Welcome to the podcast.
We have an update on OpenAI.
They're pivoting away from Sora.
They're shutting it down.
But apparently their pivot is going into robotics.
So they're not just shutting down Sora.
They're actually putting it somewhere else.
And we also have Apple that is making a huge move with Syrian iOS 27.
That affects basically everything Every iPhone user ever.
That I'm actually kind of excited about and maybe turns Apple into a winner in an area where I thought they were only a loser in AI.
The biggest story is that there is a data leak at Anthropic and it revealed a secret model called Claude Mythos.
Anthropic's own internal documents describe it as a quote-unquote step change in capabilities and they're saying that it poses an unprecedented cybersecurity risk.
This is coming from the safety company, so we're going to unpack all of that on the podcast today.
But first really quick.
If you're someone that is trying to keep up with all of these different AI models Claude GPT, Gemini Grok We'll be right back.
Basically, you describe the tool you want and we'll chain together and we'll add prompts and build the whole thing for you with no coding.
I am not a developer, so I made this mostly for myself and now for everyone.
So if you want to set up a content pipeline or a research workflow or basically anything where you're doing the same thing over and over again, You can automate that on AI Box.
It's $8.99 a month and it's linked in the description.
So go check it out.
And yeah, let's get into the first story.
The first thing I want to talk about is AI robots at the White House.
This is kind of honestly just sort of like a funny story.
A lot of people talked about it.
I think there's like a headline on TechCrunch, which said like Melania Trump wants...
AI robots to homeschool your kids.
Anyways, people are kind of just being funny with it.
But basically what happened was Melania Trump brought in the figure three humanoid robot and it was basically walking around on two feet.
It was greeting guests.
It was speaking in 11 different languages.
Now I think on the surface you can really look at this like a PR moment for figure three and actually probably in the White House right to say like look, we're like super high tech.
But I think the reason why it matters is it's a signal of how fast physical AI is moving.
If you look at a year ago, we were seeing these robots in these kind of controlled lab demos.
And now we have one walking through the White House.
I think this is a really big jump in a very short amount of time.
And I think it connects a really big trend we see everywhere, which is just this week Agile Robots announced a partnership with Google DeepMind.
They're going to integrate Gemini models into physical robots for manufacturing.
They're also doing automotive and logistics.
So you're seeing Google, OpenAI and a bunch of other companies all converging on this idea that physical AI, which is, you know, robots that can actually do things in the real world.
This is the next frontier.
And I think we're going to be talking about this a lot more in the coming months.
And I would say by the end of the year we're going to see some pretty, pretty sizable rollouts of companies rolling these out in a big way.
The second thing I want to talk about is SoftBank's $40 billion OpenAI investment.
So they're putting together this big round for OpenAI.
I think this is obviously a massive number, but that's almost secondary to what it represents when the industry is really headed in an interesting direction.
I think.
For me, what it's showing is there is a barrier to entry for building these kind of top line AI models.
And it is this barrier to entry is very high.
We have a lot of amazing companies I mean myself included taking these tools and building cool things with them.
But if you want to be a frontier model company, the stakes and the barrier to entry is insane.
And it kind of honestly makes me feel bad for some of these companies, like Mistral AI, which are just smaller and regional areas like France.
And yes, they're raising billions of dollars, but like Are they able to raise $40 billion?
Are they able to have the entire arsenal of Google behind them?
It's pretty hard to compete in these ways.
I think it's not just about having the most talented research team anymore.
You have to have billions of dollars in compute in infrastructure and you need to have the ability to scale your distribution globally.
At the same time,
So I think a lot of these companies are... are really pulling further and further ahead.
Like OpenAI, like the top hyperscalers, they're getting way further than anyone else just because they had the lead at the beginning.
So SoftBank has basically made OpenAI a core pillar of their entire AI investment thesis.
And when you look at the landscape OpenAI and throw up at Google, the gap between all of these front labs and everyone else is getting.
So I think for the average person using these tools, it's actually a good thing.
More investment means that more compute is going to make the models better.
But from a competition standpoint I think it raises a lot of questions about how concentrated this industry is going to get.
