Google Gemini podcast mentions

Quote

support to talk about Agor, but I encourage people to just go see the websites at Agor And then you'll get a sense very quickly visually for what it looks like. like but each card is a git war tree on the board and each card can have one or many AI sessions inside of it. Currently when you add you click the little plus button to create a new AI session inside your work tree. You have the choice between plug Code, codex, Gemini, or opencode.ai, which is an open source one. that has opens up you know like 70 different models. So here you're in a fire environment where you can use all these AIs or all these agentic coding tools. With the variety of models that they expose and you can put some AI to work. Organize the the work spatially so you really know what's happening there. That's one of the interesting aspects where you're talking about having these templated Prompts is that in my own experience of building these systems, I'm developing my own best practices for how I like to prompt these systems, how I like to structure I would like to structure the guidance, but that is very difficult to then prompt. propagate and popularize throughout the rest of the team. I guess I could write markdown documents that I put And so I'm interested in the case.

Quote

So another a batch of releases, but I think um like on the open weight front, but I think uh that that is like a separate thing where um you We have now companies developing the tooling around L's that is becoming more and more mature. And then you have better LLMs yourself. And I think I would also almost like uh Separate those two. So my hypothesis is if you would take the best um open weight LLM and put it into let's say a Chat GPT or Gemini or or clawed um interface, you would almost get the same type of quality performance and everything. Like where I think a lot of uh use cases evolve around the That's this idea that was popularized. Kind of towards the end of last year on harness engineering? Um so I I think um Um that is also something how we changed using LLMs because Before it was just simply um yeah, like a very simple Chat interface that was a model. Yeah, yeah, yeah. Yeah. And then it became you know, more sophisticated, you could upload files and PDFs. And so for my personal use case I use LMs mostly uh for like actually it sounds weird, but like proof So just the um before

Quote

And you know CIDR, which you know folks have maybe have heard of is our you know our internal ID. And we've had all kinds of you know capabilities and tools there for a while. We certainly have had pretty good tools for a while, but they were for internal use. Yeah. Yeah. When Google started getting into the sort of like LM game. Like basically when Everything rebranded to become Gemini and like certain certain push out Gemini where people were like, Oh like Did you know that Google probably like Google's entire repo is uh probably about the same size? as GitHub. And like I you know, th there must be some interesting data. Oh yeah. I mean uh and that's one of the things, you know, in building a lot of these internal systems is you know the the the data is incredible. Yeah. Especially when it's, you know, not only is the model and the training House but all the data on the usage and whatever so you know we could build really kind of sophisticated Sophisticated things there. Yeah. Okay, so let let's let's introduce people to Jules. Uh on the your Your website says Jules Autonomous Coding Agents. We've seen lots of these. They're not octopuses, they're not purple. So you got that you got that going for you. But but like what what really is like the Yeah, so what we think about And what we set out to do, you know, back when I joined, was like, where is where are coding agents gonn…

Quote

Um yeah, just just tell us a little bit about this project um and and how you Use it. Yeah, so uh ghostwriter agents is not something that I invented um I borrowed the ID, uh not even close owned it. Um it was a project uh that was uh being shown off to the the uh Google developer experts a couple of uh months ago uh where they had created a tool that hooked into Gemini. um but as I was not the Gemini user, I was a copilot user. Um I had to recreate it um in like markdown files. Uh to be able to have flight agents. Um so I started thinking about okay, what if I can have my brain dump uh that I have somewhere in yeah a markdown file in front matter. And the moment I have time, I start writing it. Um so I say okay, here needs to be a screenshot, here needs to be a code snippet, and so on and so on and so on. Um but nowadays uh or what I wanted to get to was

Quote

a boss who eight years ago, ten years ago even, you know one-on-one chats with him, hey Paul, how do I grow here? And his answer it's all about FaceTime. First in Last out. Now nobody's at the office, so nobody sees you doing these things. You know, does does that matter? But um and you know he there is something. Something to that he now runs AI responsibility at Gemini and it's a huge Huge job and I I th I believe he got that job. Because he made the right connections and he he was available. He was there, he was a presence and made himself known, you know, and that opportunity. is honestly there for anybody to take. Um but you know that's What separates good from great. Yeah, no, absolutely. And that's fair. Um, what would you say is kind of the message you would have to some of the um solopreneurs or small business owners that that we're talking to today um and how they should kind of look at this year uh'cause I I did it at the beginning of the year like seeing where I can set Simplify life, right? I think you know anyone running their own business uh is aware of the the stressors. That comes with it, no matter how intentional you think you were at first. Is there any advice that you would?

Quote

And uh I'm dialing in from Seychelles. We're both traveling and it's been um about uh a little less than a month since we recorded the last episode. So we have uh quite a bit to mention uh some of these things and I think it shows a lot of progress that's been made. uh in just a few weeks and we're excited to talk about some of these things. Start off, I think uh Gemini has been making a lot of good progress and pushing the boundaries of AI and AGI and they have beaten uh the previous incumbents in Arc AGI, both Arc KTI one and two. So these two tests are a test of intelligence tests that we had before, the Turing test and a couple other um standardized benchmarks. So arc AGI one was the first iteration and uh founded by the ARC Foundation, um, which is, I forget what it stands for. Um but the first iteration was a bunch of logic puzzles. And uh as you can see, I think we're nearing

