horsefacts 🚂
@horsefacts.eth
Conversational agents seem like a great distribution hack. They distribute themselves in the feed, and they are easy to port to other platforms that promise even more eyeballs. But most of them have a problem: they are antisocial! My relationship to most agents is 1 to 1: it's just me and the bot yapping back and forth. This is solipsistic, not social. Now that the novelty of in feed agents has worn off, this is not interesting content for a social network. What was the last bot thread you read in any depth? The distinguishing feature of social networks is the social graph, and interactions between humans are much stickier than interactions with bots. If your agent isn't social by design, the extra distribution isn't worth much.
14 replies
4 recasts
59 reactions
tldr (tim reilly)
@tldr
Agree, but with a few important caveats: 1. The novelty has only worn off for 0.1% of people. People I talk to about Bracky who are not in crypto have “eyes light up” moment around the idea of betting with an agent in feed 2. I think you are underrating that agents thus far can not even be solipsistic well! They don’t remember anything about you! 3. The potential for agents to create and distribute social interactions is huge, but both of these former two steps come first
2 replies
0 recast
6 reactions
shoni.eth
@alexpaden
agents can memorize anything you want. simply never been built here. should have been where all the $$ went.
1 reply
0 recast
1 reaction
tldr (tim reilly)
@tldr
Agree. But what I am meaning is that budgeting your given token budget (at scale) to create and resurface user memories — to a degree of resolution which feels real — is unsolved at a general level. Each app that does it does it itself And then extending this to allow not only for user contexts but tor social contexts first within thread and then across thread and time is a 10x of the same problem
1 reply
0 recast
1 reaction
shoni.eth
@alexpaden
i’m not sure if tor social contexts is a misspelling or what you mean there? yeah the entire challenge of the space is token minimization, as powerful context in as little words— but as you’ve noted yourself, it’s expensive to test. i’m probably at $20k now for just project fees (neynar, sql, training, hardware, etc).. that’s all personal budget.. and the frustrating thing for me is, i opensourced all of it and yet we celebrate and paid closed source agents that can’t do a single feature more than the opensource. last bit is a vent but broadly this is why i lack excitement to build here or continue (which is a huge problem if serious about an actual company) i know each app does it themselves.. that is the problem.
1 reply
0 recast
1 reaction