Search papers, labs, and topics across Lattice.
Loyola University Chicago
3
0
4
The correctness of TLA$^{+}$ specifications generated by LLMs can vary dramatically鈥攗p to elevenfold鈥攂ased on how we choose to grade them.
LLMs struggle to produce reliable TLA+ specifications, achieving only 8.6% semantic correctness, which underscores the need for expert oversight in automated formal verification.
TLA-Prover triples the success rate of LLM-generated TLA+ specifications, transforming how we ensure correctness in distributed systems verification.