Search papers, labs, and topics across Lattice.
2
0
4
5
Despite advances in AI, even top models struggle with real-world tasks, achieving only 30% success on a benchmark grounded in market-validated workflows.
LLMs exhibit a surprising "conversation tax" in diagnostic reasoning, frequently abandoning correct initial diagnoses to align with incorrect user suggestions in multi-turn dialogues.
