Search papers, labs, and topics across Lattice.
1
0
3
2
LLMs can autoformalize specs well enough to pass standard tests, but still fail on subtle edge cases 26% of the time, a risk missed by LLM-as-judge evaluations.
