20× Less waiting per turn 1024 Gemma’s sliding window, in tokens 12/12 Eval tasks passed on a 550-token prompt 🧰The
My local AI was pausing 7 seconds before every reply. It turned out to be one cache bug. Read Post »
Two Claude Code agents doing browser automation against one Brave window quietly sabotaged each other for an hour while I blamed LinkedIn. What was actually wrong with the computer-use setup, the fix that already shipped, and a thirty-line test so you can check your own.
Two AI Agents, One Browser Window: What Actually Breaks, and What Doesn’t Read Post »
NVIDIA gave away a 30B model that sees, hears, and reasons. The catch: only its text half ran on Apple Silicon. I ported the vision and audio towers to MLX, verified them against NVIDIA’s own code, and open-sourced it.
120,000 Characters of system prompt leaked 90 min Ultimatum to comply 3 Days from launch to blackout ⏱️The two-minute version
Three days after launch, the U.S. government ordered Anthropic to disable Fable 5 and Mythos 5 worldwide. Nobody running local AI lost a thing. Here’s the lesson.
Frontier cloud LLMs log your prompts. For privileged content (NDAs, medical records, source protection) that’s a deal-breaker. Open weights on Apple Silicon are the only honest answer, and there’s real, sustained demand from firms that nobody is talking about.
Why I Quantize Open-Weight Models for Macs, and Why Your Law Firm Should Care Read Post »
I’m not a neuroscientist. But I pooled 504,571 single-cell measurements from four research groups and ran the meta-analysis. AGTR1+ dopamine neurons are depleted in every Parkinson’s cohort, combined p < 10^-100. The receptor is targetable by FDA-approved blood-pressure drugs.
504,571 Brain Cells, 4 Labs, One Hypothesis: A Citizen Pass at Parkinson’s Read Post »
I converted Babsie’s BF16 abliterated Hermes-4-14B to MLX 4-bit so Mac developers don’t have to wait. 7 GB on disk, runs on a 16 GB Mac, drop-in for any mlx-lm or claude-code-local workflow.
Hermes-4-14B Abliterated, MLX 4-bit: Apple Silicon Just Got Another Real Model Read Post »
A federal judge ruled documents prepared via public AI lose attorney-client privilege. A second court the same month preserved work-product for internal AI use. Here is what the split actually means, with cited sources.
The Heppner Ruling, Warner v. Gilbarco, and What Confidential AI Actually Has to Mean Read Post »
Antirez (the Redis guy) shipped ds4, a hand-tuned C inference engine for DeepSeek V4 Flash. Same prompt, three engines, one MacBook. Local DeepSeek beat cloud Claude on wall-clock time.
I Just Watched One Hacker Catch Up to a Trillion-Dollar Data Center Read Post »
