test, fun little quiz tomorrow because I always hear, oh, I'm a chat GPT expert. Well, are you? Let's find out live. And if anyone, if anyone can answer all of our questions correctly, we're going to be giving away. We have a thousand dollar prize pool. So if one person does it, you get a thousand dollars. If 10 people can somehow get every single question, they're each going to get a hundred dollars. But you have to be here live, right? And so if, if you're an avid listener to this show, you're gonna have a huge competitive advantage. So make sure that you sign up or that you are reading our newsletter because we're going to be leaving the link for the live show in the newsletter or if you're always tuning us, tuning in live on on YouTube or LinkedIn. That is going to be our show for tomorrow. Tuesday, July 16th. Right? You're probably going to need a couple extra dollars to go out and spend on Prime Day. Anyways, now that AI Rufus is going to be spending all our money for us. So make sure you tune in tomorrow for that 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.