Search papers, labs, and topics across Lattice.
Affiliation:
1
0
3
Long-horizon agents can now achieve GPT-6 Astra and Claude Opus-level performance on complex tasks by learning from dense, progressive rewards derived from latent world specifications.
