MMartin Bowling·Oct 27The Celebrity Neuron Is Not a NeuronI tried to make Qwen3.5-4B identify as Einstein, Curie, Newton, Taylor Swift, and Keanu Reeves. The scientists worked. The entertainers didn’t.5 min read·martinbowling.com
MMartin Bowling·Oct 26I Audited the Viral Qwen Neuron ZooI tested the viral one-neuron personas on Qwen3.5-4B. 16 of 22 survived a one-digit fix, one was impossible, and three neurons become the Golden Gate Bridge.10 min read·martinbowling.com
MMartin Bowling·Oct 26Three-Neuron BangersFive combos that fuse into something none of the neurons does alone — plus two cursed loops and one random control that does nothing.5 min read·martinbowling.com
HHaithem Slimi·Oct 25Teaching a Small Local Model to Think in DifferentialsA DDXPlus and QLoRA experiment on Apple Silicon, from held-out evaluation to oMLX deployment31 min read·nightthoughts.me
MMihai_LeanZero·Oct 16Best Local Image Generation Model on Mac: ERNIE vs FLUXThe best local image generation model on a Mac, measured: six open models blind-scored and timed on an M3 Ultra, plus which licenses allow commercial use.29 min read·leanzero.hashnode.dev
HHaithem Slimi·Oct 9Complete Guide: Always-On Mac Server with MLX, oMLX & QwenPawHeadless MLX inference with oMLX, QwenPaw, Caddy, Tailscale, and Cloudflare — verified end to end.50 min read·nightthoughts.me1
DDragos Roua·Oct 3[ Video ] How To Set Up Local AI on a 16GB MacBookSetting up local AI on a 16GB MacBook involves using optimized binaries for the Metal architecture. Start with MLX-serve. Optionally OpenWebUI4 min read·dragosroua.hashnode.dev
MMihai_LeanZero·Sep 23Goose 3.0: Local Models, MLX and LeanZero LinkGoose Swarm 3.0 runs local models on its own MLX engine, joins your Macs with LeanZero Link and serves an OpenAI-compatible goose endpoint.26 min read·leanzero.hashnode.dev
MMihai_LeanZero·Sep 19Qwen3-Coder-Next + MXFP8: The 128GB Local LLM That Runs PredictablyThe MXFP8 MLX quant of Qwen3-Coder-Next (80B total, 3B active, 262K context) on a 128GB Mac: LM Studio 0.4.1 setup for Claude Code.34 min read·leanzero.hashnode.dev2B
MMihai_LeanZero·Sep 19MLX vs GGUF on Apple Silicon: Benchmarking the Same Local Model Two WaysOne 27B model as GGUF and MLX, 30 app builds graded by running them. Each worked 13 of 15, speed was a wash, and MLX was steadier: the safe default.13 min read·leanzero.hashnode.dev
