[jraph]

You don't need a car to go to your neighbor.

But it's actually worse than that: while both the car and the horse will correctly move you from A to B, you can't cite an LLM output as a source and can't use it to check something. You can use it as a source of inspiration at best.

Here your LLM output is wrong. Someone from Germany wrote they most commonly use dots, in France we use slashes and I think countries around all do one of the other.

So yes, there's no way around doing it like we always did: find a reputable source to back a statement. And ChatGPT does not do this.

Better use a horse than a roulette.

[hu3]

Roulette? It's often correct enough, if you know how to ask, so more like a superhuman than a roulette.

If ChatGPT behaves like a roulette to you I'll attribute that to operator error.

[jraph]

When evidence is needed, "often correct enough" is not sufficient. You need reliable and correct for sure.

I assume you understand how LLMs work? Probabilistic word generation? That can't possibly be used as a source. It literally works by making up things that look probable. It can be used to find a source, but you still need a source.

The LLM might be correct often enough, but you'll only know the LLM is correct in the case at hand by checking at a source…

> If ChatGPT behaves like a roulette to you I'll attribute that to operator error.

I never used it. I know LLMs can be impressive and useful, but sourcing is not among the use cases. The operator should be operating the right tool for the job in the first place.

Shit I say is hopefully correct often enough, but I still can't use myself as a source.

[hu3]

> When evidence is needed...

That's a big "when". Often it is not. Real life is often more maleable.

For similar reasons I don't need aeronautical engineering to fix my kids bike. Good enough is good enough.

> I never used it.

You should try. If at least to have a more educated opinion on the matter.

[jraph]

> That's a big "when". Often it is not

I'm not implying frequency, and it does apply here, you cited ChatGPT as an evidence, in this very discussion. I'm speaking about this, specifically. This comment where you do this was downvoted and flagged to death. And I did mention LLM can be useful. What are you trying to do here?