10 ms·
I'm kind of late to this post, but really awesome initiative and well executed project! Thank you for bringing this to people! I tested it a bit and it seems p
by Valk3_ 2y ago
I'm kind of late to this post, but really awesome initiative and well executed project! Thank you for bringing this to people!
I tested it a bit and it seems pretty decent, although for some really niche theoretical questions it wasn't successful in retrieving the answers I wanted even if alot of the results were really good in other aspects. It could simply be because the answer is not available anywhere in hackernews.
I'm wondering if someone were to build a similar project but for other sites, what would your advice be? For instance what technical difficulties did you stumble on that you think would be good to be aware of?
Thanks in advance and once again congratulations on the project!
- jnnnthnn 2y agoThanks for the kind words! Yes, the underlying dataset very much conditions the quality of the responses. Additionally, the retrieval strategy is also a really important factor (and that is something which I haven't had time to extensively optimize). I'm writing a blog post that will answer your questions! Will post it here when it's fully baked.
- Valk3_ 2y agoAwesome, looking forward to it!
- jnnnthnn 2y agoBlog post is now live: - https://news.ycombinator.com/item?id=40442039 https://news.ycombinator.com/item?id=40442039 - https://jnnnthnn.com/how-to-build-your-own-perplexity-for-any-dataset.html https://jnnnthnn.com/how-to-build-your-own-perplexity-for-an...