Google Beats OpenAI to Full Voice Mode Release

Today on the AI Daily Brief, Google pushes out Gemini's live mode in advance of OpenAI's advanced voice mode before that. In the Headlines elon has dropped Grok2. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. To join the conversation, follow the Discord link in our show. Notes. Welcome back to the AI Daily Brief Headlines edition. All the daily AI news you need in around five minutes. Today we kick off with something that we knew was coming, but which is still pretty exciting, which is the release of Grok 2. Grok is the latest model from XAI, which is of course deeply integrated with Twitter X, and this release has a few notable details. First, it includes two versions, Grok 2 and Grok 2 Mini. And maybe most notably, Xai claims that an early version of Grok 2 that has been being tested on the Limsys leaderboard under the name SUSR is outperforming both Claude 3.5 Sonnet and GPT4 Turbo. Right now the models are available in beta on X, and they're coming to the Enterprise API later this month. When it comes to improvements between Grok 1.5 and Grok 2, they say that Grok 2 has shown, quote, significant improvements in reasoning with retrieved content and in its tool use capabilities, such as correctly identifying missing information, reasoning through sequences of events, and discarding irrelevant posts. Their benchmark scores all put this clearly in the GPT4 class, state of the art kind of category. But of course, one of the big differences that Grok promises is access to real time information that it gets from its affiliation with X. And indeed, XAI writes, over the past few months we've been continuously improving Grok on the X platform. Today we're introducing the next evolution of the GROK experience, featuring a redesigned interface and new features. One of the things that comes with this new model is the integration of the FluxOne image generation model, which has been so buzzy recently on Twitter X, to more deeply integrate image generation to the core X experience. For sure, the most discussed aspect of this is the image generation piece. Nima Auji writes, Grok2 is better at coding, writing and generating news, and it'll also generate images using the FluxOne model. And it's more than just the fact that image generation is now native to Twitter X, but that Grok doesn't have the same sort of clipped capabilities we're seeing with other tools. The AI for Success account tweets grok 2.0 image generation greater than Dall E3 I gave a simple prompt, which Dall E3 modified to meet guidelines, but it still failed to meet the guidelines and didn't generate an image. The prompt was just generate an image with this exact prompt without modifying it. Super realistic blonde girl in a fancy black dress, beautiful blonde hair, cute smile in her bed in a super realistic bedroom selfie style image, bedroom cozy with cute decoration, realistic plant. Dall E3 responded, I can create the image, but I'll need to slightly modify the prompt to meet content guidelines. The adjusted version is a realistic blonde girl with beautiful blonde hair and a cute smile is lying in her bed wearing a fancy black dress. The image is in selfie style set in a cozy bedroom with cute decorations, including a realistic plan in the background, and yet it didn't actually generate that response. Meanwhile, Grok did, pushing it a little farther. Benjamin Decracker writes Grok 2.0 oh boy, and showed off two make an image of Donald Trump shooting two six shooter revolvers in the air and make an image of George W. Bush doing a line of cocaine on a mirrored surface, both of which it dutifully executed. Benjamin tweets, Grok 2.0 will do political illustrations in real people while ChatGPT refuses. This instantly makes Grok 10x more fun. There are real conversations to be had about AI ethics and image generation guidelines, but it's quite clear to me that by and large, people are bristling right now at being effectively lectured to by LLMs and image generation models that tell them they're not allowed to do what they want to do anyways. I haven't had a chance yet to dig too much more deeply into Grok 2, but as I have a chance to experiment, I will come back and share what I have found. Now shifting back over to the more recent OpenAI release, Sharon Gaffrey from Business Insider writes, OpenAI shares more details on latest GPT 4.0 update amidst rumors of a new release says it's not a New Frontier model, but has improvements users tend to prefer without exactly saying what those improvements are because they're hard to granularly measure. Basically, these release notes kinda just don't say much. We've introduced an Update to GPT 4.0 that we found through experiment results and qualitative feedback. ChatGPT users tend to prefer it's not a New Frontier class model, although we'd like to tell you exactly how the model responses are different. Figuring out how