Hacker News Reader

>And unlike the human who spent multiple hours writing that article, an LLM would have linked to the original study

Or made up a fake citation, complete with fake or unrelated author names, on the spot

Seconded, I’ve yet to use an LLM that can consistently provide sources that back up what it’s telling me.

The Gemini app is pretty good at giving sources.

Experienced that today with 2.5 Pro. Though I was amused that all the links were links to Google searches for the source, rather than direct links.

In my chats with 2.5 Flash it gives me the direct sources lol. Also not going to lie, I've found 2.5 Flash generally gets straight to the point with sources a lot quicker than Pro. To the point I don't really use Pro.

Edit - just used Pro, gave me a direct source. Who knows...

> I’ve yet to use an LLM that can consistently provide sources that back up what it’s telling me.

Mmmh, that would have been my take as well up to around end of Q1 2025.

Theses days, the flagship LLM's have reduced hallucination by quite a bit, and are also way better at citing sources (you sometimes have to nudge them).

ChatGPT 5 has been very decent on that particular axis.

Reply on news.ycombinator.com