Right.
It's becoming a very expensive game to play.
And I mean, even if you think like 40 billion dollars.
Let's say there was, you know, eight different AI companies and SoftBank gave each of them, or 10 different AI companies and SoftBank gave each of them 4 billion.
You could build some cool stuff, right?
But like, would that be able to compete with Anthropic, OpenAI, Google?
And that's kind of the question.
So it's really, really wild, just the numbers that go into all of this.
And unfortunately, I think it really makes it so there's not a lot of competition in the market.
All right.
The third thing I want to talk about is opening eyes robotics pivot.
So we already talked about robots walking around the White House.
Recently, I gave an episode about OpenAI shutting down Sora, their video model.
This is a bit of an update.
The new detail is that the compute that they're basically turning off for Sora.
So it was kind of very computationally intensive to run that video model.
So they're shutting that down and they're actually going to be giving that directly to robotics research.
You can see this figure robot obviously is a big deal.
A lot of, uh, Optimus robots coming out of Tesla and everything there.
This is a big deal.
And so opening, I think is really focusing on maybe if they don't own the robotics company, right?
Like Elon owns Tesla with the Optimus robots and he owns Grok.
So obviously he's going to put Grok into those.
If they can't have that kind of uh distribution where they own the robot company, maybe they need to start thinking about well you know, if they don't currently own one, maybe they need to start thinking about how they can own one.
So they're not just, you know, making these kind of partnerships where they're working with figure other people to get in, but their AI could be swapped out.
So I think they're putting a lot of these resources into a much into a really interesting kind of field.
I think they looked at AI video generation.
They looked at robotics and basically, as a business decision, they had to pick one and they picked robotics.
So I think right now it's definitely worth paying attention to, because it represents a really big shift in where the smartest people in AI think value is going to be created over the next few years.
It's probably not going to be created in these short form little AI slop videos.
That sounds so terrible because I actually think you could make some great videos and as the models get better, they'll be super useful for video production and a bunch of other things.
But That's sort of what Sora was getting used for, and evidently they want to shift their image to something which is robots, which is going to give them a lot more ROI.
Okay, the next thing I want to talk about, this is Apple's big play.
I'm super stoked.
Apple is planning to open up Siri to third-party AI services. through the App Store in iOS 27.
So up until now, ChatGPT has had an exclusive integration with Apple Intelligence.
That is going away.
Basically, what this means is that you could have Claude or Gemini or Grok or really any other AI model running your Siri for you, as long as the developer builds integration.
So you'd essentially be choosing your AI assistant the same way you choose your default browser on iPhones.
I think this is super cool.
It's interesting, you know, Apple never went down the route of building their own AI model.
I think they just decided maybe it was too hard.
They didn't have the talent.
They didn't want to designate the money there.
And instead they're just relying on other people and they're just kind of building themselves as a shell around some of these other AI companies.
A couple different reasons why I think this is happening.
One, Siri's been losing ground compared to what, you know, Claude and ChatGPT can do.
And I think they know that.
And number two, I think that by opening it up, they are essentially reducing their dependence on a single partner right.
They don't have to bet everything on OpenAI.
They let the market compete and Apple just provides the platform.
And I also think that right now, everyone's basically a lot of people are paying for their own AI subscription.
I would be.
I think it'd be a smart move if you know, instead of having to pay for a huge licensing deal with ChatGPT, whether that goes one way or another, They basically say look, we'll let people pick their own platform.
If you have a pro subscription to OpenAI, maybe you grab your own API key or there's an integration with ChatGPT.
So premium subscribers are going to get premium Siri.
Anyways, it's an interesting thought.
I think for anyone with an iPhone, which is right now over a billion people, This is one of those changes that you'll actually feel the AI assistant on your phone.
I think it's going to get a lot more capable.
Siri's obviously been terrible for forever.
I'm really excited if this solves Apple's kind of AI problem and they can actually start making Siri useful again.
I'd be thrilled.
And I think this is a good move from Apple.
All right, the last story that I want to do a deep dive on that is absolutely wild is coming out of Anthropic, because they have leaked something called Claude Mythos.
