Hacker News new | ask | show | jobs
by thomashop 1163 days ago
What makes you sure that the quality of AI generated content (that has potentially been the result of prompting and editing by a human) is and will be inherently worse than purely human generated content.

Where do you draw the line? Is using translate as a foreigner a problem?
1 comments
zvmaz 1163 days ago
> What makes you sure that the quality of AI generated content (that has potentially been the result of prompting and editing by a human) is and will be inherently worse than purely human generated content.

For the time being, I don't think you can trust AI generated content; quite often, when I asked chatGPT something I had to be sure of, it made mistakes. Take erroneous citations and references: do you think humans fake them the way chatGPT hallucinates them?

link
maroonblazer 1163 days ago
> do you think humans fake them the way chatGPT hallucinates them?

They don't need to 'fake' them; they can just be inadvertently wrong.

I have good, human, friends who tell me erroneous things all the time. I don't take them at face value, I check them. I do this for pretty much nearly every piece of information I get where it's going to inform a decision or point-of-view I'm going to take. Why should we be any less vigilant with a technology like ChatGPT?

link
zvmaz 1163 days ago
What I implied is that we should be more vigilant with chatGPT. I don't think it is common for an article to completely invent a reference that does not exist, but it is common in chatGPT.
link
creata 1163 days ago
Yeah, it's like how I'd prefer JPEG artifacts to an upscaler that completely fabricates the details.
link
zvmaz 1163 days ago
How did I not notice this analogy?! It is in fact true that a JPEG artifact upscaled with fabricated details is the same thing as a totally forged reference of a paper describing when to diagnose appendicitis in children. Thank you.
link
creata 1162 days ago
1. You know images can be used as evidence in court, right?

2. You know people can just follow up on references in papers, right?

3. You know things don't have to be exactly the same in severity to be comparable anyway, right?

4. I was going to delete my comment because it didn't say anything useful, but by then two hours had passed, so HN didn't let me.

Thanks for being a dick, pal.

link