Atomistic foundation models hit data-quality wall: fit error now below DFT label error
matt_slotnick · x · 2026-11-18
Potential Sciences published its first technical lab note arguing the bottleneck in atomistic foundation models has shifted from model fit to data quality: models trained on OMol25 (2025) now sit closer to their DFT labels than those labels sit to the gold standard.
On GMTKN55 neutral reactions, OMol25-era models achieve fit errors of just 0.37x the DFT label error for reaction barriers, 0.66x for intermolecular binding, 0.79x for reaction energies, and 0.26x for conformers — all firmly in the 'data-limited' regime, versus 3–13.8x for SPICE-era models.
Implication: further architecture gains yield little; progress now depends on higher-quality quantum-chemistry label data.
More from Research
- Author clarifies "nothing looks interesting" remark: it's about reviewing, not condescension — rishabh16_ · 2026-11-18
- Researcher laments reviewer fatigue as coding agents flood academia with incremental papers — rishabh16_ · 2026-11-18
- Stanford team predicts viral evolution and designs antibodies that neutralize future escape variants — BrianHie · 2026-11-18
- GPBam: New German Public Law Benchmark Finds Retrieval Gains Limited, Essay Quality Tracks Statutory Knowledge — aaire · 2026-11-18
- AI agents uncover first known genome-wide frameshifting in bacteria — GaryMarcus · 2026-11-18
- Google launches Antigravity Science plugin, gives 10,000 scientists free top-tier access — DynamicWebPaige · 2026-11-18