Los Alamos and NVIDIA release 250-question ICF benchmark for LLM scientific reasoning
geoffwolfe · x · 2026-11-02
Los Alamos National Laboratory and NVIDIA published a paper using inertial confinement fusion (ICF) as a test case for language models in science. They built an expert-curated benchmark of 250 questions spanning qualitative reasoning, numerical calculation, symbolic derivation, and equation formulation.
Responses are graded with a combination of deterministic numerical/symbolic checks and LLM judges, going beyond endpoint-only evaluation.
The poster's key takeaway: a correct number can come from wrong physics — training and eval signals should test assumptions, units, and reasoning, not just the final answer. ReasonCore offers 5,000 CritPt-style tasks based on this idea.
More from Research
- BioCompute opens its remotely controlled wet lab, pitching 'Lovable for biology' — vaibhavbetter · 2026-11-04
- Legora CTO hand-writes 100k-param model weights with Fable, one-shot beats training — dair_ai · 2026-11-04
- 20-40% of alignment data is noisy: a loss-reduction scoring framework to fix it — burkov · 2026-11-04
- Scientists image quantum entangled photons in real time, and the result looks like a taijitu — CurieuxExplorer · 2026-11-04
- kalomaze: crude diffusion autoencoders may lose low-frequency info at the latent stage — kalomaze · 2026-11-04
- Structured captions boost text-to-music alignment, new Music DSG metric beats CLAP — affige_yang · 2026-11-04