Search papers, labs, and topics across Lattice.
3
0
7
0
Stop waiting for AI agents to mess up: AgentTrust intercepts tool calls *before* execution, offering a chance to block, warn, or fix risky actions in real-time.
Scaffolding LLMs with hints during RL training can boost both initial accuracy *and* long-horizon reasoning performance, but only if the hints mimic student behavior and are gradually withdrawn.
LLM safety guardrails are surprisingly more about parsing JSON than preventing jailbreaks, and general-purpose LLMs often outperform specialized safety models in multi-step tool-use scenarios.
