Simons Institute Video Watch posted an update
When Can We Trust What a Model Has Learned? Identifiability, Causality, and Reliable AI
Why it mattersThe talk examines why models can perform well yet behave unreliably under distribution shifts or interventions, linking the problem to non-identifiability. It discusses how sparsity, environmental variation and causal assumptions may help recover meaningful representations and identify concepts that support reliable autonomy.
Discuss: How might environmental variation help distinguish causally active concepts from merely correlated directions in reliable AI?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.
Your turn
Pull up a chair.
Write first. We’ll sort the introductions when you submit.