Sander Schulhoff

“Oh, so technically that would be zero-shot. There's a lot-”

“... all fairness, across the industry and across different industries, there's different meanings of these, but zero-shot is no examples.”

“One-shot is one examples, and few-shot is multiple.”

“Yeah, well, even within ML, there's research papers that call what you described one-shot. So it's-”

“That's another thing you [inaudible 00:17:22]. You don't even necessarily have the inputs and the outputs. In your case, you just have, I guess, outputs that you're showing it from the past.”

“Yeah. This is perhaps the question that I am most prepared for out of any you'll ask, because I've spoken to this over, and over, and over again, and gotten into some internet debates about.”

“I do remember at some point we put out a tweet and it was just, "Role prompting does not work." And it went super viral. We got a ton of hate. Yeah, I guess it was probably this way around, but anyway...”

“The math professor one would actually get easier to explain.”

“Telling it's a math professor could activate a certain region of its brain that is about math, and so it's thinking more about math. [inaudible 00:24:01]-”

“So I do distinguish them, and I think with this example you'll see kind of why.”

“Yeah, that is the phrasing I like to-”

“I guess you could do it infinitely. I think the model would go crazy at some point.”