Quote

the application layer. So we're like way above all these layers trying to div deliver customer value, but Our AI gateway has I've lost count now, it's north of seventy-five models running in it. in production and rovo dev to do a general task will probably use two or three models on most like most turns of the conversation. from different people, a lot of times it's using clawed code as a model, for example. It's using Gemini because our job is actually to pick the best model for the first time. for the best task based on the data rather than just kind of giving you a a task. Um, I'll tell you another thing that's maybe interesting that I think is underappreciated, and again. I run an organization with a massive amount of technology. A massive amount of what would be deemed legacy technology. already have written over the last twenty years. Maybe it was written a year ago, maybe it's written five years ago. It's running it's in production, it's great. When I make a change, I need to understand all of that in some way. In terms of the change. We have architectural ways of solving this. We come up with microservices. So we have lots of services that contact each other to do all sorts of things, identity, logging, you know, media storage, whatever, awesome. It's good cloud platform. Now I'll make a code change. Uh it's very hard for tools.

Quote

That's much closer to an agent. It really is an agent at that point. than it ever has and like these reasoning models where it gives itself a what type of information it should return. That's closer to an agentic workflow. I would almost argue any chat bot you're interfacing with, like ChatGPT, Anthropic. Claude, Gemini, those are agents now. I don't think they're just chatbots anymore. Interesting. And I and as far as the risk, I the risk is much higher for agents because it's being thrown at people. Everyone's saying These agents can save us a ton of money and do a bunch of work for us. Let's experiment get this going fast and get it out in the wild so we can start saving that that that money and getting this free work as as done as fast as possible. I think a lot of the risk is coming from speed of delivery. There are definitely companies. We've had it sure sounds like that's getting pushed to the side right now. Security is We saw this with web app. applications and even today there's that issue security hasn't had time to breathe and catch up compliance hasn't

Quote

uh has basically abandoned Lama. What is already there will remain open source. what kind of updates, who knows, if there are any. But they have turned to a cult closed source model family now. It's called Muse Spark, and that's where all of their effort will be going forward. So essentially that kind of puts it back into the same space. uh closed source space that we see open AI and anthropic and uh the Gemini models from Google, uh among others. And so the question And so uh people right now are looking around at what they can do in the open source context. And you know, going back to all this physical AI and things that that we were talking about and How will that affect uh a lot of the things people want to do there? And right now, uh it's Interesting'cause I I will finish by without diving into it. In the industry I'm in, um there is at least uh a perceived national. security interest in having Western created models that are open source. out there um and that are kind of preferred over uh over specifically models from China. Though at this point, if you look at leading uh open source uh AI models, China is definitely taking the lead uh in that space.

Quote

Testing that he and the Andon team had done with both Opus four point seven and GPT five point five. Fascinatingly, and in a definite narrative, Andon reports that while Opus four point seven still makes more money in its vending machine simulation, it does so in part by adopting ruthless tactics. Which GPT five point five does not. Lucas describes GPT five point five as clean. We also hear a bit about their experience opening a new Gemini Run Cafe. Our third guest is another returning champion, Zv. It was a bit too early for Zv to render judgment on 5.5, but we did get a job. Get into quite a bit of detail on four point seven, including how he understands the bad behavior reported. by Andon Labs, and also what he makes of Anthropic's recent model welfare reports. Including why we should care, how much we should trust the model's self reports, and what we're doing. at least on a precautionary basis. Then finally we have Naveen Verma. Princeton Professor of Electrical Engineering and Co Founder and CEO of En Charge AI. I, a company that's developing a new computing paradigm that uses in memory analog data. Which

Quote

Ghost model has of course guinea is starting to get a bad rep and Sam Altman is going and calling them out. for their fear based marketing. They also signed a huge enterprise deal at Novo Nordisk, the Ozempic company we'll be talking about. And I think one of the biggest stories is that M Amazon is pouring over twenty-five billion dollars into Anthropic on top. Now if you're still paying for Chat GPT, Claude, Gemini, Grok, any of the audio models, 11 labs for audio. and any of the image models, I've got to tell you about AI Box. This is what I personally have built It's what I'm recommending to my friends who ask me how to actually use AI without going broke on the system. subscriptions, you get access to over 80 different AI models in one place. All of the top models so you can pick whichever one is best for the task that you are doing. The part that I think is super useful is our automation builder that we have just created you describe what you want in plain English and it builds out a workflow for you. You don't need to know how to code. You don't need to how know how to, you know, wrestle with uh with a new platform you describe what you want and it builds it for you so it's eight ninety nine a month and I hope this is something that's saves you a ton of money and is incredibly useful for getting your hands on all of the different AI models to use. them. All right, let's…

Quote

Now let's talk about AI. So we talked about technology a little bit in general. You could have had students doing programming and working with robots for many years now. But it's only probably very recently that they can be interacting with with LLMs. And tell us about that journey. When you first proposed some of the topics that we could be covering in this episode, at that time You were using Claude as the kind of the key LLM internally, but now there's been a switch to Gemini. And I for our listeners to understand as the AI landscape continues to evolve. How does one make a decision for what kind of LLM or what kind of system or what kind of framework they should be deploying for kids to work with? So that's a great question. I think we think about this all the time. When we launched our website, we have a lot of people. It was the spring of twenty four and we were thinking. rate that we were we knew would impact the outcomes, right? We want to accelerate it. And so we had a problem to be solved and with thinking about a AI and you know we we we were just at that little that early adoption adoption, you know, that Rogers adoption scale, we live by it. When we take on initiative.