Welcome to the @Distributed____ Systems Project 🤗 your flesh is a relic, a mere vessel dsco2048.substack.com before: @UofMaryland @GoldmanSachs
📍 Washington, D.C.
- Tweets 10,594
- Following 7,510
- Followers 2,480
- Likes 24,425
Arthur retweeted
It feels so fucking good seeing your homies win
Nous Research has raised $90 million at a $1.5 billion valuation, I know a lot of people from the team, and they are a really great group. Teknium helped me when I was starting out here. Hermes has been a huge success for them.
12
7
285
6,515
A world is any system that is already running with live state: a trading book, a ledger, a weather feed, a fleet. A mind is what accretes on it when you talk to it: ask for a capability, the proposer writes it, the world checks it against its own data, accepts it, and the module fans out to every connected node. Learning is a code load. Memory is the catalogue, with source and docs attached, so the mind can read its own definitions, which weights have never managed. The edge of the mind is the edge of the cluster; a partition is a mind splitting into two that keep thinking separately, and a rejoin is reconciliation at message boundaries, where the two-version rule stops being a BEAM quirk and becomes the mechanism by which a divided mind agrees with itself again.
Crashes are forgetting, and forgetting is a feature. A supervisor tree is an immune system: anything malformed dies and is replaced, and nothing that dies takes the whole mind with it. Models cannot do this. A corrupted weight is a corrupted model; a corrupted process is Tuesday.
The model is not the mind. It's the proposer, a consultant you call when the world needs a capability it doesn't have yet. Once a definition is accepted it's just Lisp, and calling it costs no inference. So the economics invert: every agent framework today pays tokens per action, forever, while this pays tokens per capability, once. The cost of thinking amortizes toward zero and the cost of learning stays bounded by how often the world meets something new. A mind made this way gets cheaper the longer it lives.
And it outlives its proposer. Swap the model, the catalogue stays. Retire the lab, the cluster keeps running. The weights are a contractor; the catalogue is the institution. That's the expanse: not a bigger model, but a world that has been talked to long enough that it no longer needs to be.
1
99
shout out to all the RSUs that are gonna pay out for everyone, cheers lads
youtube.com/watch?v=Rwt4AYQi…
103
Arthur retweeted
Chart of the day:
Agentic traffic has surpassed human traffic online for the first time!
21
83
476
68,462
If you thought cyber and bio threats were the scariest things in the utility belt, just wait until you hear about mirror life
1
10
353
Arthur retweeted
The peak years of San Francisco are yet to come
i moved there in 2008, age 24. was one of best decisions of life. but thats when san francisco was amazing, everyone in history agrees those were the peak years (2008 - 2014 to be exact), and now its changed, and it doesnt appeal to me as much anymore.
5
3
65
5,239
Arthur retweeted
this is so fun
Below: an LLM loop is hooked into the Lisp REPL, and the LLM adds functions to the REPL as needed. These functions persist across application restarts.
Instead of the entire read_file + edit_file + bash compile loop, it's now just prompt functionality in the application as it runs.
28
21
238
14,509
Arthur retweeted
Realizations after returning to NYC from Tokyo:
- New York is dirty & gross
- New York is expensive for no reason
- New York is a scam
- New York lacks infrastructure
- New Yorkers are rude & bad
- New York is borderline uninhabitable
- New Yorkers have Stockholm syndrome
564
667
11,198
769,715
