1. In the Same Week, OpenAI Made Its Biology Model Smarter and Published a Plan to Contain Biology OpenAI shipped two posts this week. One teaches its models more biology. The other asks how to stop biology from being turned against people.

2. He Pulled the rsync Git Log to Settle Whether Claude Broke It Before stating a single conclusion about rsync, the developer behind a blog at alexispurslane.github.io spent his opening paragraphs on something else: explaining how he built his report.

3. Where the Answer Came From Now Matters More Than Whether It's Right Four papers posted this week share no authors, benchmarks, or methods. They converge on one move anyway: stop scoring agents by their final answer and start finding where the work went wrong.

In Brief

• Google will pay SpaceX $920M a month for compute Google signed a deal to buy compute from SpaceX at $920 million monthly. A Google representative tied the spend to demand for its recently launched AI products outpacing internal capacity.

• AirTrunk commits $30B to 5GW of Indian AI data centers AirTrunk, the Australian data center operator, will spend $30 billion building 5GW of capacity in India. The plan ranks among the largest single-country data center bets announced this year.

• New York passes one-year ban on new large data centers The New York legislature passed a one-year moratorium on new large data centers, the first statewide ban of its kind. It awaits Governor Kathy Hochul's signature, and sponsors say the pause buys time to study energy-price and environmental effects.

• Musk's SpaceX IPO positions him to become a trillionaire The Verge's Decoder examined how Musk is structuring the SpaceX IPO and index-fund inclusion to push his net worth past $1 trillion. Reporter Ryan Mac, coauthor of Character Limit, walked through the financial mechanics.

• Mira Murati starts speaking publicly again Murati, who left OpenAI to found her own startup, has begun making selective public appearances after a long quiet period. The move follows a stretch where staying silent stopped paying off for visibility.

• Anthropic launches Services Track and Partner Hub Anthropic added a Services Track and Partner Hub to its Claude Partner Network. The program lets consulting and integration firms register as certified partners for Claude deployments.

• Mathematicians warn AI is closing the gap on their field Science reported that mathematicians are raising alarms as AI systems solve problems faster than expected. The piece documents researchers reassessing how much of their work machines can now reproduce.

• Quilty's script-scoring tool draws skepticism from users Quilty promised it could predict a film's box-office success by reading the script. Industry testers who tried the product found its predictions unreliable, even with extensive data.

• Verge calls on platforms to add AI-content filters The Verge argued that YouTube, Instagram, and TikTok already label AI-generated images, video, and music but refuse to let users filter that content out. The piece presses platforms to add an opt-out toggle.

• Founders raise money for anti-phone, in-person startups While AI fundraising sets records, some founders are building products that pull people away from screens. Mirror founder Brynn Putnam raised for Board, focused on in-person games, and DIY cyberdeck makers are gaining traction.

• Code2LoRA generates per-repository adapters for code models Researchers introduced Code2LoRA, a hypernetwork that produces repository-specific LoRA adapters without adding inference-time tokens. It targets the cost and brittleness of per-repository fine-tuning as codebases change.

Read the full edition →