[skeptrune]

I'm interested. Could you use a LLM function call to decide whether or not to respond instead of randomness so it feels more intelligent?

Also, I have no idea what the use case for this would be but making it work sounds cool and kinda fruitful.

[lgas]

> Could you use a LLM function call to decide whether or not to respond instead of randomness so it feels more intelligent?

Yes, and once you collect enough samples from the LLM you could train a SLM to do it faster and cheaper.

[Jowsey]

> Could you use a LLM function call to decide whether or not to respond

I've implemented this for a toy project before and it worked surprisingly well, yes! It can take some creative prompting to have the model understand which messages are actually directed at it, though, which I guess makes sense since every User message is _supposed_ to be?

[vvoruganti]

Curious what kind of things you had the model look for? Did they have any kind of goal or was it more a "vibe" thing.

Also would it only respond if a message was directed at it or could they proactively decide to join a conversation?

[thierrydamiba]

What model did you use for this?