Hacker News Reader

Comment by riku_iki | Hacker News Reader

Some people don't trust current quality of AI

Yeah, but this is essentially the same as Google results now but instead of just links it uses AI to summarize the results into a text blob. It still contains links to where it got it information in the summary and it allows you to ask follow up questions.

It's not like it generating new information. It just summarizes information based on the same links you would have gotten without it.

One may decide that "summarization" be lower quality, add more errors to results, and pollute results page.

then just click the link and visit the page like you would have without the summarization? It doesn't seem like you've lost anything

Well, the way I see it, you shouldn't really trust any information that's out there on the internet. It's your own responsibility to put search results into perspective, whether it's someone's blog or AI generated content.

Correct, and I have my workflow to reduce noise/signal ratio, for example by adding: site:stackoverflow for programming questions, or by adding site:ycombinator.com for tech queries, and reading discussions how answer was derived.

For AI, I expect noise-signal will be not as good as I expect.

Right, so I want a list of results where I can assess the information from each in the context of the source it came from, its reputation, etc. And other metadata like how old the article is.

It's a non starter to have that critical context stripped out and be given an opaque answer by an AI without having seen the underlying information in the context it originated from.

AI is faceless, a website is not. The other content on the website will provide a clue as to whether the site is trustworthy or not. This is how web search has worked since the beginning.

Sure, but do we trust Google's non generative summaries / search results for some reason? It's all a black box to us on the outside. It's already heavily ML'ed, like at no point was a human ranking these things or determining relevance

It's not the quality of the AI that's at issue. In fact the higher quality Google's AI is, the less I will trust it. Google wants to extract maximum value from me. I want to find useful information on the Internet. These two goals are frequently at odds with each other. The better Google's AI gets, the better it will be at doing it's job (which I do not like).

That is not a very compelling argument.

Why? They tune their algos/metrics to specific metrics.

Main metrics for their business is engagements(how many queries you do on google.com) and Ads clicks, and not how precise answer you got.

Can you be more specific? Just saying it's not compelling doesn't add anything to the conversation

AI as a mechanism for allowing Google (or whoever) to deliver the same information we already have through an inevitable additional opaque layer of bias and advertising would be seen as a horror if so many people weren't impressed by the gee whiz factor of it.

And yet I find it vastly more compelling than your addition.

Reply on news.ycombinator.com