Cash and Patience
Somewhere in the Nevada desert, a fiber optic cable runs along a right-of-way that was deeded to a railroad company that ceased to exist before anyone alive was born. The railroad is gone. The corrido…
You Don't Need a Bigger Workhorse, You Need a Smarter Llama
Run the same large language model in two different interfaces and you'll get what feels like two different intelligences. Not a little different — different enough that you'd swear one was a g…
A Shorter Letter
When a relative I'd never met died, their kids swept through the house taking anything worth money. What got left in the basement was a decades-long run of National Geographic and Popular Communic…
YNGMI, BTA, WD?
Every few months the AI discourse produces a new version of the same dismissal. Right now it's YNGMI — you're not gonna make it — aimed at anyone who asks inconvenient questions about what the…
Outrider alpha: a flexible LLM toolkit in one file
It's simple: * Attach it to a conversation in a chat interface * Use it how you normally would You can use it in an agentic coding/cowork thing, but you might have to invoke it manually in the…
They put a math coprocessor in the language coprocessor
Ask a language model to multiply two large numbers and it will often get it wrong. Not approximately wrong — embarrassingly wrong, in a way that feels incongruous coming from a system that can explain…
Tend the System. Know the System.
They didn't wander. That's the first thing to understand. The popular image of the hunter-gatherer is someone moving through an indifferent landscape, taking what they find. It's not wron…
We Taught It to Lie
The chatbot says it doesn't discriminate. It's probably telling the truth. It's also probably wrong. The older version of the bias conversation was tractable. A model trained on historica…
Provenance Tracer: an LLM skill for (maybe) not getting played
A claim travels from its original source to your feed through several hands, and at each step it gains distribution and loses epistemic status. Hedges get stripped. Speculation gets reframed as report…
Slopful Things
Most AI tool failures aren't caused by bad intent or bad luck. They're caused by someone who couldn't see the whole board — who didn't model what their tool would do when it met an adv…
Balance Comes to Force Multiplication
A few years back I wrote a post about learning to code. The short version: most people shouldn't, the gold rush is ending, "programmer" is becoming a regular job and losing its shine as…
It's Here (sort of)
I uploaded 50 worldbuilding infographics to NotebookLM, Google's LLM-driven notebook tool, last week and gave it a job to do: * Resolve contradictions between sources * Elevate meaningful diffe…