I Trained My Own Vision Model to Read Mail Out Loud. It Almost Learned the Wrong Lesson.
Sep 5
I Added Day-One Muse Glimmer Support to Apple MLX-LM
Aug 11
I Took Down Six of My Own Benchmark Videos. Then I Built the Local Agent Leaderboard.
Aug 10
My local AI was pausing 7 seconds before every reply. It turned out to be one cache bug.
Aug 7🧠Local AI & Open Models
Open-weight models, MLX ports, quantizing, and benchmarks on Apple Silicon.I Trained My Own Vision Model to Read Mail Out Loud. It Almost Learned the Wrong Lesson.
You point the phone at a piece of mail and press one button. A second later it says one sentence out loud: what the…
Sep 5I Added Day-One Muse Glimmer Support to Apple MLX-LM
here is the thing nobody tells you about adding a new model to an engine like this. it is not magic and it is…
Aug 11I Took Down Six of My Own Benchmark Videos. Then I Built the Local Agent Leaderboard.
My benchmarks are filesystem-judged: the model does real work in a sandbox — create the file, fix the failing test, rename a function across…
Aug 10
NVIDIA Shipped a Model That Sees and Hears — It Just Didn’t Run on a Mac. So I Wrote the Missing Piece.
NVIDIA released Nemotron-3-Nano-Omni-30B-A3B — a "tri-modal" model, which is a fancy way of saying one brain with eyes and ears attached. You can hand…
Jul 16
Why I Quantize Open-Weight Models for Macs — And Why Your Law Firm Should Care
I publish a few MLX quantizations a month under huggingface.co/divinetribe to close that gap. Two of mine have crossed 1,000 downloads in the last…
May 24
Hermes-4-14B Abliterated, MLX 4-bit — Apple Silicon Just Got Another Real Model
Hermes 4 is NousResearch’s instruction-tuned model family. The 14B variant is built on Qwen3, so it inherits Qwen3’s tokenizer, chat template, and architectural quirks…
May 24
HumanEval on a MacBook — 81.7% pass@1, Wi-Fi off
The Qwen team didn't publish HumanEval scores for any Qwen3-Coder variant. They consider the benchmark saturated — and for cloud-served frontier models, fair, it…
Apr 29💻AI Coding
Running coding agents locally — costs, setups, and war stories.My local AI was pausing 7 seconds before every reply. It turned out to be one cache bug.
I maintain claude-code-local, a repo for running coding agents against local models on a Mac. No cloud, no API key. The original approach pointed…
Aug 7
M5 Max + 128GB = a 30B AI Coding Agent Running Locally. Wi-Fi Off.
Until now, running a 30B+ parameter model meant a GPU rack — or paying a cloud API per token. The M5 Max changes that:
Apr 29
Pulling 10x My Subscription Value Out of Claude — While Quietly Building the Backup Plan
Every Sunday night I watch the meter tick toward 100% again. That's been the rhythm for months — five days of heavy work, one…
Apr 26
Cloud AI Coding Costs Keep Climbing — How to Pay $0 and Still Use Claude Code
I got tired of watching that number climb so I built a way around it. Claude Code still works great. You just run it…
Apr 22
Three Generations of Running Claude Code Locally on a MacBook — What I Actually Learned
The project is open source — it's at github.com/nicedreamzapp/claude-code-local if you want to skip the story and just run it.
Apr 22🌅Ambient Computing
Computing without the screen: voice, agents, and the post-desk workflow.
Two AI Agents, One Browser Window: What Actually Breaks, and What Doesn’t
I was posting a technical write-up to a few places. Three things went wrong, and none of them looked related.
Aug 7
A Field Guide to Ambient Computing — The Words for the Thing That’s Coming
https://youtu.be/PLbV4QtFmFY Watch: My Mac mini at home is the AI. I just talk to it from any browser. (50 sec)
Apr 22
What It’s Actually Like to Code By Voice — With the AI Replying In My Own Cloned Voice
A compiled Swift binary wraps Apple's SFSpeechRecognizer — the same engine that powers macOS dictation — in a continuous-listening daemon. It transcribes everything I…
Apr 22
The Era of Hunched-Over-A-Screen Computing Is Ending — Here’s What’s Replacing It
Sitting in a hot tub a few weeks ago, I sent a text from my phone: "find me the best rated electric guitar at…
Apr 22
This Is What a Robot Can See Now — 601 Objects, Live, Offline, on Your iPhone
That's the part I keep coming back to. We've quietly crossed a line where a machine running on the 6-ounce thing in your pocket…
Apr 22🔒Private AI
Confidential AI for law, medicine, and anyone who can't send data to the cloud.
A Federal Order Switched Off Anthropic’s Best AI Overnight — and Made the Case for Private AI
Update: The fuller, verified story is here: Anthropic's biggest backer just killed its best model.
Jun 13
The Heppner Ruling, Warner v. Gilbarco, and What Confidential AI Actually Has to Mean
This article describes a recent federal court ruling and what it implies for confidential AI use. It is not legal advice. Talk to your…
May 22
Your Medical Practice Is Probably Using Cloud AI on PHI Right Now — Here’s the HIPAA Problem Nobody Is Talking About
For years, the assumption was that nobody on staff would use a "ChatGPT" on actual patient data — it'd be obvious that PHI shouldn't…
Apr 22
If Your Law Firm Is Using Cloud AI on Client Files, You Probably Have a Problem
There is another way to run AI tools inside a firm — locally, on the machines the firm already owns, with no frame of…
Apr 22📰AI News & Commentary
What's happening in AI and what it means.
Why No PC Laptop Can Touch a 128 GB MacBook for Local AI Yet, and the Three Chips Coming for It
I do all of my AI work on a MacBook Pro with an M5 Max and 128 GB of memory. Songs, videos, vision models…
Sep 5Anthropic’s Biggest Backer Just Killed Its Best Model: The Real Story Behind the Fable 5 Shutdown
Anthropic launched Claude Fable 5 and Mythos 5 on June 9, the most capable models they had ever shipped. Three days later, on Friday…
Jun 15
504,571 Brain Cells, 4 Labs, One Hypothesis: A Citizen Pass at Parkinson’s
The short version: when you pool 504,571 single-cell measurements of human midbrain tissue from four different research groups, one specific subtype of dopamine neuron…
May 24
I Just Watched One Hacker Catch Up to a Trillion-Dollar Data Center
Today I gave the same prompt to three different AI engines. The same prompt, on the same MacBook:
May 10