How do "actions" work

[chilly berry]

im new here and not that knowledgeable with ai. how is neuro able to do stuff like control her voice speed and use the sound board?

[spark kettle]

from what it looks like, and with the game api, she just has the option to use a command that controls it, I'm guessing it embeds in her text something like {spin} or {sound:vine_boom.mp4}, and she's given that option when that function is enabled to use whenever she wants

its most likely very like this
https://github.com/VedalAI/neuro-game-sdk/blob/main/API/SPECIFICATION.md

[chilly berry]

thanks alot will be pretty usefull for me

[jade basin]

they’ve probably been trained to be able to format stuff like that with a high success rate. the twins will still occasionally fuck it up.
how vedal made iteration 18 without severe latency increases? magic maybe

[hidden token]

you forgot that vedal obsesses over l*tency to a high degree

[obtuse lotus]

look up tool-use in LLMs, basically what bluecat said

[chilly berry]

do i need to close this?

[hidden token]

nah just leave it I think it's fine

[chilly berry]

ah ok

[neon cosmos]

AKA 'function calling' in most popular LLM's

[chilly berry]

is this also called agents?

or is that a whole different thing

[neon cosmos]

no, thats completely different

function calling allows a LLM to interact with any given tools (aka functions you can write in a programming language), API's, etc.

agents are autonomous AI models, executing tasks on their own once provided a goal, which are also able to utilize any tools provided

[chilly berry]

ah that makes sence ty

[chilly berry]

wow i was stupid 3 months ago