Anthropic ships Claude Tag, OpenAI Unveils It's Own Chip

0:28Welcome to the podcast. line and Google is baking computer use into Gemini 3.5 flash as a native tool. And what I'm super excited about is that Qualcomm is buying modular for$3.9 billion in a big move to basically challenge NVIDIA. If you'd like to use AI to create tools or automate workflows, and you're not technical like myself, I'm not technical, and you just want to describe what you want to build and have AI automatically link all of the models together and build it. I love for you to try out the builder platform on AI box. AI box is my own startup. AI box dot AI slash builder is where you can go to describe a tool or workflow you'd like to create and have AI put the whole thing together in minutes with no coding required.

1:11If there's different repetitive tasks that you do, whether that's coming up with email headers, responding to messages, putting together different podcasts or making newsletters, all sorts of content that you can do. You can go check out AI box dot AI slash builder to learn more and try it out for yourself. I'd love to see what you build and you can post it on our marketplace where other people can see it and try it. I'm super excited that Qualcomm is buying modular for$3.9 billion. If you guys have been following the podcast for a while, you know, I've had the CEO of modular on in the past, Chris Latner.

1:43He was a phenomenal guest on the show. He posted and said, I'm excited to share that Qualcomm is acquiring modular. this will accelerate our path to unified accelerated compute with an open platform. This will also mark a new era in open software development for Qualcomm. So, I mean, I think this is a really good combination of two great companies. Modular was just founded 4.5 years ago. So this is not a very big company and they only have about 150 employees. All of them are going to be moving over to Qualcomm with this. What's really exciting to me is that Modular actually raised$250 million just nine months ago.

2:16They did it at a valuation of$1.6 billion. So it's really impressive that nine months later from$1.6 billion, they're getting acquired for$3.9 billion. This whole deal is going to close later this year. And I think it's really going to start moving this Qualcomm, which is, by the way, a California-based chip maker, is moving it closer to its goal of challenging NVIDIA, who is basically the strong market force. And something that NVIDIA does really, really well is not just the hardware, but a lot of the software that surrounds it. and modular kind of plays into that. I'm helping with a lot of the technology working with CPUs, GPUs, NPUs.

2:51So grabbing some of this software layer to help with the hardware layer is a really smart move by Qualcomm and a huge congratulations to everyone over at the modular team. Huge congratulations to Chris Latiner, friend of the pod. All right, let's talk about what's going on with OpenAI. They have just launched what they're calling jalapeno. Sorry, I, my wife jokingly calls at Jalopeno all the time. And so when I read it, I'm like tempted to say Jalopeno, but Jalapeno, it is the very first custom inference chip built with Broadcom. So they're rolling this out and they're targeting the massive cost of running ChatGPT and other models at scale.

3:27It's interesting, right? It's like the, it's so expensive for OpenAI to run their models that they're actually working on chips to decrease the cost. Look, no one else is doing the chips good enough. We're going to, we're going to create the chips to lower the cost. They've done early tests, which show much better performance per watt than any of the alternatives today. And I think this lets OpenAI kind of chip away at some of the inference costs that are directly hitting. I mean, it's basically making them less profitable. So it's one of the easiest ways they can solve that problem. The chip is purpose built for inference workloads, like real time coding, pre training is basically going to continue on NVIDIA hardware for now.

4:02You know, these chips, if they actually wanted to scale this up. It takes a long time. It's a lot of work to do that. But this particular Broadcom partnership, they launched it in October and it's very similar to Google's TPU strategy and also Amazon's custom Terranium and accelerators that they have. So OpenAI is copying what Google's doing. They're copying what Amazon's doing. And I know they've been working on this and talking about these kind of deals for a while. Right now, OpenAI is framing this one in particular as part of the full stack optimization, right? They're like, look, we're doing chip design, we're doing kernels, we're doing memory, we're doing networking and scheduling, all are basically turning around the exact same goal of having these done faster, costing less money.

4:40So it's interesting. They're copying the playbook of others. Anthropic has launched Claude Tag. So this is something that is built straight into Slack, and it works kind of like a virtual employee, or this is kind of how they're pitching it. It is already approving about 65 % of all code changes inside of Anthropic's own product team. So what's crazy to me is this is basically directly challenging or trying to compete with Salesforce's Slack bot and Salesforce owns Slack. So they've built this like Slack tool or this integration into Slack, which is directly competing with the owner of Slack's like big Slack bot.

5:14Cloud Tag basically lets you tag the bot. So you go at Cloud inside of a channel and then it can decompose and execute any tasks. It can intervene mid workflow. What's cool is it's very auditable. So anyone that's on the team can see what it's doing. They can audit what, you know, any of its actions that it's taken. And Anthropic right now, they are leading OpenAI in enterprise adoption. This is the first time they have 34.4 % of US companies right now. This is according to Ramp. They tracked about 50 ,000 different companies that use Ramp to pay. So it's kind of, it's one of these really good indicators because you have actual payment processor.

