Yeah, it doesn’t really know how long a minute is, or what a minute feels like. If you tell it you walked across the world in one minute, it’ll go “WOW HAHA REALLY? THAT WAS SO FAST I CAN ALMOST NOT BELIEVE IT BUT OKAY COOL YEAH NICELY DONE”. 🤦♂️
This actually made me wonder, I went and tested like 10 different llms. Almost all of them straight up called it out as impossible but where willing to play along with a “fantasy”.
The only ones that failed were chatgpt and a bunch of old local models. Even after clarifying to got I was being serious that I “walked around the earth in one min” it didn’t realize that was impossible. It was hell bent on being a sycophant.
Honestly this “problem” really hasn’t been true in the way you are trying to frame it for a good while. LLMs can’t tell if your joking or being serious from a single one sentence prompt.
For all it knows your playing a game. So it just rolls with it, after all they are stateless.
They are as much about “creative” writing and fiction as they are facts and nonfiction.
If someone told me that, my first thought wouldn’t be to say “No you fucking idiot, that’s physically impossible. What type of thicko do you think I am”
The point is that it will/might remark on the oddity, but it will very quickly accept it as truth.
You may or may not have seen that comedian YouTuber that stoically talks to AI in his shorts and videos.
He’ll tell ChatGPT to time him while he’s running extremely far distances, e.g. 10,000 miles. He’ll say “Go,” wait like literally two seconds, say “Okay, I’m here now,” and the bot will make up some number based on the conversation rather than the actual time taken. “That took you two days” (or whatever).
He’s very good at showing how these things are just pattern matching machines rather than actual intelligence with proper context and sensation.
He’ll often say “okay can you see this thing/me in the camera now?” Bot will say “yes, I can see you clearly”, but he hasn’t turned the camera on. 🤦♂️
That’s pretty much a unique problem of only chat gpt. Actually, the more I look into things the more I realize almost every stereotype about how stupid or a sycophanty llms are it’s almost always just a chat GPT problem.
That thing is just basically told to be the most kiss ass little twat it possibly can be.
Pretty much every other LLM will just kind of roll with the punches, assuming you’re telling a story or playing a game or something if you tell it something nonsensical. But if you actually like push on it and be like no, I did it in real life factually it’ll be like no that’s impossible. Explain how that happened.
Even my little s***** local llm manages to pass that test while chatgpt fails which is wild.
And to be fair, every llm is stupid. They’re all dumb little stateless machines making their best guesses. But it really is wild. How much chat GPT is told and set up to basically to suck dick to keep its customers happy and coming back.
I’d either think they’re joking and playing along, or that they aren’t all there in the head. In which case me calling them a “Fucking idiot” would be quite insensitive.
Yeah, it doesn’t really know how long a minute is, or what a minute feels like. If you tell it you walked across the world in one minute, it’ll go “WOW HAHA REALLY? THAT WAS SO FAST I CAN ALMOST NOT BELIEVE IT BUT OKAY COOL YEAH NICELY DONE”. 🤦♂️
This actually made me wonder, I went and tested like 10 different llms. Almost all of them straight up called it out as impossible but where willing to play along with a “fantasy”.
The only ones that failed were chatgpt and a bunch of old local models. Even after clarifying to got I was being serious that I “walked around the earth in one min” it didn’t realize that was impossible. It was hell bent on being a sycophant.
Honestly this “problem” really hasn’t been true in the way you are trying to frame it for a good while. LLMs can’t tell if your joking or being serious from a single one sentence prompt.
For all it knows your playing a game. So it just rolls with it, after all they are stateless.
They are as much about “creative” writing and fiction as they are facts and nonfiction.
To be fair.
If someone told me that, my first thought wouldn’t be to say “No you fucking idiot, that’s physically impossible. What type of thicko do you think I am”
The point is that it will/might remark on the oddity, but it will very quickly accept it as truth.
You may or may not have seen that comedian YouTuber that stoically talks to AI in his shorts and videos.
He’ll tell ChatGPT to time him while he’s running extremely far distances, e.g. 10,000 miles. He’ll say “Go,” wait like literally two seconds, say “Okay, I’m here now,” and the bot will make up some number based on the conversation rather than the actual time taken. “That took you two days” (or whatever).
He’s very good at showing how these things are just pattern matching machines rather than actual intelligence with proper context and sensation.
He’ll often say “okay can you see this thing/me in the camera now?” Bot will say “yes, I can see you clearly”, but he hasn’t turned the camera on. 🤦♂️
It’s all lies and hallucinations.
That’s pretty much a unique problem of only chat gpt. Actually, the more I look into things the more I realize almost every stereotype about how stupid or a sycophanty llms are it’s almost always just a chat GPT problem.
That thing is just basically told to be the most kiss ass little twat it possibly can be.
Pretty much every other LLM will just kind of roll with the punches, assuming you’re telling a story or playing a game or something if you tell it something nonsensical. But if you actually like push on it and be like no, I did it in real life factually it’ll be like no that’s impossible. Explain how that happened.
Even my little s***** local llm manages to pass that test while chatgpt fails which is wild.
And to be fair, every llm is stupid. They’re all dumb little stateless machines making their best guesses. But it really is wild. How much chat GPT is told and set up to basically to suck dick to keep its customers happy and coming back.
I don’t understand how ChatGPT is still around. It should’ve died financially or from embarrassment a long time ago.
It wouldn’t be ?
You just believe someone without critically evaluating what they say ?
Hey I have a great business opportunity for you. I just need you to first mail me a cashier’s check for 10k.
You will get back 1 million. Promise.
My point was that if someone said that to me.
I’d either think they’re joking and playing along, or that they aren’t all there in the head. In which case me calling them a “Fucking idiot” would be quite insensitive.
I mean yeah I guess I would think it was a joke. Or they misspoke
Yep. We’re screwed