38 AI-for-Science Papers, Six Lessons: From mRNA Stability to Label-Efficient Neutrino Models
bravo_abad · x · 2026-10-03
This week's Discovery at Scale briefing covers 38 AI-for-science papers and highlights three results:
- mRNA stability: new formulations retain 100% bioactivity after two months at 37°C, with storage stability built into the objective from day one.
- Evaluation pitfalls: a surgical-skill model scores AUROC 0.888 on a random split but only 0.571 when the test surgeons are new — same data, different question.
- Label efficiency: a pretrained neutrino-detector model achieves more with 1,000 labels than a from-scratch counterpart does with 10,000.
The recurring lesson: progress depends on decisions around the model — what to optimize, what to reuse, and what evidence to require before acting on a result. The full briefing includes a four-question checklist and 23 shorter reads organized by field.
Related event: Weekly AI for Science Roundup Distills 38 Papers into Six Research Lessons(2 posts)→
More from Research
- Building a Minimal On-Policy Distillation Setup Focused on Training Engineering — bclavie · 2026-10-03
- Schmidhuber cites three decades of papers on formal theories of creativity and curiosity — SchmidhuberAI · 2026-10-03
- Projection sampling: transforming expert data so SFT learns new skills without forgetting — burkov · 2026-10-03
- Grok 4.7 in Cursor yields numerical candidate for open sphere-inspection problem — xiaosun86 · 2026-10-03
- Pairscore: scoring multiple items at once beats independent ranking for LLM uncertainty — yeewhye · 2026-10-03
- willcb: neurosymbolic world models beat single-rollout RL with value models — willcb · 2026-10-03