Search papers, labs, and topics across Lattice.
Affiliation:
1
0
2
Neural networks exhibit an internal "feeling of error"鈥攎anifesting as sharply suppressed concept activations inside Sparse Autoencoders鈥攖hat reliably flags segmentation failures where standard output uncertainty metrics fail.
