Welcome to your Friday Intelligence Digest – the AI news espresso shot that saves you hours of scrolling so you can prepare for the AI news week ahead!

We’ve got model upgrades, corporate food fights, government intrigue, desktop AI invasions, and yes… a shoe company turning into an AI stock rocket. Buckle up (or should we say, boot up).

— Jay Chrawnna

Kicking things off, Anthropic dropped Claude Opus 4.7, delivering a marginal boost over the uber-popular Opus 4.6 - Though not quite the homerun the previous release was.

Reports of greater token consumption have been met with rate-limit increases for subscription plans. But Opus 4.7 is still nowhere near the Mythos model we’ve been buzzing about.

A version of that “scary-powerful” AI is heading to US government agencies, expanding beyond the limited tech and finance firms previously whitelisted in the Glasswing cyber-defense program.

CEO Dario Amodei is scheduled for a White House visit to untangle last month’s safety concerns with the Pentagon around AI in warfare.

And, with much backlash on X, some advanced Anthropic tools now require government ID verification.

In a leaked memo, OpenAI roasted Anthropic for “meager computer investments,” allegedly inflating revenue numbers, and building a brand on fear and restriction.

To match the Mythos hype, they unveiled their own cyber-defense program - available to all highest-tier customers instead of just whitelisted big tech and gov (though also with requiring gov ID).

On-trend of narrow AI, they released Rosalind a model focused and tuned for scientific research!

And speaking of female names: the original gender divide among early ChatGPT adopters (determined by usernames) has not only closed, it’s flipped since the beginning of this year!

But all these users need to generate revenue. A Starbucks integration for ordering drinks in-app is now being tested, and ChatGPT’s ad pilot program is introducing pay-per-click ads like those from Meta and Google.

Tesla finalized the design for its AI5 chip, destined for Optimus and supercomputer clusters.

While previously intended to upgrade Tesla’s upcoming Full Self Driving robotaxi fleet, it turns out the AI4 chip can do 10x better than distracted human drivers on its own.

Musk expects AI5 to become one of the most-produced AI chips ever once volume production hits in 2027 - packing 5× the useful compute of today’s dual A14 self-driving chips.

Drama on the Dwarkesh Podcast this week as the host pressed Nvidia CEO, Jensen Huang, on his strategy of opensource collaboration between the USA & China.

As the worlds largest AI chip maker stuck to its guns releasing 2 opensource models: Ising for stabilizing quantum compute and Lyra 2.0 for generative open worlds (think video games that generate as you walk around!)

But China wasn’t far behind!

E-commerce & fintech juggernaut Alibaba followed with HappyOyster which does the same. This came on the heels of HappyHorse, an AI video generation model that has taken the crown from fellow Chinese model Seedance 2.0!

While China lacks access to the advanced compute chips enjoyed in The West (like those made by Nvidia), their chip production capacity and political willingness to build datacentres at scale is closing the gap quickly.

Meanwhile here are all the US states legislating pauses on Datacentre builds…

To catch the news as it happens be sure to follow us on your favourite platform!

YouTube - X - Instagram - TikTok