|
|
|
|
|
|
by krapp
835 days ago
|
|
|
To be fair, most people using search aren't looking for fun, random and unrelated content, they want results specifically relevant to the terms they're searching for even if they mistype their query. The big problem with LLM and search is that the results themselves are often poisoned by factually incorrect LLM generated content being promoted from sites Google presents as sources of truth (see the recent "can you melt eggs?" debacle[0].) [0]https://arstechnica.com/information-technology/2023/09/can-y... |
|
|
You're not wrong here, and I don't think LLM has made this much better - for a reason in addition to yours.
In my experience pre-LLM there was a good blend of both relevant and unrelated even when a few letters were off, or a word was slightly wrong.
And now more often than not, the LLM they must have parsing the search query guesses my intention wrongly, making the search results ironically less relevant. Pre-LLM In that situation, I'd correct my query to narrow down the results in my next search - as I still do anyway post-LLM.
TL;DR using LLM to 'enhance' search queries hasn't actually improved anything, while also making my day less fun.