LLMs cram textbooks, genomic models discover new biology
julia_kiseleva · x · 2026-11-18
Julia Kiseleva addresses a recurring objection: "Why build a model for DNA when LLMs already know biology?"
- LLMs learn what humans have written about biology — papers, databases, variants and pathways — a compressed representation of the scientific record. That makes established knowledge searchable and combinable, which is already transformative.
- But the literature is only a snapshot: most possible mutations have never been tested, most regulatory sequences unmeasured across cell types, most combinations of variants, environments and cellular states never observed.
- For a mutation that appears in no paper — say, how it changes enhancer activity in a specific cell type — the answer exists in no sentence. It must be predicted from DNA itself, which is exactly what genomic foundation models are trained to do.
Core thesis: language models learn what science has written down; genomic models learn what evolution has encoded. Complementary, not substitutes.
More from AGI Musings
- Peking U Paper: RSI Tax Shows Verification Eats 74.8-93.6% of Self-Improvement Cost — RobertTLange · 2026-11-18
- SF AI circle's undercurrent of anxiety: only 1-5 years of intellectual alpha left before AGI — CatAstro_Piyush · 2026-11-18
- The nonhuman economy: AI compute buildout may crowd out land, labor and everything else — thomasahle · 2026-11-18
- Researchers push back on 'agent swarm' takeover premise: real-world 80% AI success is only ~15-minute tasks — rbhar90 · 2026-11-18
- AI weather model gives hurricane targets one extra day of warning, Nature paper shows — davidmanheim · 2026-11-18
- Software singularity must be a sigmoid: compute substrate caps AI takeoff curves — tszzl · 2026-11-18