Search papers, labs, and topics across Lattice.
3
0
7
4
TRACE transforms user corrections into enforceable rules, slashing preference violations from 100% to as low as 2% in critical coding tasks.
Current LLM-based web agents are vulnerable to prompt-injection attacks, with no reliable defenses against any attack objective, revealing a critical oversight in security evaluations.
Evoflux transforms how compact agents navigate tool workflows, boosting execution success rates from a mere 3% to up to 24% in real-world scenarios.
