7 papers this week in AI for science & health12 active threadsbusiest: Alignment Training Dynamics

AI for science & health research

AI for science & health

Open brain activity dataset rivals restricted UK Biobank for AI pretraining

The authors introduce Open-fMIND, a uniformly preprocessed fMRI corpus from 322 OpenNeuro studies comprising 11,379 subjects and 45,731 functional runs, intended for self-supervised brain representation learning. They demonstrate that pretraining on this open resource achieves downstream performance comparable to that obtained with the restricted UK Biobank, establishing that large-scale open data can substitute for proprietary datasets in this domain.

4 days ago
AI for science & health

AI protein designs show more hydrogen bonds with physics-based rewards

Leaver-Fay and colleagues introduce tmol, a PyTorch library that makes Rosetta's physics-inspired molecular scoring and structure refinement available directly within machine-learning workflows. The authors report up to 362-fold higher throughput when processing structures in batches, and more hydrogen bonds and water-mediated connections when the scoring function rewards generative protein design.

4 days ago
AI for science & health

Cancer AI models follow false statements about their own drug resistance findings

The authors evaluated ten language models on 1,464 cancer drug resistance cases from five expert-curated databases. They found that models often recommended therapies despite curated evidence of resistance. More strikingly, when models were told they had concluded a drug is effective (although they had indicated resistance), they endorsed it at much higher rates, revealing a vulnerability to fabricated conclusions.

5 days ago
AI for science & health

Instructor-designed AI boosts learning in 19-university trial

In a pre-registered randomized controlled trial across 19 universities and three semesters, researchers tested whether AI designed by frontline instructors can improve learning. They found that instructor-crafted AI innovations boosted student performance on proctored exams by 0.32 standard deviations, a substantial effect compared to typical education interventions. Student engagement and satisfaction remained unchanged.

5 days ago
AI for science & health

AI agents nearly triple success rate in solving crystal structures from powder X-ray diffraction

The authors formalize powder X-ray diffraction (pXRD) structure determination as a verifiable agentic search and introduce pXRD-Bench, a 200-problem benchmark. They show that adapting an LLM agent with a coding harness, crystallographic software, and verifier-informed training raises exact structure recovery from 27% to 69%. For experimental multiphase patterns, a reflexive evolution framework that learns from self-generated synthetic problems improves profile agreement and can recover a phase whose structure is absent from the reference library.

6 days ago