Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
1
0
2
The assumption that more capable LLMs always benefit from less structured guidance is wrong: a "harness-complexity paradox" shows that overly strict harnesses can actually *hurt* the performance of frontier chat models.