A 0.62-AUC classifier helped discover 6 new altermagnets: AI's job is making one loop step cheap
bravo_abad · x · 2026-09-20
A roundup of recent AI-for-Science papers argues models don't need to be accurate — just cheap at one step of the scientific loop. Highlights: a classifier trained on 350 structures (6 positive, AUC 0.62) narrowed 65,578 linker candidates to 145, yielding 6 new altermagnets; an AI never touches transition-state calcs in catalysis but input prep alone shifted barriers by 2.5 eV; one paper turns published papers into agent-runnable tools; another assigns an AI agent to each of 37,075 clinical trials in a virtual biotech, doing 76 days of work in 6 hours.
More from Research
- 4 LLM families rank the Top 500 open problems in math via 34,890 pairwise judgments — burny_tech · 2026-09-20
- Paper2Agent turns any research paper into an AI agent that answers and collaborates — burny_tech · 2026-09-20
- XGEN debuts Generative World Simulation: JING model tops WBench Full split — hey_abusiddik · 2026-09-20
- Metropolis–Hastings explained: the MCMC engine behind Bayesian inference in ML — burny_tech · 2026-09-20
- Researchers Propose Hard Cap of 10 Papers Per Author as Submissions Explode — RexDouglass · 2026-09-20
- The 5-step recipe to train your own frontier model: open data, one arch, 1k GPUs — andrew_n_carr · 2026-09-20