Hacker News Reader

Comment by outside1234 | Hacker News Reader

Of course it did - this whole technology is AI hallucinating about facts. This is exactly what the math tells us it will do.

It is not a "fact generating" technology - it is a "scaffolding" technology to make it easier to fill in the boilerplate of the answer -- but not the answer itself.

View on HN

It's in vogue to hate Google at the moment - so everyone's just going to pretend it doesn't matter that MS had errors sometimes because "something, something Google sucks now."

Just wait until the GPT dataset includes an infinite amount of GPT generated SEO garbage. If you thought searching for relevant data couldn't get any worse, prepare yourself for the post-GPT mass hallucination.

And if you think that's bad, it's only going to get worse when those GPT-generated SEO sites get indexed by crawlers and fed into future versions of GPT as input.

That's exactly what I mean and it is inevitable without halting all non-human verified updates to the dataset going forward. There are two worlds now, pre-GPT and post-GPT. In the pre-GPT world you have human error and in the post-GPT world you have human error and machine error compounding ad infinitum.

Openai released a tool to detect if a text was generated by an ai[0]. They probably already thought about this and are trying to stop it from happening

[0] https://platform.openai.com/ai-text-classifier

Not to mention a rebellious anti-AI counter culture that will intentionally feed it nonsense.

I don't think that's an accurate description of the current state. It generates facts but those need to be evaluated by experts since tech can error as likely as a human.

> This is exactly what the math tells us it will do.

Technically it hallucinates on basic math as well. So a fitting choice of words :)

Reply on news.ycombinator.com