Olivia Moore from a16z points out that even though LLMs have exploded as the quickest-growing consumer category on record, every major chat provider still treats sharing as little more than passing a static link in a message thread.
It’s mindblowing to me that LLMs are the fastest growing consumer product of all time
…and the “multiplayer” experience (across every provider!) is basically just being able to share a chat history
Someone is going to crack this and it’s going to unlock a LOT more usage
Moore argues that moving past one-way history links to actual shared sessions would unlock substantially higher usage, yet no provider has shipped anything beyond basic URLs so far.
Technical and design hurdles are cited as the current blockers to true multi-user environments, though specifics on those hurdles are not detailed in the post.
Positive users praise existing hacks like Slack integrations and past tools as magical or usable for multiplayer AI chats, while negative users criticize simple link-sharing as pathetic and embarrassing given LLMs' rapid growth.
No Digg Deeper questions have been answered for this story yet.
there is a reason no one has a usable multiplayer AI chat experience. Do you know what it is?
It’s mindblowing to me that LLMs are the fastest growing consumer product of all time
…and the “multiplayer” experience (across every provider!) is basically just being able to share a chat history
Someone is going to crack this and it’s going to unlock a LOT more usage
@deepfates these things barely hold together using the coherence of one user, much less many.
there is a reason no one has a usable multiplayer AI chat experience. Do you know what it is?
@omooretweets i mean you can have group chats in chatgpt
just no one rly uses them but they work decently for social stuff… not so much for productive tasks
@omooretweets Way more fun when they’re in groups
Haven't thought about it this way. My instinct is it follows search.
Google was never multiplayer, even though they tried (SearchWiki, Google+ results inside search). It all failed because people search with solitary intent (most of the time). I think the vast majority of people use the models the same way today.
LLMs feel 1-of-1. Every output is generated for me and my context (from a consumer pov), so multiplayer is at odds with it on that.
But I do think that's a result of the chatbox interface, not the model. So it's probably somewhere between a use-case and design problem to be solved - super excited when it does get solved!
@omooretweets the closest thing to this right now is slack channels & telegram chats with both humans and agents, and it feels magical. It’ll be epic when there’ll be a good way to do this in imessage and whatsapp
@omooretweets The Slack based things (Centaur, Town, Claude Tag, and many others) are sort of this
@omooretweets Isn’t that just Town, Tasklet, Stilla, et al?
@deepfates instruction finetuning expects there to be only one question at a time
@omooretweets For consumer and enterprise. Diff UXs are coming
@omooretweets That’s why we built Alfi :)
@omooretweets multiplayer is severely underdone, there's so much potential. zuck sees it for sharing of codex output for eg
@omooretweets working on something cool
@deepfates imo, privacy/security for most chats that pull meaningful context
@alienpisscrack Likely true. And multiturn RL expects that one question to take several answers and then return
@deepfates On our IRC channels it works just fine. (So does it on a variety of Discords.)
It's just that noone found (yet) how to make it that valuable / useful, besides being source of some (pretty nerdy and niche) fun.
(Something something Claude Tag something. Or @MuaddibLLM.)
@IamKushagraB @omooretweets Group chats in an AI product make no sense. Low data density and single threaded.
Collaborative documents make much more sense.
@deepfates If no user they turn into weather.
@omooretweets What about remixes? From @wabi & @Lovable apps to tools in @GoogleLabs Flow, I can blend outputs into new ones
Or Templates, e.g. on @ElevenCreative Flows, or access to AI workflow-files, e.g. in @canva, to let others play w/ my work
Isn’t it multiplay? Or do u meant text only
@anselm_io One might assume Network effect would counteract this tho