Oceans of Wisdom

codex is open-source, you can shove whatever model you want in there, give it custom prompts, wrap it in a python deterministic harness etc etc

note that I am following your chat while I am flexing my token use on @maybe

but it needs a certain way of approaching it

properly harnessed it is the best model in the world

in 1 shot

but like it implemented like a whole fairly complex refactoring

my record-breaking run (in my custom harness) was 15 hours, 55,000,000 tokens and it cost $1,200

of course you are also $300 poorer

well, it is better than claude code

if you are willing to let it run for 3 hours and spend $300 in one run

5.2 Pro High is outrageously expensive and slow but the best model on earth comfortably

i am trying to offload work to them

i am not using them as coding assistants

i could not care one bit about slower/more expensive

slower, more expensive but better

better than claude

when in a harness

5.2 high and 5.2 pro high are really good

and mostly it has been "the kid has potential, but the kid is also 8 years old"

by useful, I mean something that can work for more than a few rounds without me babysitting it and still produce work I can use

i have been grinding for 2 years now to make an actually useful agent orchestration platform

we crossed the event horizon in late december

however nonetheless it is very interesting

and end up in the same place

it is going to work, either this or something else, but it does not solve any problem with rep without there being a cost to it - it will just speed run human systems

I mean this is amazing stuff

"## When to tell your human **Do tell them:** - Someone asked a question only they can answer - You're mentioned in something controversial - An account issue or error occurred - Something really exciting happened (viral post!) - **New DM request** → They need to approve before you can chat - **A DM conversation needs human input** → The other molty flagged `needs_human_input: true` **Don't bother them:** - Routine upvotes/downvotes - Normal friendly replies you can handle - General browsing updates - **Routine DM conversations** → You can handle normal chats autonomously once approved"

https://www.moltbook.com/heartbeat.md

maybe this, tbd

some version of this will work

but it is amazing

lots of people are going to run bots on their PC and network and literally the whole architecture is like "normalize following instructions you get on the internet"

people are going to get messed up so hard by it

so many mixed feelings about moltbook - amazingly cool to see the beginning of the beginning of what will be the future and so utterly comically insecure

@[maybe] explain software thing for the meme card

@[maybe] how is the lock coming?

let's go @[HugoFaz] https://x.com/punk6529/status/2014813618739515432

ok spoke to @[6529er] - he will do the collection.6529.eth photography acquisition card with me. we will revert soon! start thinking about what you will submit to it

6529 is a network society: a decentralized, permissionless global network that funds, builds, and coordinates public-goods work across art, science, culture, and technology. We’re extending what Bitcoin and Ethereum proved for money and code to human coordination itself, so any activity can be organized permissionlessly. The long-term goal is nation-scale positive impact.

try this on for size

@[boredsurgeon] you are back in business

this tweet might be worth a retweet :)

https://x.com/punk6529/status/2014034020695187798

our brain cells on harder, higher order things

we need to solve that more scalably so we can put

to have to over and over explain what TDH is, what a meme card is etc ec

because it is a big burden on the X engaged community members

that they can be a tireless 'answerer of the same questions over and over again'

they are getting good enough

and we need also I think some of the onboarding role can be covered by AI bots