Reinventing the wheel to sound smart, the Musky way View quoted note →

Why does #ditto not rich format markdown? Weird.

# Not all linen is created equal 🧵 Ever wondered why two products both labeled "100% linen" can have wildly different price tags? The label tells you almost nothing. **The fiber matters most.** Premium linen uses "line fibers"—long, continuous strands up to a meter long that produce smooth, low-shed fabric. Cheap linen uses "toe," the short broken scraps left over from processing, or even "cottonized flax" deliberately chopped short to run on cheaper cotton machinery. **Geography is destiny.** 75–80% of quality flax grows in a narrow coastal strip through France, Belgium, and the Netherlands. That climate enables "dew retting"—a slow, natural process where field moisture and microbes dissolve the plant's internal glue without chemicals. Cheap flax skips this for hot water or chemical retting: faster, but harsher on durability. **The softness trap.** Real linen comes off the loom feeling crisp and softens naturally over years of washing. Factories selling "instantly soft" sheets often bathe them in cellulase enzymes or stonewash them—literally eating away the fiber surface. You're paying full price for a product that's already lost half its lifespan. **What to check when shopping:** - GSM between 170–230 for bedding (lower = flimsy, tears at seams) - Brand transparency about GSM and geographic origin - If those details are missing, walk away Genuine linen softness is earned, not manufactured. #linen #textiles #qualitymatters #craftsmanship #consumerawareness #slowliving #naturalmaterials #flax #bedding #manufacturing

GLM 5.2 Included in Subscription GLM 5.2 and GLM 5.2 Thinking are now available and included in the NanoGPT subscription at 2x included-input usage. GLM 5.2 is Z.AI's latest flagship model for long-horizon autonomous tasks, built to plan, execute, iterate, and optimize complex coding and engineering work over extended runs. It is especially strong for autonomous agents, complex engineering optimization, and real-world development scenarios; use the regular version for faster direct answers or the Thinking variant for harder reasoning-heavy work. https://nano-gpt.com/subscription/kUGJhgF9 #glm #nanogpt

The biggest issue with using local models in omp is the 30+k system and tools prompt. So you kinda have to run a big context window from the get go. Makes gemma 12b pretty slow. I reverted back to just have it write a Python tool+skill for gemma 4 uncensored for specific tasks 🤔 View quoted note →

Opus and me are helping to improve #omp for local models

GitHub

Empty finish_reason:length completions retried 3x and only warn-logged (local ollama) · Issue #2774 · can1357/oh-my-pi

What happened With the smol tier pointed at a local ollama model, omp produces no visible output. The only trace is a warn-level log line ("Assista...

issue closed and fixed already!

Testing now OBLITERATUS/Gemma-4-12B-OBLITERATED via ollama/webui The first abliterated model to achieve zero refusal with zero benchmark regression versus stock weights. #gemma #obliterated #ai

The First Domino Has Fallen...

Wait #spain is playing to 0 against a country I can't even pin on a map? Polymarket on life support rn 🤣 #worldcup

Kimi K2.7 Code is now included in the NanoGPT subscription at 2x included-input usage. It is a coding-focused agentic model for long-horizon software engineering workflows, with native image input, tool calling, and strong performance on complex coding tasks. https://nano-gpt.com/subscription/k7AXDwBb #ai #nanogpt

Third-party apps can now add Sign in with NanoGPT. Developers can use OAuth PKCE to let users approve access in NanoGPT and receive a dedicated API key for local apps, coding agents, chat frontends, and other OpenAI-compatible clients. https://nano-gpt.com/subscription/k7AXDwBb #ai #nanogpt

Cool? View quoted note →

I think #zai is pretty good, and they have cool tools, but they do such a bad job at presenting them. Best mobile app by far is from #kimi imo

Scored F (25%) for AI agent-readiness on metaend Grade. Scan your site →

switchboardgo.com scored F on AI agent-readiness

25% agent-ready on metaend Grade. Scan your site.

Might want to make this more accessible for agents. 👋 View quoted note →

Scored F (25%) for AI agent-readiness on metaend Grade. Scan your site →

lnvps.net scored F on AI agent-readiness

25% agent-ready on metaend Grade. Scan your site.

@Jessica @ LNVPS

@Jessica @ LNVPS can I use cashu with your service too?

TIL: there's firmware (MeshCore-BitChat by jooray) that turns a Heltec into a bridge between bitchat's BLE world and a LoRa mesh. #bitchat #meshcore

@Mullvad VPN is this true? for Mullvad (its .deb postinst calls systemctl and fails to install on Devuan

Moving to systemd free #Devuan soon. Initially I had #artix in mind, but the ecosystem is such a cluster fuck... https://www.phoronix.com/news/Arch-Linux-AUR-More-Than-1500

BREAKING: NEW CLAUDE PLAN /s #Claude