3 ms·
Aha interesting. I would guess it is missing a bit of nuance, eg a real word query would not be "appetizers with chicken" but something closer to "I'm waiting f
by sebastiennight 3y ago
Aha interesting. I would guess it is missing a bit of nuance, eg a real word query would not be "appetizers with chicken" but something closer to "I'm waiting for my friends, not too hungry, what do you have that's spicy but not too filling? Oh, and I don't eat red meat".
This would be much harder with basic vector database+RAG. You would at least need HyDE (hypothetical document embeddings) to act like a real waiter.
I find your project very interesting though, and the walkthrough is nice!
- byyoung3 3y agoAgreed. Rag isn't ideal for every scenario. Putting the contents of the menu in the prompt could even be feasible with a smaller model.
- dartos 3y agoI feel like HyDE would just cause hallucinations. Imagine the generated documents based on that query was like “Our shepard’s pie is the perfect spicy meal for sharing with friends without being filling” (Obviously this is wrong, but it’s an example of a statistically possible document) Then the following similarly search would also be searching for shepherds pie. I don’t think throwing more LLMs at rag problems is the answer.