... is to not embed an LLM in Group Chat.

If my friends start adding LLMs to our group chats I'm gonna be in the market for some new friends.

I can't imagine trusting any LLM for private messaging that doesn't come as part of OS

I find the first two responses "Perfect!..." and "Based on..." intensely annoying. They add nothing to the discussion and act as noise. This article crystalises a problem with AI: words are cheap to generate and expensive to read (just like spam). The AI should just do as it's asked and succinctly generate the poll. If asked "Why did you make these suggestions", it should explain why it chose those restaurants.

Edit: I'd also add that done "right" the AI should be clearly labelled as such. Its chat handle should include something like "AI", so the other chat users know exactly what they are dealing with, without having to know what "TripJam" or "TJ" is.

The only thing I can imagine being funny is image generation, but its usually too slow to be relevant as your conversations change.

I have a sports betting group chat "Degen Bets" that I made a funny image for the thumbnail based on our fantasy football teams.

Having an assistant everyone can co-drive is actually kind of useful, but I don’t want it in my conversational flow.

In a temporary thread about something yeah.

yeh I type in Paris, now I get a paragraph to read with random suggestions with no context to verify if these are good for me or not. If User has to figure out and make choices and tell AI to plan the trip then what is the point. It is easy for me to see nice UI with reviews, image grid and make choice based on this type of data. Why is the purpose of back and forth discussion with AI ?

nice write up! covers every consideration I had to go through while building group chat myself for https://olly.bot