I think it would be interesting to share the system prompt and also think about if it makes sense to rewrite the question in a style of the article, so less a question, more a statement that could appear in an article, and only use the question at the end again to generate the final answer?
Your video hit at the right time when I was thinking of rewriting the query. I notice Perplexity using query rewriting to get the best results on search display, and was wondering if I should implement this on my LLM projects. Thanks for sharing the info, and now I know I should. Btw, the metrics hit_rate seems interesting as a way to measure the successful output versus input.
Could be worth playing around with the prompt a bit more to see how it does. e.g. maybe ask it to evaluate 5 different rewrites and tell you the best one. Only thing will be to make sure it doesn't take too long on the query rewriting step.
Mark, you are about to push the quality of your content beyond a point where I start to feel incompetent. Hence, I am hesitant to say "keep on going", but I'll say one more time "keep on going" 🤭
I think it would be interesting to share the system prompt and also think about if it makes sense to rewrite the question in a style of the article, so less a question, more a statement that could appear in an article, and only use the question at the end again to generate the final answer?
Your video hit at the right time when I was thinking of rewriting the query. I notice Perplexity using query rewriting to get the best results on search display, and was wondering if I should implement this on my LLM projects. Thanks for sharing the info, and now I know I should. Btw, the metrics hit_rate seems interesting as a way to measure the successful output versus input.
Could be worth playing around with the prompt a bit more to see how it does. e.g. maybe ask it to evaluate 5 different rewrites and tell you the best one. Only thing will be to make sure it doesn't take too long on the query rewriting step.
Mark, you are about to push the quality of your content beyond a point where I start to feel incompetent. Hence, I am hesitant to say "keep on going", but I'll say one more time "keep on going" 🤭