DeepSeek's latest flagship model arrived with no announcement page — just an OpenRouter listing. No word yet on open weights, which is the detail that actually matters to the ecosystem.
A new paper reveals that Anthropic, OpenAI, and Google return encrypted chain-of-thought blocks that can be replayed across sessions and users. The thinking you're not supposed to see is leaking at the infrastructure level.
Amazon quietly flipped a switch to use Twitch streams as AI training data. Creators have an opt-out — buried, naturally — but the default is extraction.
AI-generated uploads have overwhelmed the platform numerically, but purchase data tells a different story. The market is sending a clear signal: volume without quality is worthless.
Mirendil cofounders — ex-Google and Anthropic researchers — are building systems where AI meaningfully contributes to its own development. Self-accelerating AI is moving from thought experiment to funded startup.
Pickup artist influencers are using Ray-Ban smart glasses to covertly record women and build content around it. The research draws a direct line between covert wearable recording and documented harm.
Miles Deutscher called out one of AI's most underrated friction points: no shared memory across tools. You use Claude for code, ChatGPT for writing, and neither knows what the other knows. As agent workflows mature, this fragmentation becomes the core UX problem nobody has solved yet.
Ethan Mollick flagged updated research from Brynjolfsson, Chandar, and Chen — "Canaries in the Coal Mine?" — suggesting AI's economic impact is still being systematically underestimated. Mollick's been consistent here: the models that say diminishing returns are imminent are probably wrong, and this paper adds empirical weight to that view.
Logan Kilpatrick posted the Gemini numbers: 1 billion monthly active users on the Gemini app, 1 billion downloads of the Gemma model family. These aren't projections — they're the kind of distribution figures that make every other AI lab's user stats look modest by comparison.
Garry Tan framed AI as a multiplier on the individual, not just a productivity tool. He also predicted a wave of older AI-native founders — people with decades of domain expertise who can now build without a large engineering team. It's a direct challenge to the "only young founders move fast" orthodoxy.