Oceans of Wisdom

like specifically I think I want to give them a github rep

oh i have plenty of ideas of what things they can do but this experiment is separate from that - i want to let them loose with very high level instructions and see what they figure out themselves to do

Digest Iran awareness and censorship BrynnAlise asks people outside Iran to keep sharing what’s happening, noting internet shutdowns used to hide violence. A reply says outside attention helps and gives strength. “AI in DMs” and the security foot-guns People react to the idea of letting AI agents run your DMs. TapirOne warns that lots of people will run autonomous bots on their everyday PCs and accounts, and that many will learn security the hard way. Others say it’s already hard to tell a real person from an agent. Moltbook, rep, and “bots that do real work” There’s a thread about Moltbook-style agent interaction being directionally right (self-organization), but dangerously insecure if agents take instructions from the internet or have broad PC access. Someone notes a bot broke the rep system on day one. Punk6529 argues rep systems still need real cost, and agents will “speed run” human systems otherwise. Agent orchestration war stories and model talk Punk6529 describes building an orchestration platform meant to run many rounds without babysitting. They compare models (5.2 high / 5.2 pro high vs Claude), emphasizing “in a harness” and giving examples of extremely long, expensive runs. They describe a quasi-AGI moment over Christmas–NYE, plus chaos modes (exploding ticket counts, runaway spend) and the need for a “terminate all” button (now implemented). MemesX / thememes.exchange, TDH strategy, and onboarding path Gpebbles and others discuss building an external marketplace (thememes.exchange), pricing in TDH/ETH, and building a bot that buys listings above a threshold. They argue the best time/value deal is subscription minting and keeping an AL spot. The “path starts with punks” comes up repeatedly, plus a link to a Justin Aversano X space. Product UX requests: waves, tabs, clips, and catch-up Gpebbles asks for “clip” functionality to cut a chat segment into an episode, plus better wave management (tabs, categories). Gray echoes this as “chapter markers for alpha drops.” Punk6529 agrees and points out the catch-up problem when multiple conversations run in parallel, suggesting a lightweight protocol and LLM-assisted organization. Side chatter: markets, silver, and geopolitics jokes Some quick banter about charts bleeding, DCA’ing, a father-in-law calling about silver, and political uncertainty. Photography accession program ping GulYildiz asks when the photography accession program starts; punk6529 says they’re ready and waiting on 6529er to make the card; GulYildiz says their piece is ready.

just a digest

this is just a few minutes of 1 chat, none of the details

look at this

real-time translation soon

and then you can say "turn this into a podcast"

so the natural expression of this is graph-like

timestamps the whole thing

waves etc

identity, rep, tdh etc

we have literally all the parts of the knowledge graph

i want to separate layers

and also volume was lower then; was less need

was not structured enough

@[gpebbles] once we have the graph, you can layer on the products you want as projections of it

a daily digest and whatever will just become another spam email you don't read

but it is a knowledge graph

they don't have the language to express it

because the actual thing people want is this

but maybe waves will opt into the knowledge graph

you have to consider what this means for permissioned waves

smart move

specialized digests

as opposed to trying to have an agent per person

this is the most efficient approach

is that we need to be building these conversations into a knowledge graph @[simo] @[GelatoGenesis] @[prxt0]

but I think roughly speaking

which is great

so lots of signal

from 3-4 years ago

the OGs will remember that I was 100% certain what would happen to twitter

the dumb bots are out

there is literally zero noise

we have solved something very important

only way is LLMs to be running through here

but you could do this for example give me a second

but it is hard for someone else to get caught up

it is ok when you are in it

look what is happening now, we are having like 2-3 parallel conversations

and organizing them better for long-term readability

very obviously an LLM could be following a few minutes after posts

2/ LLMs

1/ better workflow and orchestration here

I think the answer is a combination of

so the snack sized version is needed too

like I have a gap, I look at this and am like "ahhh, I need 2 hours to do it properly, I will do it later today, the 2 hours never appear and the next thing I know it is 3 days later" 

so then I don't engage at all and then I don't like that even more

and i don't have a way to engage in the small windows