AI Interpretability Frontier
Reading the internal structure of learned systems.
What trained models represent, how they compute it, and whether the methods for finding out actually work — including on brains.
Drag to pan · Ctrl + scroll to zoom
Questions being explored
Expeditions in progress
Signals at the edge
- Emerging regionInterpretability × systems neuroscienceConfidence 0.28 · low
- Field convergenceLearned surrogates in physical scienceConfidence 0.35 · medium
- Active research frontierWhat fundamentally limits transformer scaling?Confidence 0.40 · medium
0 charted findings and 1 bridge candidate touch this field. See what is charted →
Why this field matters
We build systems whose internal workings nobody designed. Being able to read them would change how they are evaluated, corrected and trusted, and would give neuroscience a testbed with full observability.
How your support is allocated
Backers who chose a focus are honoured first; the rest follows these weights (demonstration values), then each target’s expected usefulness. Every batch is recorded with its reasons — most recently A-862, < $0.01 routed.
- Active expeditions33%
- Replication17%
- Frontier signals15%
- Open questions12%
- Scouting8%
Popularity chooses attention. Evidence chooses knowledge.