49 ms·
It has to generalise just to be able to fill in the gaps within-domain. If it was not so, I could copy-paste lines from this into google and find out where it
by ben_w 2y ago
It has to generalise just to be able to fill in the gaps within-domain.
If it was not so, I could copy-paste lines from this into google and find out where it was quoting from: https://chatgpt.com/c/3105db2c-b88d-42c6-83bb-ea2b9d6d65a1 https://chatgpt.com/c/3105db2c-b88d-42c6-83bb-ea2b9d6d65a1
> unlike a human
You know the phrase "thinking outside the box"? It's a thing people look for specifically because it's rare.
I can't tell how good or bad LLMs are at this compared to a human due to LLMs having a ridiculous advantage when it comes to breadth of knowledge.
My guess is they are indeed worse than us at generalising.
But they do generalise. They have to generalise even to be at this level.