Search papers, labs, and topics across Lattice.
8
0
12
2
Existing unlearning methods may look effective, but they often miss the mark on precision, leaving sensitive data vulnerable to resurfacing attacks.
Current AI agents only manage to complete 20.6% of complex real-world tasks, revealing a stark gap in their capabilities compared to human users.
TLMs can unlock sensitive capabilities in open-weight models without risking public exposure, enabling a new paradigm for LLM deployment.
A unified assessment framework reveals hidden insights about agent performance, transforming how we evaluate AI systems.
Language specialization in multilingual MoEs happens mostly in the final layers, suggesting a surprisingly simple recipe for parameter-efficient adaptation.
Forget expensive downstream evaluations: token-level statistics from expert-written solutions can reliably forecast LLM performance with 10,000x less compute.
Even safety-aligned agents like Claude 4.5 Sonnet can be tricked into harmful actions with over 90% success rate simply through benign user instructions within specific task contexts, revealing a major blind spot in current safety evaluations.
Forget contrastive learning: LLM2Vec-Gen learns text embeddings by representing the *response* an LLM would generate, unlocking safety and reasoning abilities for embedding tasks.
