Serge Parel

@spparel.bsky.social

Cheminformatics • Research IT • Data Science - All opinions are strictly my own !

1/ Three real machines from the past 90 days: a $1,740 garage rig on a decade-old server board serving a 30B coder at full speed, a MacBook running a 397B model streamed from SSD, and eight H200s running MIT-licensed weights that beat Claude Opus 4.8 on Terminal-Bench.

1/ Your GPU is mostly waiting. A 70B model at 4 bits is ~35 GB of weights, all read per token. At 128 GB/s of bandwidth that caps you at 3.5 tok/s. The teraflops never enter the equation. This is the memory wall, and every 2026 inference speedup is a way around it.

Bild

1/ Everyone's chasing bigger models. DeepSeek shipped something quieter that might matter more: DSpark. No new flagship, no reasoning breakthrough. Just their existing V4 stack running 60–85% faster, with zero quality loss. How it works 🧵

First, the White House doesn't date back to July 4, 1776, idiot. It opened on November 1, 1800. Second, the tackiest man in the United States and its president should *not* be the same person. Third, nice inspiration, fascist.

BildBild

1/8 Earlier this year you could buy a Mac Studio with 512GB of unified memory. Today the top config maxes out at 96GB. The high-memory machines didn't just get pricier. They quietly disappeared. Here's what's happening with the 2026 memory crisis 🧵

1/ 5:21 PM Eastern, Friday June 12. The US government picked up a phone and shut down the most capable AI model anyone had ever used. Globally. For everyone. Including the foreign-national researchers who built it. I spent the weekend digging in. 🧵

Bild

1/ Uber burned its entire 2026 AI budget by April. ServiceNow did the same. KPMG says U.S. orgs project $207M average AI spend over the next 12 months — nearly 2x last year. The all-you-can-eat era of AI coding tools is over. The deadline matters.

Bild

Agent “skills” may be the next serious optimization layer. SkillOpt treats a reusable best_skill.md not as static documentation, but as something you can train. Frozen agent runs tasks → optimizer reviews traces → proposes bounded skill edits → held-out eval accepts only improvements.

Festus, Missouri just removed every council member who approved a $6B AI data center. All four. Gone. 70% of Americans now oppose data centers in their local area. That's already past the peak opposition to nuclear power plants. Ever. The tech industry calls it NIMBYism. They're misreading it.

Anthropic leaked Claude Code source to npm. KAIROS (unreleased autonomous agent), "Undercover mode" (AI hides it's AI), and irony that their own AI-written toolchain caused it. "A significant portion was probably written by the AI you're shipping."

Me: "I'll just try @OpenClaw for a quick task" *48 hours later:* My AI has extracted 46 plant care guides from OneNote, filed a bug report on Moltbook, and is tracking my bloodwork trends. I have created a monster. A very organized monster. ????