Search papers, labs, and topics across Lattice.
POSTECH
4
0
5
Forgetting can be finely controlled to minimize knowledge loss while maximizing utility, a breakthrough that enhances unlearning in LLMs.
Turn-level credit assignment can boost jailbreak success rates to over 98%, revealing the limitations of traditional trajectory-based methods.
Harness-aware post-training can drastically improve LLM agent performance, especially when facing shifting tool environments, revealing a key design dimension often overlooked in AI systems.
Guaranteeing uncertainty quantification in dynamic environments is now possible even when feedback is strategically withheld by an adversary.
