# Study tests how AI weighs scientific evidence

> Researchers introduce a 96,000-case biology benchmark and report that calibration training improved models’ forecasts and experiment selection.

Oossa · 2026-11-02 · https://oossa.com/en/study-tests-how-ai-weighs-scientific-evidence

Researchers introduced EDB-Bio, a benchmark built from published studies in three biological fields, to test whether language models can judge evidence for scientific hypotheses. In the researchers’ tests, untrained models were systematically underconfident and scored about as well as a simple base-rate forecast; training them to give better-calibrated probabilities improved their results.

The researchers say their Ibis0 model selected experiments 2.5 times more efficiently than the best frontier model in a sequential experiment-selection task. They released the code, data and trained models.

## The facts

- EDB-Bio contains 96,000 episodes drawn from transcriptomics, CRISPR screens and human genome-wide association studies.
- The paper was posted on arXiv on 2026-10-29; the listing gives its publication date as November 2, 2026.

## Why it matters

For scientists using AI to help choose experiments, the results suggest that probability-calibration training may improve experiment selection, though the reported advantage is limited to the study’s task and has not been independently validated here.

## Sources & references

1. [Do Language Models Know What Compelling Scientific Evidence Looks Like?](https://arxiv.org/abs/2610.41015) – arXiv cs.LG, 2026-11-02

Last updated: 2026-11-02