10 Comments

But how can we correct its mistake like making it legit is there any prompt for that?

Expand full comment

No. Unfortunately: the only solution is for you to check the work. There are partial solutions like: Asking the same question in different ways to see if you get the same answer. Asking the same questions to different LLMs to see if you get the same answer. etc. But none of these is a guarantee.

Expand full comment

It depends, right? I mean if I doing some creative work, I wouldn't care about that. In fact, I wouldn't want it to. But if I am doing a fact check, I would want to tell it to not be "creative".

Expand full comment

Yes, of course. Only in cases where accuracy of the answer is important.

Expand full comment

Problem is perhaps that in your prompt you are not asking it to push back if it's not sure.

Try this

https://share.cleanshot.com/jhXsnBHY

Expand full comment

Thanks Mayank.

In general, I guess this line should be included in all prompts to ChatGPT...

Expand full comment

Now ‘care’ itself is a trait that we will not associate with AI ( atleast the current AIs). So yes it’s bullshitting all the way.

Expand full comment

Well, we know ChatGPT is a habitual liar when it comes to Cinema (among other things). ¯\_(ツ)_/¯

https://twitter.com/_zenman/status/1630835105995554816

Expand full comment

Wow ! I just loved the way GPT lied on the Sadhana song question. So confidently! Sounds like a pathological liar 😃

Expand full comment

It is not a liar. It is a bullshitter. The difference is "The liar cares about the truth and attempts to hide it; the bullshitter doesn't care if what they say is true or false" (Harry Frankfurt).

ChatGPT wants to please you. And it will do anything to do that irrespective of whether it is true or not.

Expand full comment