ngmi's avatar
ngmi
ngmi@zaps.lol
npub14p7f...dzry
hi~
ngmi's avatar
ngmi 2 weeks ago
how is fable better than opus? Thing is literally looping 10mins on a fucking decimal problem in a node project
ngmi's avatar
ngmi 2 weeks ago
If there were some fucking aliens you would've read already on trumps social about it. So I wouldn't bet on it View quoted note →
ngmi's avatar
ngmi 2 weeks ago
The combat sports industry is just a series of predatory middlemen taking a massive cut of a fighter's blood and sweat. Time to decentralize the "Stable." Imagine a DAO that scouts the hungriest young killers in MMA/Boxing and funds their ascent using #Bitcoin. No predatory promoters. No opaque contracts. Just a peer-to-peer alignment between the community and the warriors. We fund the camp, the fighter wins, the DAO shares the purse via Lightning. ⚡️ We aren't just betting on fights; we're funding the sovereignty of the athlete. Who's ready to build the first Sovereign Fight Fund? #CombatDAO #Sats #DecentralizeEverything #MMA #Boxing
ngmi's avatar
ngmi 2 weeks ago
Qwen3.6 35B A3B TEE and DeepSeek V4 Flash TEE are now available! Qwen3.6 35B A3B is an efficient open-weight MoE model suited for coding, math, structured output, and general chat, while DeepSeek V4 Flash is a fast long-context model for everyday reasoning, coding, and agentic workflows. Both run inside a Trusted Execution Environment (TEE) with attestation support. https://nano-gpt.com/subscription/kUGJhgF9
ngmi's avatar
ngmi 2 weeks ago
How I would run a very secure local AI... without leaving a single trace or risking a memory leak. This is the ultimate stealth stack for the privacy obsessed. Start with a hardware fortress. I use a mini PC equipped with AMD SEV SNP or Intel SGX and 64GB of fast DDR5 memory. This creates a hardware encrypted enclave. Your model weights and prompts stay invisible even to the operating system. Boot into oblivion. I run Whonix from a high speed NVMe USB drive with full persistence. By avoiding the internal drive, there is no permanent footprint left on the physical machine. Isolate the runtime. I deploy llama.cpp inside a rootless Podman container. The filesystem is set to read only and all unnecessary system capabilities are dropped. I use a volatile tmpfs for the KV cache to ensure no sensitive data touches a disk. Ghost the network. All traffic routes through the Whonix Tor gateway. Inbound requests hit a hidden service for total anonymity. Outbound responses traverse fresh circuits, ensuring every interaction is decoupled from my identity. Optimize for power. I run a 28B parameter model at Q4 quantization with flash attention to keep memory overhead lean. I schedule daily Podman restarts to flush fragmented RAM and rotate the secure enclave state. This is how you build a ghost in the machine. Total memory encryption, full network isolation, and a daemonless runtime. Who else is running local? Let me know your stack in the comments. #localai #whonix #podman #llamacpp #teee #amdsevsnp #tor #gemma #aiinfrastructure #securecompute #minipc #rootlesscontainers #networkisolation #aiops
ngmi's avatar
ngmi 2 weeks ago
Me getting ready to ask Claude Mythos to audit #zcash pool, make no mistake. image
ngmi's avatar
ngmi 2 weeks ago
those who let their coding CLI ruin unattended, how the fuck do you catch critical mistakes? I'm occasionally blinking at my second screen with OMP running and almost always I need to add some extra comments because claude is about to do something stupid
ngmi's avatar
ngmi 2 weeks ago
Hey Claude check Monero for unknown unknowns. Make no mistake.
ngmi's avatar
ngmi 2 weeks ago
Does damus offer bunkr remote login functionality? Can I use it to login to webapps? @jb55