Notes
The notes answer specific texts and moments in the discourse around large language models and writing.
A measurement study asks who drives the story when human and language model write in turns. It arrives at the amplifier — and the declaration still does not rest on it.
OpenAI publishes solutions to ten open problems in mathematics; decades of search time, priced at two thousand dollars. The substrate's contribution resolves into the dimension of time — Hartmut Rosa's diagnosis of acceleration is confirmed at its youngest link, and a rest stays outside.
A study measures the flood of machine-generated books on Amazon. It documents the dilution of the market — and cannot measure what declaration would change, because there is none to be found.
The border between human and AI is sought in the material of speech. It lies in its mode — and calculation is the part that can be handed over entirely.
Why a metaphor holds although nothing in the world connects heart and fire — and why AI knows the net of language but cuts no tears into it.
Article 50 of the EU AI Act has applied since 2 August 2026. It requires labelling for deep fakes and for text of public interest. Of the novel it requires nothing.
AI language models score above the human average for creativity, a study finds — and hit a ceiling. On combinatorial versus generative novelty.
Anthropic's AI Fluency framework demands honesty about the machine's role. A confirmation from an unexpected direction — and the word that does not hold.
Why Maschinenschrift refuses the AI detector. A note on the papal encyclical, the Snopes verdict, and what detection cannot establish.
Why Maschinenschrift says language substrate and not AI, LLM, or assistant. The word grants the machine no intelligence it does not have.