Search papers, labs, and topics across Lattice.
2
0
1
17
Traditional benchmarks miss 82% of LLM performance, revealing a vast underestimation of their true capabilities in diverse tasks.
Despite advances in agentic systems, they still lag far behind human performance in complex tasks, achieving only 19.1% success on a new benchmark designed to probe their limitations.
