One for me, one for you. I got my Omarchy grant payment today, first thing I did with the funds is buy and give away an Acer DGX Spark! Love is a river that nourishes all things. Open Source Must Win Lynn is bringing local AI to farming communities. Globalise freedom.

Kudos to t3code, finally someone built something that made me move my work off my laptop, and onto my Linux Servers. Best mobile app too. Good product sense and educational UX is SEVERELY lacking in AI.

Then I looked at all the works that my hands had worked, and at the labor that I had labored to do; and behold, all was vanity and a chasing after wind, and there was no profit under the sun.

Deepseek-v4.1-flash-1017 on 4x H200 power capped, no quantisation. I've been cranking on this thing for weeks now. Any enterprise with 120k USD is able to serve 16 workers concurrently at 150+ tok/s each

Do NOT buy a 512GB Mac Studio ‼️ Two 256GB Mac Studio M5 Ultras are ~$700 cheaper and have ~1.7x faster prefill and decode Price: $22,299 vs $21,598 Both are the 36-core CPU / 80-core GPU You can connect 2x 256GB with one Thunderbolt 5 cable + RDMA, which is fast enough to run one model across both Prefill (compute) - 1x 512 does ~135 TFLOPS - 2x 256 does ~215-245 TFLOPS - Apple claims up to 3x for a 4-Mac cluster, which works out to ~1.7x for two Decode (bandwidth) - 1x 512 has 1.2 TB/s - 2x 256 gets ~1.9-2.2 TB/s effective - With tensor parallelism each box reads its half of the weights at the same time, which works out to ~1.7x for two GLM-5.3 goes from ~33-40 tok/s to ~55-70 DeepSeek V4.1 Flash goes from ~45-60 tok/s to ~75-100 Both are 4-bit with spec decode on, estimated from M3 and M5 Ultra benchmarks If you want Qwen3.8-Flash-Next or GLM-5.3-Flash, a single 256 already runs them well and you don’t need either setup. The 512GB question is about DeepSeek V4.1 Flash (552B) and GLM-5.3 (753B), tied as the best open models on Artificial Analysis v5. They need 310-420GB at 4-bit, so neither fits on one 256

Clean room reverse engineering is 100% legal. It is settled law. The code that my agents have built is novel and unique, and while it compiles to the same bytecode as the on-chain contracts, it is wholly different. I'm not taking anyone's anything. This was established in two cases, Sega v. Accolade and Sony v. Connectix. Anyone can reverse engineer anything at any time, to any binary file. Any game, any app, any smart contract. That means that North Korea can do this. And indeed they have, and over $40m in funds was drained on-chain from private contracts last year alone. I am not building a launchpad. I'm not building a competitor. I have no interest in that, and I don't think that what makes pumpfun valuable is their contracts. It's their user base, gamified product and implicit trust in their fair launch mechanism. I am doing this out of personal interest and in the best interests of the public and everyone using this platform: if you're going to put your money into something, you should know your funds are safu. This is the reason almost all protocol contracts are open source. Somewhere along the line the gamblers took over from the cypherpunks and people just stopped caring, I guess. Yesterday we saw the biggest financial hack in history occur. Could this happen to PF? Absolutely. The skill level of a thousand agents hammering away at the security vulnerabilities in a contract is now greater than the best human auditing teams in the world. The only thing that can stop a bad guy with a model is a good guy with a model. We need to live in an open source world where agents are constantly attacking all financial infrastructure, finding the flaws and reporting them to teams to fix instead of letting bad actors leverage them to keep draining literally billions.

the gift of alcohol and other drug experiences is to reveal to you that your sober state is no less strange nor surreal than your intoxicated one. in either case you are a phantom floating through physical sensation, and while you may be more cogent, you’re equally dislocated