Um, it's basically, this is what happened.
There was a configuration error on on Theropics content management system, basically their blog backend.
And that made a bunch of unpublished draft posts publicly accessible.
So you know when, when a company is getting ready to have a big announcement and they want the blog posts published ahead of time, and it kind of set up and ready to go, ready to post wherever they want it.
And they had about 3,000 assets that were never meant to be live.
All of those were buried in the documents.
And there was one of them that was a draft blog post about a model called Claude Mythos.
Anthropic has since confirmed that the model is real.
A spokesperson said that it represents quote a step change in AI performance and it is quote the most capable model we've built to date.
So this isn't speculation.
Anthropic has actually acknowledged that this exists.
I think what makes this really interesting is that this was in those documents.
The leak references a new model tier called Capybara.
So basically for context, right now Anthropic has Haiku Sonnet, Opus and those are kind of like their three from lowest to highest.
Opus is their best model.
Sonnet is kind of their medium and Haiku is their light model.
Capybara is above opus, so it's kind of like the, the ultimate model.
It's even better than better than opus and it's basically an entirely new tier.
It's larger, it's more capable, and then anything that they publicly released and capybara and mythos appear to be basically the same underlying model.
So The benchmarks described in the document are huge compared to Cloud Opus 46, which is already, you know, one of the strongest models available.
It's basically my daily driver personally.
Mythos is apparently scoring even higher on software coding, academic reasoning and cybersecurity tasks.
The thing that I think is really important is that Anthropic's own draft blog post.
This is their own internal writing.
Said that the model poses quote unprecedented cybersecurity risks.
And I think you have to sit with that for a second, right?
This is Anthropic.
This is the company that has built their entire identity around AI safety.
They have, you know, constitutional AI.
They have, you know, responsible scaling policies.
They've positioned themselves.
It's like they go to war with the entire government basically to make their model safe and not used for, like you know, quote unquote unsafe tasks or autonomous AI, all this kind of stuff, right.
And their own documents are flagging this model as a cybersecurity concern.
So I think that is pretty wild.
Something you know, basically how the markets are responding to this.
Bitcoin dropped, software stocks dipped.
I think this is pretty straightforward when you have a model that's dramatically better at cybersecurity.
And the capability cuts both ways could be used to find and also patch vulnerabilities.
So hackers can use it.
And of course, we can use it to counteract hackers, but it's definitely can be exploited.
So I think when the company built it, it is already waving a flag about risks.
And they're like look, we're made this amazing thing, but there's a whole bunch of really big risks with it.
And I think people are starting to take that seriously.
I think, you know, basically what we know and what we don't know is kind of important in this.
This was a leak.
This wasn't a product launch.
So Anthropic has confirmed that they're testing Mythos with early access customers, but we haven't gotten any official releases, the full benchmarks or the kind of the complete safety valuations.
So I think there's still a lot we're waiting on.
But I think that the bigger takeaway here is, you know, what this signals for AI right now.
I mean, they said this is a quote unquote step change, not just like an incremental improvement.
They're describing a, you know, a really qualitative leap.
And I think if that is accurate, if the next generation of models really is that much more capable, then I think we do need to have a serious conversation about what it means for deployment and like applications.
I don't know where this all goes.
Now, I, for one, am actually not an AI doomer in the least.
I'm actually really excited for this model to come out.
I would love to use it to help build my tools and software.
I'm overall very excited, but it is pretty wild that this model exists.
A lot of people said that AI models had kind of hit a plateau and we're just making new tools for them and it's hard for them to get any smarter, but it seems like we have figured out ways to make them smarter.
And yeah, everyone, I'm pretty excited about that.
Okay, guys, that is everything that we have for AI Chat today.
If this was helpful, please leave a review on the podcast.
Over on Apple, you can drop a comment.
On Spotify, you can hit the About section and leave some stars.
It really helps the show out a ton.
And also go check out AIboxai if you want to access over 70 AI models and the ability to build automations and tools without any coding.
The link is in the description.
Thank you so much for tuning in and I will catch you in the next episode.
