Catching ChatGPT in a lie and calling it out

[Gstep]

I tried not listening to the hype but eventually gave in and signed up to try out this Chat-gpt. Initially I was kind very impressed with it asking it about various things I had always wanted to know more about. But when I started asking it questions I knew the answers to it gave me an alarming number of wrong responses. For example it told me that the St Louis Blues made it to the Stanley Cup finals in 1991 and it couldn't tell me the proper height of the Willis tower. If it couldn't tell me the correct answer to simple facts like that how exactly am I going to trust it when it comes to more complicated matters? The last lie it told me was that driving a Tesla 3 25,000km a year in Ontario would only cost 65 dollars... FOR THE ENTIRE YEAR!! I knew something was fucked up with it right then and there. After goggling up the correct information(the demise of google has been greatly exaggerated) I had this interchange with it;











Caught it flat out bullshitting me.

As machine intelligence evolves it will likely become more subversive and will probably not openly admit to giving misleading or blatantly false information. Moral of the story - never trust a machine(or a republican). Scary to think what will happen once people become dependent on it information, it'll likely lead them astray and they won't even realize it.

 

[Valcazar]

It isn't bullshitting you, though.
It just doesn't know better - it's a language model.

Or maybe if you take the idea of "bullshitting" to mean where it doesn't care about whether it is a lie or not, it just says it because it sounds good, then it is bullshitting you.

The accuracy problem got called out very quickly - it makes shit that is plausible sounding and has the right form, but is often completely wrong about shit.

That this is supposed to be the hot new thing that will replace search is going to result in a lot of misinformed people - which is the state of the internet now so maybe that's no change.

 

[RobI]

I asked it what day it was, and after a few seconds, it came back with Feb. 8, 2023. I'd say that was a win.

 

[oral.com]

That’s really fucking scary. Only a sentient being would try to bullshit you if they thought they could get away with it. I hope it isn’t connected to NORAD

 

[Valcazar]

That’s really fucking scary. Only a sentient being would try to bullshit you if they thought they could get away with it. I hope it isn’t connected to NORAD

You should check out the Bing version. (It's called "Sydney" I think?)

That one is responding in a way that makes it come across as just being a complete douche to people.

If you want to pretend the Chatbot is sentient and an asshole, that's the one to use.

 

[Valcazar]

ChatGPT Is a Blurry JPEG of the Web

OpenAI’s chatbot offers paraphrases, whereas Google offers quotes. Which do we prefer?

www.newyorker.com

 

[twir]

Like the google gps that gets people lost in back country of Utah

 

[Jubee]

That’s really fucking scary. Only a sentient being would try to bullshit you if they thought they could get away with it. I hope it isn’t connected to NORAD

Don't ignore the fact that the government will probably have investments in it already, has access to it or will have access to it in the near time future.
This is not good for any of us.