ChatGPT Is Dumber Than You Think | Hacker News Reader

ChatGPT Is Dumber Than You Think

theatlantic.com

18 pointsbehnamoh14 comments

I never thought it was accurate to begin with, but maybe it's because I asked it questions like "Tell me the history of podcasting" and it had the dates and people involved all wrong. Had I asked it to tell me about the Punic wars I might have been more impressed but I don't need an Open AI bot to get up to speed on the battles of ancient history.

I use it to generate text that is more complex than I typically would generate on my own. And then manually fix the facts in post-processing.

All that means is that it's better at stringing together words than you are

I also like that it uses some words that I am less likely to use.

It might be dumb, but it probably can still write better than most of the students in a college English class.

In other words, it takes an astounding amount of hubris to believe humans will remain universally smarter than this LLM or its successors.

If anything, I think we are reaching a point where humanity will be put in its place by AGI. We are not that impressive as a species, and we sure as hell are not special.

> we are reaching a point where humanity will be put in its place by AGI. We are not that impressive as a species, and we sure as hell are not special.

We probably will have bad stuff happen to us due to ai. But should we create true AGI, that would be special. We’d be gods, the creator. Our creation would likely destroy us, but it still would be a impressive feat.

https://archive.ph/2022.12.12-105719/https://www.theatlantic...

So the article is behind the paywall and immediately behind the paywall is the line "Hold up, though. I, Ian Bogost, did not actually write the previous three paragraphs. A friend sent them to me as screenshots from his session with ChatGPT".

At least Ian understands that intro is indeed extremely boring and in general not more than just a toy. Too many repetitive articles recently, repeating marketing statement from OpenAI from 4 years ago: “GPT2 is too dangerous to release”.

What I really worry about is uninformed Dunning-Krugerites using it to sound more articulate and intelligent than they actually are to disseminate more BS than we already contend with online. Online debate is folly to begin with, now let’s throw gasoline on the fire.

The only way to win is not to play at all. Ignore people talking about things they don't know about. Take everything with a grain of salt and verify everything, even this.

Isn't the goal ultimately to be able to continuously generate content automatically to placate the masses? Think Wall-e but in a more distopian kind of way where the humans don't even bother interacting with each other.

I'm envisioning partisan bots just arguing with each other ad nauseam. It will be interesting to say the least.

ChatGpT does not know anything. It vomits words that are statistically likely to convey meaning, but there is no guarantee that it is truthful: it is always a probability, not a certainty.

How is that different than an average human? Wanna take odds on which one appears to be smarter on average?

Reply on news.ycombinator.com