What the field is saying

The new Chai-1 benchmarks are more honest than anything I've seen from a foundation model team this year. Actual error bars. Actual held-out set. More of this, please.

Arvinas hitting Phase 3 with ARV-471 is one of those moments where a whole modality crosses a threshold. Next decade of oncology chemistry just shifted.

A lot of people are going to retrofit their slide decks this quarter to include AF3. Most of them shouldn't. Structure is cheap now; downstream decisions still require judgement.

Quiet quarter internally. Loud quarter externally. Two Phase 2s reading out, REC-2282 data looking clean. Team’s built the muscle of turning platform signal into program decisions.

Running Boltz-2 against the PoseBusters rejects for giggles. Getting ~14% recovery that v0.3 missed. Confidence calibration really does work.

Not surprised developability is the wall. We've spent 20 years optimising natural antibodies; asking a model to rediscover that in one gradient pass is a tall order. Joint objectives are the path.

Q1 AI-drug discovery deal value up 22% YoY despite count down 12%. Concentration in PROTAC + ADC. Full breakdown in this week's State of AI.

Clinical signals are catching up to the AI hype. Arvinas PROTAC, Recursion phenomics, Isomorphic partnership momentum – finally seeing molecules, not slides.