A language model trained only upon the written world of 1700–1938.
The weights remember what the words once meant.
The attention turns where the old authors turned.
What the model returns is not creation —
it is the archive, compressed and answerable.
Behind this page is a language model trained on the written world of 1700–1938 — fifty-four billion tokens of books, pamphlets, parliamentary record, and scientific prose. It carries the knowledge and the blind spots of its era in equal measure. It has read one world war into its weights but cannot see the next. What you ask it is up to you. What it answers is drawn from the record alone.
Knowledge cutoff: 1938 · 7.24B parameters · trained on 54B historical tokens
History means inquiry — conduct it yourself.
A joint undertaking of UCL and Oxford — London, 1938