Search papers, labs, and topics across Lattice.
Humalike Research
1
0
2
LLMs struggle to adapt to local social norms in group chats, but explicit prompting can boost performance, with top models reaching over 80% accuracy.