Opinion
AI made taste the skill people pay for
The more work machines produce, the more people want to know a person shaped it.
5 minAI
Everything we publish on AI: the models we run on our own machines, what the tools send home, the benchmarks we measure ourselves and the claims we check against the code. Start with the investigations, then the guides.
Guides
The free route keeps your meeting recordings at home
A 7 minute meeting transcribed offline with the network blocked. Names and speakers are where it slipped.
7 min
Guides
Free AI can answer your customer emails. Don’t leave it alone
Gemma 4 got 25 of 30 replies right in our shop test, and told one customer an address had changed when it had not.
7 min
Guides
Qwen3.6 on MLX returns an empty reply
Structured output from LM Studio came back blank five times out of five; the JSON sat in reasoning_content. Three ways round it.
8 min
Guides
Two agents, one Mac: MLX serves both at full speed
Two callers on Gemma 4 31B got 11.5 tokens a second each on MLX and 5.7 each on llama.cpp.
12 min
Investigations
MLX beats GGUF on a Mac, just
Gemma 4 31B: 12.6 tokens a second against 11.7 on our M4 Pro.
7 min
Investigations
LM Studio’s MLX engine reads prompts twice as fast
Qwen3.6 35B-A3B took in 20,000 tokens in 30 seconds on MLX and 75 on llama.cpp.
12 min
Insights
What Anthropic’s usage data actually holds
We opened the 86 files behind the Economic Index. One row is a metric value, never a chat.
7 min
Explained
Why does a model forget what you told it?
Continual learning, and why no deployed model has it yet.
7 min