Notes & long reads
Things I worked out by writing them down. Mostly about NLP, occasionally about the job of explaining any of it.
Most of what is here started as something I did not understand well enough. I read about it, tried it, got it wrong, and then wrote it down until the explanation held together. If a post reads clearly, that is usually because the confusion came first.
I write about natural language processing because that is where my day goes: tokenisers, embeddings, models that behave beautifully in a notebook and badly in production. But the explaining interests me as much as the thing being explained. A technical idea that only makes sense to people who already understand it has not really been explained.
So I try to stay concrete. Worked examples ahead of analogies where I can manage it, and an honest note where something is still unclear to me. I would rather leave a question open than close it with a sentence that sounds confident and means nothing.
Some posts are long, some are one idea I could not shake. Take whichever is useful and skip the rest. 馃檪
Showing 5 of 5
A Taoist thought experiment arguing that a life is closed from outside but open from inside, finite yet edgeless, and that the afterlife question dissolves rather than gets answered, since there is no "outside" of the Tao to go to and no passenger to make the trip.
Read itA rigorous proof that nothing is good enough to describe anything at all, conducted entirely with the tools it disproves. Four equations, one unknown, and a growing suspicion that the thing writing this was never on the list.
Read itFor years my portfolio said "Python 85%". I finally asked myself what the other 15% was, and could not answer.
Read itBefore a language model sees a single word, something has already made hundreds of decisions on its behalf. Most of them are invisible until they break.
Read itI write documentation in English and 啶啶曕啶む in Nepali. They have turned out to be very different kinds of thinking.
Read it
