GOLLuM Jointly Trains LLMs with Gaussian Processes to Curb Hallucinations in Science

GOLLuM jointly trains language models with Gaussian processes, letting LLMs learn from uncertain experimental results to address overconfident hallucinations. It shows strong results across synthesis, materials, and molecular design tasks.

2026-08-29 ~ 2026-08-30 · 2 related posts

• GOLLuM Fixes LLM Overconfidence in Science via Uncertainty Training — bravo_abad · 2026-08-29

1 near-duplicate retellings: DrDatta_AIIMS