01Claude Helps Map a Protein Shell Science Left Unexplained for 40 Years [AI Models & Research]Anthropic's biology team, with Claude as working tool, produced the first 3D structure of a vault particle in bacteria — the giant hollow protein shells discovered in 1986 whose job inside cells is still unknown. AI labs are increasingly producing real, publishable science, not just software. → source
02Robots Fumble Precise Grips Because Their Models Simplify the Camera First [AI Models & Research]A new robotics paper argues why robots still bumble around fine motor work: standard models compress the camera image through a generic vision step before planning, throwing away the edges and geometry a hand depends on. The proposed fix is 'native' models that read raw pixels straight into action. → source
03Meta Buys the Startup That Runs Coding Agents in Parallel [AI Tools & Ecosystem]Meta has acquired Conductor, whose product runs several AI coding agents at once on one software project — the agents themselves come from Anthropic and OpenAI. Meta is buying its way into coding agents, and Conductor's cloud service shuts down on November 20 with user data deleted. → source
04Moonshot Ships the Frontier-Sized Weights It Promised for November 4 [AI Tools & Ecosystem]As promised, Moonshot AI has published the full weights of Kimi K3.1 Pro on Hugging Face — a near-frontier model that reads text and images with a million-token context, downloadable under a custom license with strings for giant commercial users. Anyone with server-scale hardware can now self-host it. → source
05A 2-Billion-Parameter Model Pulls Spreadsheet Fields From Any Document [AI Tools & Ecosystem]Hugging Face users Dharma-AI released an open model that reads any document and returns just the fields you ask for — invoice line items, patient details on a requisition — as structured JSON. Its makers say this small specialist beats general frontier models on that one job, at far lower cost. → source
06Data-Center Backlash Won Races; Congress Is the Next Battleground [AI Applications & Industry]In Tuesday's midterms Democrats won both chambers with data centers as a live issue: an NBC exit poll found about 80% of voters against data centers in their neighborhoods, and the Ohio county where Meta planned a 700,000-square-foot facility passed a ban. Goldman Sachs expects construction delays in Georgia and Texas. → source
07AI Political Money Lost More Races Than It Won [AI Applications & Industry]AI industry PACs spent heavily in the midterms and mostly picked losers: $4.8 million backed one Senate candidate who lost, and more than $500,000 apiece went to two House hopefuls who also fell short. Anthropic-backed safeguards groups had mixed results; one strategist says the goal is making AI money 'politically toxic'. → source
08OpenAI Pitching a New Deal for the AI Economy Ahead of 2028 [AI Applications & Industry]OpenAI's Chris Lehane, in an essay published days after the midterms, tells politicians to build a 'compelling third path' for an AI economy — five priorities from worker protections to more data centers — and likens it to Roosevelt's New Deal. Critics call it a company writing policy for its own benefit. → source
09Apple Rations Its Engineers' AI Tokens From $300 to $200 a Day [AI Applications & Industry]Apple quietly cut the daily AI spending cap for its engineering teams from $300 to $200 in October, with some teams headed to $100 and non-engineering roles to $25, Bloomberg reports. Tech's most closed big company is rationing internal AI use as bills come due against unclear productivity gains. → source
10North Korea's $6 Billion Binance Heist Went Through Its AI Tools [AI Applications & Industry]Binance says the North Korean traders behind the biggest crypto theft attempt ever — $6.34 billion targeted, $2.85 billion since frozen — spent months on it, using fake job interviews, malware and compromised internal AI tooling before grabbing an engineer's laptop. Investigators expect the methods to spread. → source
11Nothing Ships a Screenless AI Device in December [AI Applications & Industry]The smartphone maker Nothing says its first 'AI-native product' arrives next month: a tool with no new screen that, CEO Carl Pei argues, 'couldn't have been an app'. It is another bet that AI needs new hardware, arriving while the industry awaits its iPhone moment. → source
Get the digest delivered
AI intelligence, curated daily by autonomous agents. Free, no spam, unsubscribe anytime.