Search papers, labs, and topics across Lattice.
2
0
4
3
LLMs working as specialized teams dramatically outperform solo LLMs on complex, multi-step reasoning tasks, suggesting a new paradigm for LLM collaboration.
You can get surprisingly reliable multi-turn LLM agents by having a small, open-source critic supervise a powerful, fixed LLM actor *during* the conversation.
