Sources, not vibes
Probably about provenance, uncertainty, and why “the model said so” is not a method, even when the sentence is pretty.
Shorter texts on AI, archives, cultural heritage, provenance and noise: opinions, research notes and unfinished arguments that are not papers (yet).
Index
Posts, notes, opinions
Essay
LLMs in archives: fluent, useful, suspicious
Large language models can help read damaged collections, but a fluent answer is not automatically proof. Useful, yes. Innocent, absolutely not.
Essay
Documents are witnesses, not just data
Receipts, insurance claims, page layouts, outbreak reports and Wikipedia revisions: why AI systems should keep the trace between a claim and where it came from.
Essay
Noise is not always a bug
OCR, period spelling and normalisation: why cleaning historical text can also erase what matters.
Currently open
Read now
Longer pieces are added one at a time.
Personal
Personal texts
A story stitched together from personal writings, notes, diary pages and short pieces from over the years, some more Kafkaesque than others.