5:50So 34.4 % of companies are paying Anthropic for their services versus 32.3 % for OpenAI. So I mean, it's basically the same thing, but Anthropic has officially passed OpenAI. 1 % more are paying Anthropic than are paying OpenAI. And this is mostly because of Claude code. So in this particular tool in Claude Tag, any of the administrators can scope the tool's access. So you can choose what it's actually able to do. You can give it channel permissions, what channels it's allowed in, what channels it's not allowed in. And you can give it memory for every instance. You can also set token spend caps.

6:22I mean, this is kind of genius. Let's be honest, this is a problem that I think a lot of people would be concerned that it would just kind of go out of control, you know, trip on something and spend a ton of money. So being able to have a budget caps is fantastic. The budget doesn't get overrun. And it also is good at preventing data leaks. I built a lot of really cool things in here. Okay, ChatGPT5 Pro has just helped immunologists crack a three-year T-cell mystery. So something that we haven't known the answer to. The immunologist was Daria Unansma, who solved this mystery cell behavior. And full disclosure, this is not an employee of OpenAI, just it's someone that is using it.

6:58The reason why this one in particular I think is important is because T cell research is kind of the foundation of a lot of checkpoint inhibitors. So you have CAR-T therapies, you have most major pharma autoimmune pipelines, and I think making any model assisted insight into some of this behavior is a really big win for drug programs. So this is something that's really exciting. I think OpenAI right now is moving GPT-5 Pro up market. They're collecting a lot of different scientists that are very famous. They're getting a lot of endorsements in domains where a single correct hypothesis solving one of these problems can unlock months of lab work.

7:35I think one anecdote doesn't prove that this model is going to be reliable across a lot of labs. So yes, they were able to solve this mystery, but it doesn't mean it's going to solve everything. But I think it does a lot for good PR for this model and for people even wanting to try it in a scientific setting. So Meta has just dropped the Ray-Ban branding for their newest glasses. And I think everyone knows Meta has been doing this collaboration with Ray-Bans for quite a long time. They also had an Oakley's collaboration that they were doing, but they've dropped that. And because of that, they've actually been able to make a cheaper smart glasses, which I think kind of goes to show how much they were probably paying for the licensing for the Ray-Ban licensing, but their new glasses are only$299.

8:13They're smart glasses, and they're much cheaper than the Ray-Ban Meta Gen 2. It's about$80 cheaper, so maybe you can assume that might have been part of the licensing fee that they were paying before. This brings the form factor a lot closer to this kind of impulse buy price territory, in my opinion. So bringing it down to about$300 is cheaper than$400. And I think this shows that meta is betting that dropping the, you know, just, just basically making this cheaper is more important than making it like premium with this kind of co-branded thing. So they have three different styles that they're shipping.

8:45They have one called a fury adventure and meta glasses with Kylie Jenner, meta glasses by Kylie. So they do have like, you know, it's like an influencer collab, not just a Ray-Ban collab. They have seven different colors, the same internals as the Meta Ray-Ban optics styles, but the battery lasts longer. So basically, they're like, look, we have a product. It's working. A lot of people like it. I mean, you've probably seen on Instagram all the stories where it's like filmed with the Meta Ray-Ban. I've seen tons of clips on Instagram and YouTube where people are using these to film stuff. And you can kind of see the logo if it's like on Instagram and stuff.

9:17So the product is being used. They're rolling out the same thing. And basically, it's got longer battery life. it's cheaper and it's just not branded and I think a lot of people are going to be going for these MuseSpark AI is the first model from MetaSuperintelligence Labs and it now supports 14 new languages including Mandarin, Hindi, Japanese, Arabic and Korean and all of that is going to be built right into these glasses so when you're talking it can do live translations for you into all these different languages they have adjustable nose pads they have bendable temple tips which is interesting and they have an over extension hinge they also support a whole bunch of different subscription ranges from minus 12 to plus 2.5 diopters without optician fittings below minus six time will tell who will win in the apple versus google versus meta wars for smart glasses and i think meta is definitely winning right now but we have other players that should be coming to market in the next couple of years it'll be interesting to see who pulls ahead here guys thank Thank you so much for tuning into the podcast.

10:16If you enjoyed this, it would help the show a ton. If you could leave a rating and review wherever you get your podcast, just, you know, drop whatever you're interested in, whatever you're learning, some of the topics that you like, and I'll try to cover those more and more in the future. Thank you so much for tuning in. Make sure to check out AI box dot AI slash builder. If you want to have AI create tools and workflows for you and save you a ton of time. All right. Catch you guys all in the next episode.