Hacker News Reader

Comment by kolinko | Hacker News Reader

A nice approach - estimates the randomness in text. I tried with various cases - blog posts, wikipedia, scientific docs, and with a few examples from chat, and it figured it out accurately.

Respect. Even though I disagree with a need for such tools - it doesn’t matter if content was written by a human or by a machine. What matters is whether it’s easy to read and worthwhile.

View on HN

Disagree, this is very much needed. The problem with AI-generated content is that it looks superficially worthwhile and plausible, but in fact it often says things that are not correct.

> The problem with AI-generated content is that it looks superficially worthwhile and plausible, but in fact it often says things that are not correct.

This is also a problem with human-generated content.

Depends on the human

Sort of. But if I read a five page article by a human, and everything on the first couple pages checks out as correct, I expect the rest of the article to be at least reasonable. If it’s written by AI, it’s entirely possible that it’ll start in on something that doesn’t even make sense. Then I’ll realize that I can’t even trust the part that seemed correct.

But that is true also of lots of content not generated by AI! Fact checking always needs to be done, AI generated or not. But does it matter that it was generated by AI?

Which is true for most human-generated content, too.

Depends, only the one that do the 10 minutes YouTube video and/or 500 words article to make money from ads.

If you dig well, there are people who do really good content out there

What I find fascinating is that aspect of AI-generated content actually captures what humans do all the time, confidently making incorrect statements, "bullshitting" with filler text that only pretends to be meaningful, making illogical statements that contradict what was stated earlier, etc.

> it doesn’t matter if content was written by a human or by a machine

It matters a lot instead. On the top of my mind I can think about few reasons:

- if you've to train a NLP model you'd rather do with data that is not autogenerated as AI generated content is rarely used to train new models (like generating a bounce of dogs pic with dall-e and use as input for image detection may not create a precise model)

- if you pay content creators to generate content and they use ChatGPT that's definitely a breach on contract and also a problem

- many search engines (e.g. Google) already heavily penalize auto generated content

- avoid cheating at exams / officials test/certs

> Even though I disagree with a need for such tools

“Need” may be arguable, but why disagree?

Yeah what we really need is a language model that can take some text and remove everything that is void of substance, and run both human and machine writing through it.

Reply on news.ycombinator.com