Tacit Labs' AI-for-biology thesis: optimize drug discovery, not biology itself
jefrankle · x · 2026-11-04
Tacit Labs' April launch essay laid out a three-point thesis for frontier AI in biology, now revisited with a "call your shots" check:
- If models are to help cure disease, evaluate them on tasks of roughly that shape and complexity — optimize directly on what you care about: discovering and developing a drug
- Model the biotech process, reasoning, and decision-making rather than the underlying biology; biology is an empirical science, so build better empirical methods instead of treating it as an analytical domain
- The hardest part is verifying empirical methods with signal that actually matters — does the drug work in humans? "Biological time" is the most expensive forward pass of all; the key is turning scientific problems into learning problems
More from AGI Musings
- Mathematician slams AI peer review used as a 'Gish gallop' — littmath · 2026-11-04
- New Podcast: Seb Krier on How to Think (and Not Think) About the AI Revolution — sebkrier · 2026-11-04
- Margaret Mitchell: Weight Communication Without Human Decoding Is the Wrong Path — mmitchell_ai · 2026-11-04
- Mitchell: Fixing Weight-Communication Risks Is Counted as "Slowing Down" — mmitchell_ai · 2026-11-04
- Ted Xiao: Robot Progress Is Limited by Verification—One Policy Can Take 100+ Hours to Validate — lukas_m_ziegler · 2026-11-04
- Margaret Mitchell: Advancing weight-based communication without human-readable decoding is the wrong path — mmitchell_ai · 2026-11-04