Observed arrival · 2026-10-23
After42: AI looking for questions worth asking
After42 studies whether AI can propose research questions that domain experts find genuinely interesting, developing candidates through attempts to solve them.
- For
- Researchers and scientists assessing AI-generated questions
- Worth noticing
- The project reports that model judges’ rankings do not consistently prioritize the questions experts find genuinely interesting.
Field notes
The workflow starts with a researcher-written subfield goal, generates literature-grounded candidates, and develops selected questions through exploratory solving and revision. The site says the first batch is browsable with expert assessments, spanning mathematics, statistics, theoretical computer science, and physics. It reports that some questions passed expert judgment, while model rankings still fail to consistently prioritize the questions experts find interesting.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue