Hacker News Reader

The argument is not that Google's AI summaries are infallible.

The argument is that there's so much garbage on the open web that clicking through to the source has a lower signal-to-noise ratio than the AI summary.

This rings very true in my experience.

Google's AI summaries definitely make mistakes, but in most cases they're higher signal than the source websites.

View on HN

You're underestimating how obviously, laughably wrong the AI summary often is.

I really think Google is making an error using such a lightweight model for the page. That's one of my only firsthand exposures to genAI, and it's awful. Use a decent model, or if that's too expensive don't put up a summary at all.

They used to prioritize accuracy first, then speed second.

Now it's just speed. This is a grave error.

Once it quotes a gov website, exactly the contrary of the official source. Why ? Because on the official website you have checkboxes and the div shown is the one corresponding to your case and most surely the AI summary didn't have information about which one is which and just mix everything up.

Reply on news.ycombinator.com