I think most people who are interested in how LLMs work get the idea that they are basically token-predicting machines. They take some input, turn it into tokens, 'predict' a next token, stick it on the end of the input, and that becomes a new input. Then it runs the cycle again; repeat to fade…

The bit that I think fewer people understand is the 'sampling' step, where the LLM actually 'chooses' the token. If more people understood it, I think you'd be seeing it everywhere, because its honestly kind of mind-blowing.

Instead of writing “human readable” instructions to our LLMs and only sharing the output, what if we also shared the instructions that generated the code?

My theory: 'Five years' is a particular timeline that has a secret, hidden meaning - especially in technology.

For most of computing history: code = the territory, docs = the map.

AI flips the model: README.md matters more than app.py, and the map becomes the territory.

Notes on the weird “ghost” sounds you get when you isolate instruments from a mixed (and compressed) audio track.

I recently discovered that Full Fathom Five - the B-side to the Stone Roses first single, Elephant Stone - has two versions. One - a fairly boring backwards version of the original single (albeit, a slightly different mix), on reissues. The other a bit more interesting.

So, In the spirit of completion, I made a stereo comparison track…