How authorship is made and named: a meta-aggregator / CoAuthor
CoAuthor
https://coauthor.stanford.edu/
empirical comparatorMachine participation in authorshipAttribution, credit and identity infrastructurehistorical
The archive's reading. The paper records co-writing as a dataset of sessions and reads the model's part as contribution under stated definitions of collaboration; the saved abstract keeps no writer's name. The archive reads it as process evidence of distributed contribution held at the session level, with the persons counted (63) and the question of whose name the text bears left outside the record as saved.
| Relation to this census | empirical comparator |
|---|---|
| Record type | dataset |
| The question it answers | contribution: what a language model contributes to a writing session, read from recorded interactions |
| Kind | in the paper's words, 'a dataset designed for revealing GPT-3's capabilities in assisting creative and argumentative writing', with an interface for replaying the writing sessions |
| Operator | Mina Lee, Percy Liang and Qian Yang, authors of the paper that presents the dataset (CHI 2022); the dataset's home page could not be read |
| Evidence | interaction dataset; research paper |
| Unit of record | a writing session: the interactions between one writer and one of four instances of GPT-3 |
| What it records | interactions between 63 writers and four instances of GPT-3 across 1445 sessions of creative and argumentative writing |
| Whose authorship it keeps | not recorded in the saved source: the abstract counts 63 writers and does not say whether or how they are named |
| Under what name | not recorded |
| Who holds it | the paper states the dataset and a replay interface are 'publicly available'; the dataset's home page and terms could not be read |
| Authorial function and responsibility | the paper frames GPT-3's part as a contribution 'as a writing "collaborator" under various definitions of good collaboration', read from recorded interactions; how authorship of a session's text is assigned is not recorded in the saved source |
| Method | curated interaction logging of writers working with GPT-3 instances, analysed for the model's language, ideation and collaboration capabilities (per the abstract) |
| Coverage | geography: not stated in the saved source; period: the paper: v1 18 January 2022, v2 25 January 2022; CHI 2022; the session dates are not stated |
| Count | value: 63 writers, four instances of GPT-3, 1445 writing sessions; as_of: arXiv v2, 25 January 2022 |
| Holdings | the dataset of 1445 writing sessions; an interface for replaying the writing sessions; the CHI 2022 paper |
| Stated limits | none stated in the saved abstract; the paper notes that LM capabilities are 'highly context-dependent' and 'often subjectively interpreted', the problem the dataset addresses |
| Data access | 'publicly available' per the paper; the dataset home page (coauthor.stanford.edu) could not be read |
| Licence | not stated in the saved source |
| Lifecycle | state: historical; detail: a dataset published with a CHI 2022 paper (arXiv v2, 25 January 2022); whether the site and replay interface answer today is unrecorded, since the home page fetch failed on 2026-11-01 |
| Version and time | arxiv_v1: 2022-01-18; arxiv_v2: 2022-01-25; read: 2026-11-01 |
| Founded | 2022 (paper and dataset release) |
What its record can settle
- how GPT-3's contribution to a writing session can be read from recorded interactions, under the paper's definitions of good collaboration
- how many writers, model instances and sessions the dataset holds
What it cannot settle
- who the writers are, or whether they are named
- how a sustained named poetic practice works: the study population and task setting differ
- the dataset's license and current availability (home page unread)
Kept apart
- NOT a record of named authors: the saved abstract counts writers without naming them
- NOT a sustained literary practice: writing sessions in a study setting
- NOT the model's authorship: the paper reads GPT-3 as 'collaborator' under definitions it sets
Evidence
- kind (the resource's own statement):
we present CoAuthor, a dataset designed for revealing GPT-3's capabilities in assisting creative and argumentative writing.
https://arxiv.org/abs/2201.06796 · read 2026-11-01 · http 200 · sha256 540bc42e51f2e67f… - count (the resource's own statement):
CoAuthor captures rich interactions between 63 writers and four instances of GPT-3 across 1445 writing sessions.
https://arxiv.org/abs/2201.06796 · read 2026-11-01 · http 200 · sha256 540bc42e51f2e67f… - function (the resource's own statement):
We demonstrate that CoAuthor can address questions about GPT-3's language, ideation, and collaboration capabilities, and reveal its contribution as a writing "collaborator" under various definitions of good collaboration.
https://arxiv.org/abs/2201.06796 · read 2026-11-01 · http 200 · sha256 540bc42e51f2e67f… - problem (the resource's own statement):
their highly context-dependent capabilities are difficult to grasp and are often subjectively interpreted.
https://arxiv.org/abs/2201.06796 · read 2026-11-01 · http 200 · sha256 540bc42e51f2e67f… - access (the resource's own statement):
The dataset and an interface for replaying the writing sessions are publicly available at this https URL .
https://arxiv.org/abs/2201.06796 · read 2026-11-01 · http 200 · sha256 540bc42e51f2e67f… - authors (the resource's own statement):
Authors: Mina Lee , Percy Liang , Qian Yang
https://arxiv.org/abs/2201.06796 · read 2026-11-01 · http 200 · sha256 540bc42e51f2e67f… - venue (the resource's own statement):
Published as a conference paper at CHI 2022
https://arxiv.org/abs/2201.06796 · read 2026-11-01 · http 200 · sha256 540bc42e51f2e67f… - dates (the resource's own statement):
[Submitted on 18 Jan 2022 ( v1 ), last revised 25 Jan 2022 (this version, v2)]
https://arxiv.org/abs/2201.06796 · read 2026-11-01 · http 200 · sha256 540bc42e51f2e67f…
Unchecked
the dataset home page and replay interface (coauthor--home, coauthor.stanford.edu): the fetch failed with a TLS certificate error (http 000, no text); unread and unquoted; the paper's full text (only the arXiv abstract page was saved): how writers were recruited and paid, whether they are named, the session format and any keystroke-level logging; the dataset's license
Cite: Lee Sharks, ed. "CoAuthor." How authorship is made and named: a meta-aggregator, leesharks.com, compiled 2026-11-01. https://www.leesharks.com/apparatus/aggregators/coauthor