[kirab]

The English explanations you find first for emergent are not really fitting here. Interestingly the Germany explanation for the same word seems to be much better.

I'm pretty sure they mean emergent as in emergent gameplay.

Here's a good explanation for that:

> Emergent gameplay refers to complex situations in video games, board games, or table top role-playing games that emerge from the interaction of relatively simple game mechanics.

So emergent is not inherently surprising. Games like Dwarf Fortress or Rimworld have a lot of emergent complexity but most of it was not surprising.

So in the context of LLM it means that complex reasoning is emergent based on simple underlying mechanics.

And since we thought humans’ complex reasoning skills were very unique.. We are surprised by the emergence of it in LLMs.

PS: The explanation for "emergence" is much better:

> In philosophy, systems theory, science, and art, emergence occurs when a complex entity has properties or behaviors that its parts do not have on their own, and emerge only when they interact in a wider whole.

[mjburgess]

Yes, so this is the meaning of the word I'm rejecting. NN models do not acquire emergent properties.

Human reasoning (, game play, etc.) is emergent in the sense that it is irreducible to its parts. "Generalisation in high-parameter training regiemes" is fully reducible. It is not emergence in any relevant sense of the term.

[irthomasthomas]

Sorry, I meant the way I read peoples colloquial use of it toward these llms. They aren't actually displaying emergence, as this level of performance was predicted.