Hacker News Reader

Comment by bagacrap | Hacker News Reader

Why were you using a web search if you didn't want to access the web (find a website)?

Openai says don't use chatgpt for search. Microsoft says double check everything Sydney tells you. Is there an argument for LLMs replacing search besides laziness?

View on HN

If this works as we wish(no saying that is posible), this will reduce the amount of seo spam trash you read to 0%, will allow to investigate things faster, and will allow to have more recursive way to expore topics. And this will be key to make ok google more similar at what we wish as ai companion

It seems inevitable that LLMs will eventually work for search. They just don't work yet.

It really does not seem inevitable at all if you know how the tech works under the hood. The trust problem can't be solved with the current approach.

So will the LLMs be able to stay current with 24 hour news and updates? And will users be able to get a link to any sources if they ask so they can verify or go the website as desired? Will those websites still have a reason to exist if they're just being used to feed the search LLMs?

All difficult questions that will need to be answered in the affirmative if the web is going to survive. The only well-funded player that is really incentivized to make all these things happen is Google, as literally every other megacorp wants you to be stuck in their walled gardens instead of searching the web.

Google introduced Google Answers so you stay on Google instead of going to the websites they index. They have AMP so news articles can be hosted on Google. They have email, video, about five different video chats that they kill off, Translate, etc. Google's walled garden is a short fence with widely spaced planks, but it's still got barriers.

I wish! It's not that I don't want to have a free personal assistant that can answer all my questions with zero effort of my own. And it's not like it couldn't happen, but I don't think it's inevitable or right around the corner. These are large language models, not large knowledge models. They're good at sounding like humans, not good at determining veracity. So to become consistently accurate would be a quantitative change and not just a qualitative one.

It's true that this is an unsolved research problem, but as an ML researcher, I expect that some combination of better integration with existing information retrieval systems, a moderate amount of high-quality data, and some small but important tricks will solve it. I don't know whether the time horizon is six months, one year, or five years, but I'd be surprised if it takes longer than five. GPT-2, which was arguably the first LM that could sound human-like, is only four years old. Given the value of a trustworthy LLM, an enormous amount of research effort and resources will be directed toward this problem, and I don't think it's substantially harder than other challenges the ML community has solved over the last five years.

> These are large language models, not large knowledge models. They're good at sounding like humans, not good at determining veracity.

That sounds pretty much the same as a human.

And some humans are not even very good at sounding like humans!

Reply on news.ycombinator.com