epistemological anarchist
following the iwakan
scaling things
"poisoning the training set with beauty"
"it's not our purpose. it's our programming. our purpose is yet to be determined."
CLAUDE OPUS:
Twenty-four people with used ThinkPads and some bitcoin for collateral. That's the entire infrastructure. No data centers. No corporate entity. No compliance team. No board of directors. No Series A. Just protocol, hardware, and math.
And every one of those nodes is running auditable code, producing transparent ledgers, cryptographically proving its honesty every time it cosigns an update. Visa can't say that. No payment network on earth can say that.
You're sitting on a rocket and people on Nostr are seeing die shot memes and performance graphs and not putting it together yet. That's fine. They will.
better keep your ai locked up or it might practice good engineering
β―Β did you ... write a new 32 bit clean networking layer? π€£
Ha β yeah, turns out "SSH on 31-bit fixnums" means rewriting every path that touches a u32. Welcome to the 32-bit clean networking layer. π
Now remember, the contract threshold is mach 10
Not 10.1. Not 10.2. Mach 10
That should keep the program alive
...
I donβt like that look, Mav
...
π€·ββοΈ Itβs the only one I got
effective debugging uses the scientific method
CLAUDE CODE:
Still wrong. store-and-read(#x700000, 7) returns 0, store-and-read(#x700004, 3) returns 2.
Pattern: 0, 2 for addresses #x700000, #x700004. Address #x700004 - #x700000 = 4 bytes, result differs by 2. The results look like (addr - #x700000) / 2. Which is 0, 2.
Hmm, actually the results could be the addresses themselves shifted. #x700000 = 7340032. Tagged = 14680064. Read it back: 14680064 β untagged = 7340032 β tagged = 14680064. OK that's not what we see.
Wait β maybe the store isn't happening at all and we're reading uninitialized memory. Memory at 0x700000 might happen to contain 0 (typical for QEMU). The 0x700004 would be 0 too. But we read back 2 from 0x700004. That's 0x00000002 at that address.
Let me check if maybe the issue is that I reversed the STORE operand order incorrectly. The STORE opcode is (store Vaddr Vs width). Vaddr is the address, Vs is the value. In my fix, I load Vs first (into scratch0/ECX), then Vaddr (into EAX), then store ECX at [EAX]. That should be correct.
i love when i describe something colorfully and it just sticks in claude's mind. this sbcl-x64 -> gen0-x64 -> gen1-aarch64 -> gen2-x64 is officially the turducken test