Hi all,
How can an LLM determine when to use its own knowledge versus conducting an external search (e.g., browsing Wikipedia by a tool)?
Would allowing the LLM to assess its certainty in an answer through token probabilities be an effective method? Does anyone have any experience with this?