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.
Serge Parel
@spparel.bsky.social
Cheminformatics • Research IT • Data Science - All opinions are strictly my own !
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.
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.
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. 🧵
A study finds that an mRNA vaccine is highly effective at preventing recurrence of this dangerous skin cancer, when used in combination with Keytruda, an immunotherapy drug. n.pr/4u3FHCZ
A cancer vaccine made just for you. mRNA is back and it's fighting melanoma
A study finds that an mRNA vaccine is highly effective at preventing recurrence of this dangerous skin cancer, when used in combination with Keytruda, an immunotherapy drug.
n.pr
Reaction prediction models are only as good as the reaction data they learn from. This paper is worth a look: CompleteRXN: Toward Completing Open Chemical Reaction Databases 1/5 arxiv.org/html/2605.00...
CompleteRXN: Toward Completing Open Chemical Reaction Databases
arxiv.org
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.
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.
102 organic chemists shown real vs AI-generated molecules. Couldn't tell them apart — 62% accuracy ≈ random. That's CoCoGraph: graph diffusion with 100% chemical validity, 534K params vs 4.6M for comparable models. Constraints in the math, not the model. www.nature.com/articles/s42...
A collaborative constrained graph diffusion model for the generation of realistic synthetic molecules - Nature Machine Intelligence
The collaborative constrained graph diffusion model CoCoGraph generates novel molecules that are guaranteed to be valid and more realistic than state-of-the-art outputs, while achieving faster perform...
nature.com
K-Dense released 134 scientific skills for AI coding agents, incl. #rdkit, molecular property prediction, VS, ADMET analysis, molecular docking. Open source, works with Claude Code/Cursor/Codex… github.com/K-Dense-AI/claude-scientific-skills #cheminformatics #RDKit #AIagents #drugdiscovery
GitHub - K-Dense-AI/claude-scientific-skills: A set of ready to use Agent Skills for research, science, engineering, analysis, finance and writing.
A set of ready to use Agent Skills for research, science, engineering, analysis, finance and writing. - K-Dense-AI/claude-scientific-skills
github.com
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."
Billionaire Howard Marks questioned if AI was a bubble. Then he watched it build itself. Level 3 arrived: AI gets goals, not instructions. Your survival: Pivot to novelty, cultivate judgment, become the orchestrator. #AI #AutonomousAgents #TechInnovation medium.com/activated-th...
The Trillion-Dollar AI Illusion: A Billionaire Investor’s Warning on the Autonomous Agent Era
Legendary investor Howard Marks questioned if AI was a financial bubble. Then he watched it build itself. Here is why the tech market and…
medium.com
"...if you’re typing natural language prompts into Claude Code 🧑💻 and hoping for the best, you’re rolling dice too. Except the dungeon master is an AI with zero context and infinite confidence." #AI #PromptEngineering #Coding #VibeCoding #ClaudeCode #Opencode medium.com/p/i-stopped-...
I Stopped Vibe Coding and Started “Prompt Contracts” — Claude Code Went From Gambling to Shipping
Last Tuesday at 2 AM, I deleted 2,400 lines of code that Claude Code had just generated for me.
medium.com
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. ????
Wenn du Dinge, die Donald Trump macht, „sehr komplex“ findest, dann möchte ich dir nicht beim Schuhebinden zuschauen müssen.
On a scale of 1 to “invading Russia in winter”, how stupid was this idea ? #neom #TheLine www.heise.de/en/news/Mega...
Megaproject The Line has likely failed
Construction on the 170-kilometer-long city of The Line has stopped. However, the megaproject is still being pursued by officials.
heise.de
Meanwhile in Russia… “The naked arsonist who set fire to the canopy of a tuberculosis dispensary turned out to be a patient with a criminal past.” What a headline www.fontanka.ru/2025/10/19/7...