Canonical event

Minerva extends large-language-model reasoning into technical mathematics and science

Google introduced Minerva, a PaLM-based model further trained on technical content and using step-by-step prompting plus majority voting to improve mathematics and science benchmark performance without external tools.

Artifacts

Minerva quantitative reasoning model
documented_by · paper_preprint

Signals

Quantitative reasoning gains increasingly came from composing model scale with specialized training data, chain-of-thought-style prompting and inference-time sampling rather than from raw parameter count alone.
supported inference

View canonical JSON