I live in London and keep two notebooks: one for things a model did, and one for things people claimed it could do. The distance between those notebooks became the starting point for Model Dissent. My work spans internal reporting tools, small evaluation projects and the sort of spreadsheet cleanup nobody puts on a conference stage.
I write about AI benchmarks, changing work habits and the difference between a useful demonstration and a conclusion that survives questioning. I like an experiment with a clearly described task, an honest failure log and enough context for someone else to disagree intelligently. I get particularly impatient when a result from one model, one month or one team becomes a universal prediction about everybody's future.
Outside the screen I enjoy secondhand science books, late cinema screenings and walking home through the city without listening to another product announcement. My lavender-haired anime avatar reflects the playful side of this notebook. In the writing underneath it, I am skeptical, curious and willing to revise a position when the evidence changes.
I keep encountering an AI productivity number that has escaped its original date, task and population. It wanders around the internet announcing that the argument is over. Sometimes it says AI makes everyone spectacularly faster.
The prediction arrives wearing the expression of an accomplished fact: developers are finished, juniors are especially finished, and anyone still learning is rearranging deckchairs. It is an excellent sentence for a feed. It is a
The smallest model in the comparison is an elderly bicycle. The cloud model is a new motorbike. The reviewer announces that wheels have been settled forever. Everyone in the comments then starts arguing about whether