Terra Incognita
Artificial intelligence

Can we understand what happens inside a neural network?

Serious leads

Why it matters

Understanding what happens inside neural networks could give greater assurance about how AI systems will behave, and could shed light on scientific questions about the nature of intelligence.

What we know

We can isolate internal concepts and circuits in large language models.

What blocks us

Models are huge, and each neuron mixes several concepts.

Current leads

Feature dictionaries, circuit tracing, automated analysis tools.

Sub-questions

Related questions

Sources

Not yet reviewed by a specialist.

Last checked: 2026-11-18

History of this question

Back to the map →