Comment by ndarray
19 hours ago
Do you want to know how many identical questions ChatGPT gets asked every week, recomputing its answer each time because people stopped sharing results? I don't, but I'm sure it far exceeds 3 billion. "How to center a div", "how to install this and that package", "what is this pop culture reference", "who is this famous person",...
I ask a question to ChatGPT and i get an answer.
I ask a question to google and i spend 10-100 times longer visiting multiple websites making multiple servers generate webpages wasting electricity and time reading them finding my answer, trying different solutions
Sounds like a complex question. On average you'll visit like 1.5 sites. Those SQL queries are peanuts compared to an LLM burning through tokens. Besides, for hard questions, the LLM itself might download 30-60 websites to find the info on your behalf, e.g. checking the state of RAM prices.
OpenAI does cache the most frequent questions like this.
There is no way to cache the answer to questions that differ by even one token reliably without introducing potential problems.
there absolutely are ways to cache common-routes to fields of commonly requested paths.
(hint : it's not all regex and checksums.)
2 replies →
Doubtful. Considering features like memory, response style, custom system instructions, etc. can heavy alter a users output.
Imagine if they used the stackoverflow model of "closed as dupe"