no code implementations • 12 Jan 2024 • Shangqing Xu, Chao Zhang
In each step, it analyzes a pool of candidate examples and identifies the ones most likely to challenge the LLM's current understanding, measured by a new metric called misconfidence.
In-Context Learning