HackerTrans
TopNewTrendsCommentsPastAskShowJobs

str1der23

no profile record

comments

str1der23
·5 個月前·discuss
How do you evaluate the synthesis quality? Like if I ask about chunking strategies and it recommends method A over method B for my use case — how do you know that recommendation is actually good? Have you run any benchmarks on recommendation accuracy, or is this more of a "trust the retrieval + LLM reasoning" setup?