Instagram | Sponsor this newsletter

Monday hit hard this week. Meta dropped a new open-weight model before most of us finished our coffee, and a London neighborhood that used to be known for syringes and street gangs is now the place every AI founder wants an office. Small world, weird timeline.

Wispr Flow turns your voice into clean, professional text you can send the moment you stop talking. Not rough transcription you have to clean up. Actual polished text — ready for email, Slack, or any app.

Speak the way you think. Go on tangents. Change your mind mid-sentence. Flow strips the filler, fixes the grammar, and gives you text that reads like you spent five minutes writing it.

89% of messages sent with zero edits. Millions of professionals use Flow daily, including teams at OpenAI, Vercel, and Clay. Works on Mac, Windows, and iPhone.

Start flowing free

Meta just launched Muse Glimmer, a 30 billion parameter open weight model built to run entirely on a single GPU. It is small on purpose, and that is the whole point.

Here is what shipped:

But the model was really just the delivery vehicle for the message. In a 14-page essay titled "The Future is for Everyone," Zuckerberg pushed Washington to loosen training data restrictions on American labs, arguing foreign competitors do not face the same handicaps.

He tied it to a $1 billion community fund and a warning about ceding the open model lane to Chinese labs like Alibaba, DeepSeek, and Moonshot. On Fox Business, panelists framed it the same way, pointing to Virginia data center tax revenue as proof the US needs to move faster or watch the money go elsewhere.

But here’s the asterisk. Meta's stock is still down roughly 10% on the year, and investors have heard this "we're competing with OpenAI and Anthropic" pitch before without much to show for the $145 billion in planned capex behind it. A well-benchmarked small model is genuinely useful for developers who want local agents.

Whether it moves the needle on Meta's superintelligence ambitions, or on federal policy, is a much bigger ask than a Hugging Face release can deliver on its own.

🔮 Prediction: small, efficient, open-weight models become the default onramp for agentic development this year, not the flagship frontier models. Expect more labs to release a "runs on your laptop" model specifically to win developer mindshare while the giant models stay behind closed doors.

Would you actually run an agent locally instead of through the cloud? Let me know what you think.

Read through Meta’s full announcement here →

Twenty years ago, King's Cross in London had syringes in tree trunks and a church occupied by sex workers protesting police raids. Today it is the AI capital of Europe, rivaled only by San Francisco and Beijing.

The numbers tell the story:

DeepMind's 2016 move into the area kicked off the cluster, and now it is home to OpenAI, Meta, Isomorphic Labs, Synthesia, Sierra, Wayve, and Anthropic's new 158,000 square foot office, all within walking distance of each other and a 45-minute train ride from Cambridge talent.

But office space is one thing. Whether the UK is interested in investing further is the real question. The mood in King's Cross is not pure celebration. Investors there told TechCrunch that Anthropic's summer shutdown of Mythos and Fable access, tied to US export controls, was a wake-up call. As Phoenix Court co-founder Robin Klein put it, Europe cannot simply rent its AI capabilities; it needs to build and hold some of its own.

The bigger question hanging over the neighborhood is whether the UK invests in the infrastructure, compute, and energy to make that independence real, or whether King's Cross just stays a glossy outpost of US labs.

🔮 Prediction: watch for more European capitals to try to replicate the King's Cross playbook this year, but talent density alone will not solve the sovereignty problem. The cities that pair a cluster with actual domestic compute capacity will pull ahead of the ones that just have nice office buildings.

Should more AI infrastructure get built outside the US, or does concentration actually make the ecosystem move faster? Reply and let me know where you land.

Higgsfield's Seance 2.5 just dropped, and it animates static images into clean, professional motion graphics with almost no text errors, something older tools consistently botched. Here is the workflow:

💡 Pro tip: for anything more complex than a simple pop-in animation, have Gemini or Claude write you a longer, more detailed motion prompt first instead of writing it yourself. You can also upload up to 50 reference images in a single generation for more elaborate scenes.

Want more prompt workflows like this one dropped into your feed every week? Come hang out in Robo Nuggets, our free community, where we swap the exact prompts we are using before they make it into the newsletter.

🧠 Intel announced a $15 billion stock offering to fund AI and foundry expansion: its first public share sale since it listed in 1971.

📈 TSMC's July revenue jumped 44.7% year over year as AI chip demand keeps outpacing supply.

✈️ DARPA and the US Air Force flew an F-16 under AI control for the first time, with a human pilot onboard only to monitor.

💬 ChatGPT now offers unlimited text chats to free users, removing one of the last major limits on the free tier.

🌐 Cloudflare launched Kitesurf, a browser built specifically for AI agents to navigate and complete tasks on the web.

Would love to hear your thoughts! Send me your thoughts by replying to this email (yes, I read them all :)

Until our next AI rendezvous,

Anthony | Founder of Uncover AI