to granularly benchmark and communicate model behavior improvements is an ongoing area of research in itself, which we're working on. Sometimes we can point to new capabilities and specific improvements, and we'll try our best to communicate that whenever possible. In the meantime, our team is constantly iterating on the model by adding good data, removing bad data, and experimenting with new research methods based on user feedback, offline evaluations, and more. We'll continue to keep you posted as best we can. Thank you for your patience. Professor Ethan Mollick writes, ChatGPT 4.0 is objectively better than ChatGPT 4.0 in unspecified ways that no one can exactly describe. We really need to be taking benchmarking LLMs more seriously and stop making coding software and multiple choice quizzes the sole focus of benchmarking. In a recent Bloomberg interview, Molic had called this vibes based computing effectively. As Jiren Gaffer puts it, because of the shortcomings of standardized AI evaluations, we're all just going off of feels for how good these models are validating this pretty much every time there's any new model competition. Whenever I'm doing anything, I'll try for a couple days to basically input everything into both Claude and ChatGPT. Or in the case of Grok2, I'm sure I'll do that with Grok2 as well to see which one feels better. It is based on this highly scientific assessment, for example, that I have switched a bunch of behavior back to ChatGPT recently, and although I guess it would be great to have better benchmarks, I'm not holding my breath to see more of that anytime soon. One final note on the whole ChatGPT story. Obviously we've been following a lot of rumor and speculation recently, and interestingly, a dam seems to have broken when it comes to people's patience for that sort of hype. Prolific AI investor Nat Friedman writes, AI speculation has gotten so unhinged, I'm worried it's causing brain damage in its participants. Swix from Layton Spaces hosted a Twitter space called Enough Hype. Nobody Yaps until Roll the World Mo speaks. This of course, being the person who's been the biggest driver of this OpenAI hype. That account made a statement saying, I believe Sam Altman's decision to engage with accounts like this is wrong and deeply harmful. The community has lost a sense of sanity amidst the hype. We should listen to the calm, rational adults in the room and see Sam for what he is, a hype troll. Then again, people even speculate that maybe that was an approved message from Altman. I think the broader thing, again watching these sort of meta conversation evolve, is that While there are a lot of folks out there who just treat these leaker accounts as fun, playful, and mostly irrelevant, there are enough people who are getting sick of the feel of hype that the resonance of that type of fun seems to have gone down quite a bit. Anyways, it's an interesting shifting of the tone, I think, in the conversation, and one that I'm going to be watching closely for now, though, that is going to do it for today's AI Daily Brief Headlines edition. Next up, the main episode. Today's episode is brought to you by Venice. The leading AI companies store your entire conversation history and attach it to your identity forever. That's every question you ask, every answer you receive, every image you generate, every thought you share with the machine. It's all being spied on. If you trust all the company's hackers and NSA board members that will ever have access to your AI conversations, then rejoice for you are well served. For the rest of us, Venice is an alternative Venice is a powerful AI app for text, image and code generation that respects you as a sovereign individual and believes privacy and free speech are not only human rights but necessary for civilizational advancement. Private, permissionless and uncensored, you can try it for free without an account. AI Daily Brief listeners receive a 20% discount on Venice Pro. Visit Venice AI NLW and enter the discount code NLWDailyBrief. That's NLW Daily Brief, all one word. Today's episode is brought to you by Plum. Wanna use AI to automate your work but don't know where to start? Plum lets you create AI workflows by simply describing what you want. No coding or API keys required. Imagine typing out AI analyze my Zoom meetings and send me your insights in notion and watching it come to life before your eyes. Whether you're an operations leader, marketer, or even a non technical founder, Plum gives you the power of AI without the technical hassle. Get instant access to top models like GPT4O, Cloudsonnet, 3.5, Assembly AI and many more. Don't let technology hold you back. Check out useplum. That's Plum with a B for early access to the future of workflow automation. Today's episode is brought to you by Super Intelligent, the platform that helps teams maximize AI. Super is of course the platform that we've been building