Hacker News Reader

Comment by rstuart4133 | Hacker News Reader

I'd go further, and say I use search when I'm pretty confident I know the right search terms. If I don't, I'll type some wordy long explanation of what I want into an LLM and hope for the best.

The reason is pretty simple. If the result you want is in the first few search hits, it's always better. Your query is shorter so there is less typing, the search engine is always faster, the results are far better because you side step the LLM hallucinating as it regurgitates the results it remembers on the page your would have read if you searched.

If you aren't confident of the search times, it can take 1/2 an hour of dicking around with different terms, clicking though a couple of pages of search results for each set of term, until you finally figure out the lingo to use. Figuring out what you are really after from that wordy description is the inner magic of LLM's.

View on HN

If the result you want is in the first few search hits, it's always better. Your query is shorter so there is less typing, the search engine is always faster, the results are far better because you side step the LLM hallucinating as it regurgitates the results it remembers on the page your would have read if you searched

Most often not true in the kind of searches I do. Say, I search for how to do something in the Linux terminal (not just the command, but the specific options to achieve a certain thing). Google will often take me to pages that do have the answer, but are full of ads and fluff, and I have to browse through several options until I find the ones I want. ChatGPT just gives me the answer.

And with any halfway decent model, hallucination only seems to be a problem in difficult or very specialized questions. Which I agree shouldn't be asked to LLMs (or not without verifying sources, at least). But over 90% of what I search aren't difficult or specialized questions, they're just things I have forgotten, or things that are easy but I don't know just because they're not in my area of expertise. For example as a learner of Chinese, I often ask it to explain sentences to me (translate the sentence, the individual words, and explain what a given word is doing in the sentence) and for that kind of thing it's basically flawless, there's no reason why it would hallucinate as such questions are trivial for a model having tons of Chinese text in its training set.

It depends. Sometimes webpages are useful, but at the same time navigating the amount of fluff on webpages nowadays takes longer than asking a LLM.

I asked Claude to give me a recipe that uses mushrooms and freezes well and it give me a decent looking soup recipe. It might not be the best soup ever, but it's soup, kinda hard to mess up. The alternative would be to get a recipe from the web with a couple dozen paragraphs about how this is the bestest soup ever and it comes from their grandma and reminds them of summer and whatnot.

> I asked Claude to give me a recipe that uses mushrooms and freezes well and it give me a decent looking soup recipe.

It didn't suggest adding glue? I imagine it would freeze real well if you did that. /s

> I'll type some wordy long explanation of what I want into an LLM and hope for the best.

Interesting, I just random words. LLM not care sentence.

Reply on news.ycombinator.com