Search papers, labs, and topics across Lattice.
Stanford University
1
0
2
Skill-switching accuracy in LLMs drops significantly on complex tasks, but a new training approach boosts performance from 34.4% to 68.4% on challenging benchmarks.