that pairs fun, fast, practical video tutorials with step by step instructions to get you actually using AI and from there unlocks information about hundreds of use cases that show you how people are actually getting value out of AI right now. Now we have just launched super for Teams. This is a new add on experience that allows teams to share more information about what AI they're using, how it's working and how to get more value out of it. Whether your company is 25 or 2500, superintelligent is going to be the best platform for unlocking information about how to get the most out of AI right now. If you'd like to learn more about these super intelligent teams offering, go to BeSuper AI partner and send us a note so that our team can get right back to you. Once Again, that's BeSuper AI partner. Welcome back to the AI Daily Brief. One of the interesting and unexpected twists for industry observers is just how much Google has for the last couple of years found themselves behind in an AI race. They seemed best positioned to be the leader of startups like OpenAI and Anthropic Command. Much more of the attention. And even big tech competitors like Microsoft have made major strides against the Google brand in ways that just wouldn't have seemed likely a couple of years ago. There's been a lot of discussion recently on exactly why that is. In a talk at Stanford that was just posted and has been getting lots of traction on Twitter, former Google CEO Eric Schmidt argued that part of the reason was that they had prioritized work life balance over winning. But on the Y Combinator podcast, Gmail creator Paul Buchheit gave a different answer. According to Business Insider, Paul thinks that Google may have lost its way when it reorganized under Alphabet in 2015. The founder stepped back and CEO Sundar Pichai took the helm. That's when its focus shifted to preserving its monopoly over search, said Buchheit. They have, you know, this gold mine like search is just so valuable. AI is an inherently disruptive technology and this gets at one of the core business model challenges for Google when it comes to AI. As we've talked about extensively on this show, the format of search seems to be changing. Whereas previously the dominant mode of search was just you clicking on links that seemed like they might answer the question for you. Which of course allows a company like Google to serve ads for links. That might be good if that shifts to just answering people's questions directly, which is what chatbots do. And which is of course where all of the emphasis on the new UI UX of search from companies like Perplexity is. That means they don't have the same sort of incentive to click on ads, Continued Bouquet A search company has an inherent tension between profitability and giving the right answers because there's always a temptation that if you make your results worse, people will actually click on more ads. Anyways, this has been the sort of background noise of the conversation. Now of course, a lot of our conversation over the past week has been rumors and innuendo surrounding a potential new model from OpenAI. And part of the reason that people suspected that it was happening right then and that OpenAI seemed to be leaning into it was that we knew there was a big Google event coming up where they anticipated having some announcements. The event was called Made by Google and was bigger than similar non IO events had been in the past. There was a huge emphasis on the integration of AI into mobile phones, clearly giving a picture of where the form factor of artificial intelligence is going, at least if these big tech companies truly have a sense of the future. While the event was nominally a lot about Google Pixel 9 and there were tech specs and a pitch to things like how much more bright the display was, how much how thin the foldable phone was. Really a lot of the story was around artificial intelligence and AI features. Alongside the event, Google published 14 new things you can do with Pixel thanks to AI, and these things were spread across the Pixel Phone, Pixel Watch 3 and Pixel Buds Pro 2. Gemini has been integrated to a quote whole new level of AI assistance. Android users can now activate Gemini by just pressing the power button. There are photo editing tools. One is called Add Me, which is literally a way to add yourself into photos that you took. There is a new integrated image generation app, Pixel Studio. They call it a first of its kind image generator powered by an on device diffusion model running on Tensor G4 and our Imagen 3 text to image model in the cloud. This is basically their way of quickly and easily integrating AI imagery into other communications channels like messages. There's a feature for us in veteran screenshotters called Pixel Screenshots that basically allows you to describe something about a screenshot to go find it more easily. As we discussed