Research
Where yesterday’s theories meet tomorrow’s machines
What defines us is a hypothesis: that human sciences contain solutions machines haven't discovered, and machine intelligence offers laboratories human sciences never imagined. No boundaries on which direction insights flow. No assumptions about what's useful. Just experimenting whether centuries of human thought can debug tomorrow's machines, and whether tomorrow's machines can test yesterday's theories.
Workshop
NeurIPS 2024
System 1.5
We gave machines the cognitive gear shift experts have — knowing when to think fast versus slow.
Workshop
NeurIPS 2025
Monitor-Generate-Verify
We translated metacognition into mathematics. AI now assesses strategy before attempting solutions.
Workshop
COLM 2025
Before You 〈think〉, monitor
When AI pauses to assess difficulty first, it solves 7% more problems correctly.
Workshop
NeurIPS 2024
In Defence of Post-hoc Explainability
Philosophy says imperfect AI explanations can still advance knowledge. We proved it.
Our mission is ensuring researchers have the resources to succeed.
Request assistance
