Up to 58% of DFT-generated data is physically impossible, Valence Labs warns
josephdviviano · x · 2026-09-23
Valence Labs published an op-ed noting that physics-based simulation tools like DFT and FEP are being repurposed for dataset development: once manually run and reviewed by experts, they now produce tens of millions of synthetic data points without close human oversight. The catch: studies show that in some computational datasets, as much as 58% of DFT-generated data is not physically possible. The author Ihor describes several cheap and effective methods for avoiding subtle failure modes in physics simulation, cautioning against blind use of these methods at scale.
More from Research
- iSDFT: open-source self-distillation method enables continual learning for LLMs — hbouammar · 2026-09-23
- The Polynomial Freiman-Ruzsa Theorem Leaves an Open Algorithmic Question — gautamcgoel · 2026-09-23
- Researcher Argues Parallel Agent Swarms Are a Weak Path to RSI — gleech · 2026-09-23
- New paper proves the long-standing Courtade–Kumar conjecture with multibit extensions — abeirami · 2026-09-23
- TimePre Paper Lands in TMLR: Reversible Normalization Fixes MCL Instability in Forecasting — _vztu · 2026-09-23
- Reproducible agent evals: harbor makes configs, trajectories and logs shareable — seanwbren · 2026-09-23