Strange Research
An independent research lab exploring how intelligent systems develop new capabilities.
Current research
Learning from experience
Can AI agents get better at future tasks by reusing what they’ve learned?
We keep a faithful record of what agents do, so that lessons drawn from it stay linked to their evidence. We plan to compare agents that can draw on past experience, their own or another agent’s, with agents that start fresh.
Status Early exploratory runs have informed the setup. Controlled hypothesis tests are planned.