The Signal Report // AI IN THE LAB
AI in the Lab.
Where machine learning genuinely helps research, and where 'AI discovered X' is doing a lot of the lifting. All reports →
15 in this beat · page 2 of 2
Why AI Weather Models Miss the Records
AI has quietly taken over parts of the weather forecast and matches the physics on an ordinary day. Then a 2026 Science Advances study looked at the record-breakers, the forecasts that matter most, and found the machine consistently lowballing them. A physicist on why a pattern-matcher can't extrapolate past what it's seen.
Read the report → Report 022 · AI in the Lab · 2026-07-12 · 5 minWhy AI Drugs Ace Phase 1, Then Stall
2026 is the first year a real batch of AI-designed drugs reaches late-stage trials. The data splits in two: AI-discovered molecules pass Phase 1 safety at 80-90%, well above the norm, then fall right back to the ordinary ~40% in Phase 2. Which number is which tells you what AI actually fixed, and what it didn't.
Read the report → Report 016 · AI in the Lab · 2026-07-09 · 5 minWhen AI 'Hallucinates' a Protein
In a chatbot, hallucination is a failure. In protein design, it's the literal, published name of a method from David Baker's lab. The catch: of 129 AI-dreamed proteins they actually built, only 27 folded. A lab scientist on what that one-in-five really tells you.
Read the report → Report 009 · AI in the Lab · 2026-07-06 · 5 minDid AI Really Discover Millions of Materials?
Google's AI predicted 2.2 million new crystals. Then working chemists opened the list and found "scant evidence" the compounds were novel, credible, and useful. What "discovered" means when the model never made anything, from a lab scientist.
Read the report → Report 003 · AI in the Lab · 2026-07-04 · 5 minThe Pocket the AI Couldn't See
Several state-of-the-art models nailed a cancer protein's shape and missed a hidden druggable pocket. The lab found it by hand. What that gap says about AI in research, and how to read an "AI discovered X" headline.
Read the report →