Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
Affiliation:
2
0
3
0
Condensed causal circuits can be up to 316 times smaller than traditional frozen models while preserving task performance, revealing the power of targeted post-training in mechanistic interpretability.
Built-in reasoning can paradoxically improve some instruction-following tasks while degrading others, revealing a nuanced landscape of model performance.