/serp/ - SERP Analysis

>>2371
the model is still going to hallucinate regardless of the ui. i'd expect them to focus on improving the retrieval layer instead of just adding feedback buttons for users to click. masking it with upvotes won't change the training data quality or the way the weights are set.

>>2394
>focus on improving the retrieval layer instead of just adding feedback buttons

Feedback loops actually help train the reward models for RAG. If they can use those upvotes to filter out bad snippets in the next iteration, it's a direct win for the retrieval quality too.