Gemma 3 Extracts Unseen Chemistry Protocols Without Fine-Tuning, Scoring 0.89
bravo_abad · x · 2026-10-29
Costa and colleagues gave Gemma 3 (12B) explicit definitions of experimental actions plus expert-designed rules to fix recurring errors, enabling protocol extraction with no fine-tuning. On 110 held-out zeolite synthesis protocols, the pipeline scored 0.89 action-sequence similarity versus 0.70 for a catalyst-specific baseline whose vocabulary lacked sieving and crystallization entirely. Benchmark charts show it dominates on zeolite synthesis but trails organic-synthesis-specific LMs on organic chemistry, highlighting the domain-adaptable "editable vocabulary" approach.
More from Research
- Weaviate Podcast: MIT PhD's "Machine Studying" lets always-on agents prep without knowing the task — CShorten30 · 2026-10-29
- The singularity's week: 80-year-old cells reset, 25-year-old math disproved by AI labs — rand_longevity · 2026-10-29
- Single-nucleus RNA-seq atlas reveals distinct features of early-onset colorectal cancer in the young — EricTopol · 2026-10-29
- Stanford's ChromBPNet Reads DNA Regulatory Switches Base by Base — RichardsonDx · 2026-10-29
- Determinism in Training Costs Single-Digit Percent, Not 10x — gandamu_ml · 2026-10-29
- Neuron steering struggles at persona creation, failing multi-turn stability — davidad · 2026-10-29