SCIENCE UNDER AUTOMATION
LABBench2 v2
Nearly 1,900 tasks with more realistic biology information work; substantial difficulty increase.
What this result supports
Task versions and modalities differ; not physical experiment execution.
Evidence record: preprint. Checked: 2026-10-20.
Where this evidence is used
Experimental molecular biology
- Choose a discriminating experiment: Laboratory reasoning tasks support planning, not physical execution.
- Prepare samples and troubleshoot experiments: General physical autonomy is not evaluated by LABBench2.
- Validate a causal mechanism: The cited evaluations do not establish autonomous mechanism validation.