Short notes of two kinds: papers (what a paper claims vs. what it shows) and companies (what they sell and whether it holds up).
Recursion bet that massive automated cell imaging plus ML beats hypothesis-driven discovery. The platform is real; the open question is whether it produces differentiated clinical assets faster than the money runs out.
GEARS uses a gene-relationship graph to predict expression changes for unseen single and two-gene perturbations. The headline numbers are strong, but later benchmarks suggest simple baselines close much of the gap.
Nothing matches that filter.