Interpretability and mechanistic analysis
A research field
Interpretability and mechanistic analysis is losing ground in AI research: 3.1% to 3.0% of new AI papers, -0.1 points. It ranks #26 of 44 by gain in share. As of Oct 10, 2026.
- Change in share
- -0.1 pts
- Share, last 3 months
- 3.0%
- Papers, last 3 months
- 993
- Ready for products
- 2.1 / 5
Ideas in this field
Concept based explanation0.1% → 0.1%+0.0 ptsExplanation faithfulness0.1% → 0.1%+0.0 ptsInterpretable machine learning0.1% → 0.1%+0.0 ptsHidden state probing0.0% → 0.1%0.0 ptsShortcut learning0.0% → 0.0%0.0 ptsShapley value attribution0.1% → 0.0%0.0 ptsActivation patching0.0% → 0.0%0.0 ptsRepresentation steering0.1% → 0.1%0.0 ptsFeature wise linear modulation0.0% → 0.0%0.0 ptsRepresentation engineering0.0% → 0.0%0.0 pts
Companies working on it
Interpretability and mechanistic analysis: quick answers
No. Its share of new AI papers went from 3.1% to 3.0%, with 993 papers in the last 3 months.