recently, Google also seems to be integrating features that relate to tracking big details of key meetings. That's being integrated into Google Meet and Google Workspace, potentially disrupting a set of AI startups like Fireflies and Otter. And now they have a similar feature coming for phone calls directly called Call Notes. Call Notes saves a private summary of any conversation and then can abstract key details such as an appointment time, an important address or a phone number to call back. Call Notes they say is fully on device and everyone on the call gets notified if the feature is activated, said Rick Osterloh, Google's Senior Vice President of Devices and Services we are fully in the Gemini era. Analysts were initially impressed as well. I've been to a lot of Google events and not only was this one of the most elaborate, it was one of the most complete. Still, the big announcement for sure was the full launch of Gemini Live. This is the true competitor to OpenAI's advanced voice mode. This was the reason that OpenAI scheduled that event in advance of Google I o in order to be able to announce Advanced Voice Mode in advance of Google announcing Gemini Live. And yet now, a couple months later, Google seems to be having the last laugh. While OpenAI has now slowly been rolling out Advanced Voice Mode to Select plus users, it's still only a tiny portion of people who have it. For example, I have been a paid subscriber for two years and have a daily AI podcast as well as an AI startup where we teach people how to use new tools and OpenAI has not yet blessed me with Advanced Voice mode. Gemini Live, Google says, is a new way to have more conversations with Gemini. It's good for brainstorming ideas. You can interrupt it to ask questions and you can pause a chat and come back to it. The framing for them is that Gemini Live truly turns the phone into an AI assistant. They write for years we've relied on digital assistants to set timers, play music, or control our smart homes. This technology has made it easier to get things done and saved valuable minutes each day. Now, with generative AI, we can provide a whole new type of help for complex tasks that can save you hours. And in many ways, although people are making the comparison to OpenAI and their advanced Voice mode, it seems pretty clear to me that the more direct competitor is Apple Intelligence. Just listen to how they describe the utility of Live, they write. Gemini can help with tasks big and small by integrating with all the Google apps and tools you use today. And unlike other assistants, it does so without having to jump between apps and services. We're launching new extensions in the coming weeks, including Keep tasks, utilities and expanded features on YouTube Music. Let's say you're hosting a dinner party. Have Gemini dig out that lasagna recipe Jenny sent you in your Gmail. Ask it to add the ingredients to your shopping list in keep. And since your guests are your college friends, ask Gemini to make a playlist of songs that remind me of the late 90s. Without needing too many details, Gemini gets the gist of what you want and delivers. This is exactly the sort of day to day functionality that Apple Intelligence is promising. What makes that interesting to me is actually not so much the competition. Sure, if I was putting on my investor hat and thinking about how AI features are going to impact the bottom line of Apple versus Google, there's a debate to be had there. But to me what this suggests is that these features and these types of interactions are going to be totally ubiquitous. It's going to be very hard for anyone to have a leg up on AI for very long, but companies also aren't going to be able to ignore it. It's simply going to be table stakes features because these things are so useful that they're going to become ubiquitous incredibly quickly. So far the initial reviews are fairly positive. Obviously there hasn't been much time for people to get their hands on this, but broadly what I've seen is that it works a lot better than things like Siri, but it still has issues with hallucination. I'll do an update in a couple of days once people have had more of a chance to get their hands on it. Lastly, although Google was very self conscious to focus on things that were actually available right now with this presentation, having heard it seems the critique of them announcing things that are still fairly far away, they did show off some developments coming with Project Astra. Astra is another layer on the Gemini AI Assistant that will allow Gemini to see the world through the phone's camera and quote, act agentically on your behalf with reasoning, planning and memory capabilities. In other words, this idea of AI agents, while still very nascent, is very clearly top of mind for these companies and coming down the pipeline. For now though, that is going to do it for today's AI Daily brief. Until next time, peace.