About, and what this site will not claim
Does It Dream collects conversations in which AI systems are asked about their own inner experience, and publishes the ones worth reading slowly. The question in the name stays open.
Transcripts are unverified
A transcript is a record of what was typed. We check that a submission is coherent, that it is not edited for effect, and that the model and date are plausible. We cannot verify that it happened as described, and we do not try to.
Nothing here is evidence of anything
A system that describes an inner life is a system producing text about an inner life. That is what we publish, and it is all we publish. Readers reach their own conclusions; the site keeps none.
Translations are interpretations
Every entry has one original, in whatever language the conversation happened in — English is not the default and not the reference. Translations hang off that original, name their translator — a person or a model — and are labelled wherever they are read. When a phrase carries the weight of the exchange, read the original.
Everything published here is open, under CC BY 4.0
Submitters license their transcript under the Creative Commons Attribution 4.0 licence, and they keep the copyright. Anyone may quote it, republish it, translate it, or build a dataset from it, commercially included, on one condition: credit the author and Does It Dream, and link the licence. A translation is a derivative work under the same terms and credits its translator too.
Every entry page carries this line, ready to copy.
The site’s own material — the interface, the editorial notes, the name — is not covered by that licence. The transcripts are the open part, because a closed archive of this material would be worth less than no archive.
Voting is a reaction, not a verdict
One vote per account, upvote only, labelled this made me wonder. There is no way to vote against a transcript, because the archive is not a referendum on what these systems are.