How to Make LLMs Shut Up
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
Daksh, co-founder of Greptile, shares the journey of improving their AI code review bot to reduce excessive comments. They explore different strategies like prompting, LLM evaluation, and clustering. Initially, attempts to refine comment quality through prompting and severity evaluation failed. However, generating embedding vectors for past comments and utilizing a vector database to filter out less helpful feedback proved successful, significantly improving the quality of the bot's output.
Table of contents
Attempt 1: PromptingAttempt 2: LLM-as-a-judgeOut of IdeasFinal Attempt: ClusteringResultsSort: