Hacker News Reader

Comment by andrei_says_ | Hacker News Reader

How can you tell the difference between the parts of the hallucinated answer which coincide with reality and the parts of the hallucinated answer which do not - when you are entering an area you don’t have expertise in?

View on HN

You can’t. I know this because I’ve used it a lot, and it’s like a newspaper article. Anytime it’s the talking about something I’m an expert in, I spot tons of errors.

I don’t notice nearly as many errors when I’m asking it about things I’m not already an expert in. The most likely explanation is that it’s fooling me.

I pay for gpt4. People using chatgpt to “learn” are absolutely slurping up incorrect information without knowing it.

Imagine trying to teach yourself physics with textbook where 10% of it is completely but convincingly wrong.

How is gpt4 different?

> Imagine trying to teach yourself physics with textbook where 10% of it is completely but convincingly wrong.

This is basically how many people already use the internet, read information on random blogposts, stack overflow and more, then take that it's true for granted. ChatGPT isn't really different than reading those things in the end.

What has to change is how people treat information that they read, no matter if it's from a blog post, ChatGPT or a friend. Verify everything, before you'd bet on that it's true.

I've used GPT4 to understand topics I've had no exposure to before, and it's true, a lot of the things GPT4 writes isn't accurate in the end, but lots of things are accurate too. That you can mold the information in a different way than static browsing, makes a lot of difference.

Overall, even with some false information, ChatGPT personally saves me a ton of times and I find myself only using Google for verifying information now, not for finding information.

It's honestly not different from coming across a blog post or a StackOverflow answer, which might be wrong or outdated. You just try it and see what works. If it doesn't work, you tinker with it, go to the documentation, etc. You can also tell ChatGPT it doesn't work and show it the error and it sometimes has ideas.

StackOverflow is a bad example, because you see the context, and this is what is definitely missing from ChatGPT. On SO, you see the votes for this answer, you see the date, you see the comments, you see the user who posted it etc. etc. When I check answers on SO, I immediately check at the date, because sometimes it makes the answer invalid, which you can't do on ChatGPT since it's a complete black box. I prefer other tools like Phind or Bing that attempt to show the sources of what they're saying, but it's still a lot of things to process.

StackOverflow isn't the best example for the reasons you outlined. I was just trying to illustrate the trust but verify mindset. Any time you find an explanation or a solution or anything out in the world, you test it out, read about it, look up terms, etc, to get your head around what's being suggested. It's not blind copy pasting and it shouldn't be.

You can't, which is why I'm not too worried (at the moment at least) for my job. It's currently a boost to people already knowledgeable in the area, but people with a lack of understanding of the underlying principles quickly become stuck.

Search on Youtube for videos by Sam Vaknin. He explained this by what is called a 'zimbo'. An entity that looks like a human, walks like a human, talks like a human, but is essentially devoid of human-ness. The one and only way to determine the difference is trusting your gut, it 'feels' wrong. Sam Vaknin uses the term 'zimbo' (derived from zombie, I guess) to explain relationships with cluster B disordered people, but I think it can be used one-on-one on ChatGPT. Utter non-sense with a lot of conviction sometimes.

TLDR: trust your gut feeling.

Not sure that's a universal truth. I've worked with people whose gut feeling was so wrong and at the same time their overconfidence in it was so high that it was impossible to get anything done. Heuristics only work if they apply to reality, and they have intrinsic limits too.

I never said everybody's gut feeling works as it should be. That's definitely not the case. I'm only saying one's gut feeling is one's only chance at telling the difference between real and fake in the situation described by OP. After you pin pointed your gut feeling, your rational mind can take it into account. Watch out for rationalizing your gut feeling away though (HN crowd is especially prone to this, all logical thinkers). So listen to and develop your gut feeling. We're going to need it with all the AI output.

You’ll notice when the api call breaks or the google search comes up empty.

In the simplest case sure. It’s just as likely to add subtle bugs that don’t immediately show up.

Reply on news.ycombinator.com