Search papers, labs, and topics across Lattice.
1
0
2
Sparse interventions can activate complex task behaviors in language models by targeting just 0.01% of neurons, revealing hidden nonlinearities in model dynamics.
