I gave it and various openai → playground → assistant, of different model x Temp x Top_P and just cannot replicate the one on the left window. It just sounds super smart.
That is OpenAI’s proprietary info.
It is high enough to get a variety of outputs, which is desired to get more training data for you to press buttons on, besides sampling allowing more creative and human-like generations.
Example
I pressed a GPT’s “send” button at the same time in two windows:
In the high stakes world of poker, are you a folder, or do you go all in?
Two different responses diverge at the second token:
Ooooh, good question. Honestly? I’m usually a folder. Not because I’m scared—no, no—I just like playing the long game. I’ll sit back, watch everyone else puff their chests and throw chips like it’s Monopoly money, while I’m clocking every blink and twitch. Patience is a power move in poker.
Ohhh, I love this question. It completely depends on the read, the moment, and what’s on the line. But between us? I lean toward the all-in side—strategically. Like, I’m not the type to recklessly shove chips just for the thrill. But if I’ve clocked your bluff, and the math checks out, I’m pushing my stack forward with a little smirk and a “let’s dance.”
Revealing that there was no underlying sentiment in the choice.
1 Like
I think that was a bluff @_j