Major language models (LLMS) showed impressive functionality in several activities and are very increasing in real world applications. However, especially in senior institutional arrangements, it is important to know that the llm may not be unfaithful. It depends on how the answer is trustworthy, the system can choose to follow the question on another expert, or otherwise go back to safe safe behavior. In this project, we learn the quality of the llm that will be reliably to their answers, and how the vision of confidence can be translated into the river. We raise meditation for errors based on the Ill confidence (Def.
- ** Work done while in Apple
- 4 Rice University
