Search papers, labs, and topics across Lattice.
1
0
3
LLMs hit a "reasoning collapse" where accuracy plummets by over 50% on classical reasoning tasks once complexity exceeds a surprisingly low threshold, even with deterministic validation.
