am I missing some context here? My understanding is you're not going to get the same word for word results with each result, and even if it does give the same word for word response, how did SWB know the prompt mr. Ye used?
How did they find out or are they just flat out lyi... being deceitful?
It's a trained language model. People have already shown where is a search result and the data it brings up are frequent enough, the AI models will always produce the same result.
In otherwords, the AI was trained on enough performative apologies to spit out the same bullshit, more or less, each time it's asked something like this. Modeled after real world apologies.
I understand but here is what I found sus about the post:
Bruh SWB's post shows he used GPT3.5. GPT3.5 does not write a single paragraph it word vomits.
the google translation is word for word (minus a badly placed "respected") what SWB's posted of his gpt 3.5 result
GPT3.5 is fucking wild, you can try it yourself, I used Poe and using their gpt 3.5 it won't even generate an apology due to those naughty words used in SWB's prompt.
I know enough about LLMs to know it has randomness due to seed and temperature. I'm not saying Mr. Ye didn't use chatgpt to write an apology and nothing wrong with that imo, it's no different than hiring a PR guy to draft it for you only its free.
What I am saying is: that image of the chatgpt response being word for word the same as the google translated apology - Bullshit.
am I missing some context here? My understanding is you're not going to get the same word for word results with each result, and even if it does give the same word for word response, how did SWB know the prompt mr. Ye used?
How did they find out or are they just flat out lyi... being deceitful?
It's a trained language model. People have already shown where is a search result and the data it brings up are frequent enough, the AI models will always produce the same result.
In otherwords, the AI was trained on enough performative apologies to spit out the same bullshit, more or less, each time it's asked something like this. Modeled after real world apologies.
I understand but here is what I found sus about the post:
Bruh SWB's post shows he used GPT3.5. GPT3.5 does not write a single paragraph it word vomits.
the google translation is word for word (minus a badly placed "respected") what SWB's posted of his gpt 3.5 result
GPT3.5 is fucking wild, you can try it yourself, I used Poe and using their gpt 3.5 it won't even generate an apology due to those naughty words used in SWB's prompt.
I know enough about LLMs to know it has randomness due to seed and temperature. I'm not saying Mr. Ye didn't use chatgpt to write an apology and nothing wrong with that imo, it's no different than hiring a PR guy to draft it for you only its free.
What I am saying is: that image of the chatgpt response being word for word the same as the google translated apology - Bullshit.