3 ms·
LLMs cannot search, this is just responding with something it may or may not have read. Have you verified these numbers are in fact correct? Have you tried "sea
by verdverm 4y ago
LLMs cannot search, this is just responding with something it may or may not have read. Have you verified these numbers are in fact correct? Have you tried "searching" for anything outside of its training data? ChatGPT will happily hallucinate references. If you have not experienced this yet, you haven't seen the "true" side of LLMs.
There will be plugins for search, but will still need all the crawling and scraping and serving to be available. LLMs will offer an alternative interface to search, but not always the best one. One thing the LLM does not give you is a spectrum of options. How do you know it didn't miss something. This is where having to "troll" through search results is actually worth the effort.
- tus666 4y agoMy understanding is the ChatGPT basically swallowed up the entire internet, although it was a snapshot that might be out of date now. Of course accuracy and correctness are concerns, but that would apply to any kind of summarizer service, and even primary sources can be wrong. Extensive research would always be required to get near 100% confidence of correctness.
- verdverm 4y agoIt only had part of the internet, OpenAI is nowhere near as comprehensive at web scraping as Google, I don't think they actually scraped at all for this, using existing data like CommonCrawl. The other thing you are not understanding is that it did not memorized these things, it built representations for predicting the most likely next token. This is why it hallucinates and makes up numbers and web links or citations that do not exist.