Hacker News Reader

Comment by crazygringo | Hacker News Reader

The AI answers provide tons of source links.

At the end of the day, is it really all that different to provide a list of links, versus an answer or overview of a few paragraphs with links to lots of different higher-quality sources?

I follow those source links all the time. Not just to "check sources" but because they provide a ton more detail. And the links are usually much better than what I'll get with regular keyword search results.

> It’s about monopolizing access to information.

Not as long as there are competitors like OpenAI and Anthropic. In fact, LLM's have provided Google with stronger competition than it's ever had before. ChatGPT and Claude are doing what Bing was never able to.

View on HN

> I follow those source links all the time.

The vast majority of people don’t.

We’ve gone from Only links to the source -> Mostly links to the source, with a short summary picked almost verbatim from the source -> AI summary that mangles several sources’ information together and gets top billing -> Only the AI summary with some footnotes linking to the source.

Google has been fairly slowly been turning up the temperature of the pot and we’re only a few degrees away from a full boil. Let’s not pretend or be naive enough to think that’s not what’s happening.

Ask any publisher and you will get a resounding "yes, it is very different." On average they're able to attribute about a 33% decrease (globally) in traffic to google's (or others') AI answers. [1]

You're right that there are competitors, but those competitors are doing the same thing: hoovering up content and then not giving anything back for it. There are deals in place for some of the largest publishers [2] [3], but that leaves a ton of content out in the cold. That's going to decrease the amount of content that's out there, which will decrease the quality of AI search. I don't know where that ends, but given how leveraged the economy is in AI it seems like a good idea for somebody to figure it out.

[1] https://pressgazette.co.uk/media-audience-and-business-data/...

[2] https://futureweek.com/a-complete-list-of-publishers-strikin...

[3] https://digiday.com/media/a-timeline-of-the-major-deals-betw...

Most people don't look at the sources even though the sources often contradict the statements.

I've stopped using Google and find I'm not missing anything

> The AI answers provide tons of source links.

A lot of the time, the answer itself is good, but the links are spam blogs and Tiktok videos. I don't think there's a real connection between how the text is generated and what "references" are picked for it. I just searched for a math history topic and the reference was a literal TikTok video that's an advertisement for a sketchy mobile calculator app?

So yeah, these references are boosting web content, but it has nothing to do with the high-quality sources used to train the LLMs in the first place.

> is it really all that different to provide a list of links

Probably not, but I don't like change.

Reply on news.ycombinator.com