undefined

points

[-]

> the LLMs were trained to mimic how people speak (write)

It's a foundational model (fresh from autoregressive pretraining) that approximates the probability distribution of human texts. And, no, it's not the statistical average of how people speak. It approximates how a person who could have written a text in its context would have written the next words.

Fine-tuning, RLHF, reinforcement learning change this probability distribution. I guess, it's mostly RLHF that shapes the way LLMs write. The similarity of style is due to common providers of RLHF data.

by rogerrogerr1 hours ago|

prev|

[-]

But no _individual_ speaks like this. It is not common that a guy representative of the Average Of All People Plus Some RL writes a blog post.

by Terr_1 hours ago|

parent|

[-]

That reminds me of story where the US Air Force attempted to design a cheaper cockpit and chair that would fit the "average pilot".. but in practice it fit nobody well. They went back to the drawing-board and put in all sorts of adjustable components instead.

This is likely because the individual measures being averaged like "femur length" were not independent from one-another, even where they had the benefit of being normally distributed.

by win311fwg1 hours ago|

parent|

prev|

[-]

The specific quote that we are talking about reads like something I wouldn't find unusual in the works of an individual. Maybe not from someone who stands out as someone who writes things that you'd actually want to read, but from an average Joe? It is likely. I, being no wordsmith myself, have written things just like that. It looks quite typical of how people write to me.

In fact, I find no meaningful difference between "But no _individual_ speaks like this." and "But the ratio was never the danger.", aside from them being about different topics. They are syntactically very similar.