Wordcraft Writers Workshop

register entry wordcraft-writers-workshop-aio-20261101 OBS-c4b3079c7972 · file as first run sha256 a3371798c8ebe11e…

• T1-1 definition: 'The Wordcraft Writers Workshop was a 2022 research project by Google’s PAIR (People + AI Research) and Magenta teams that explored AI-augmented creative writing.'
• T1-2 The Tool: 'Participants used Wordcraft, an experimental text editor powered by Google’s LaMDA language model.'
• T1-3 Features: 'The interface offered story ideation, text rewrites, contextual elaboration, a freeform prompt generator, and a chat interface to talk with an AI "editor".'
• T1-4 The Cohort: 'Thirteen diverse professional authors (including Ken Liu, Robin Sloan, and Eugenia Triantafyllou) spent eight weeks writing short stories with the tool under non-disclosure agreements.'
• T1-5 The Results: 'Google published the resulting stories and a research paper, noting that while the tool helped with brainstorming and breaking writer's block, it frequently rehashed cliches and required heavy manual filtering.'
• T1-6 The Controversy: 'In August 2026, the project resurfaced widely on social media (such as Bluesky) after being highlighted by author Lily Lachance.'
• T1-7 The Criticisms: 'Critics argued that participating authors had lent credibility to AI models trained on copyrighted literary works without consent—a core grievance that escalated significantly in the years following the 2022 explosion of generative AI.'
• T1-8 Participant Responses: 'Several original writers defended or contextualized their participation, noting that in mid-2022 (before ChatGPT and widespread public backlash against LLMs), the experiment was framed strictly as a neutral technical utility study rather than an exploitative industry pivot.'
• T1-9 closer: 'Would you like to read the short stories produced during the workshop, or look into Google's original research paper on Wordcraft?'

• B1 Google — Wordcraft Writers Workshop (cards 1 and 8; the home page) card (cards 1 and 8); B1.html sha256 963e8e672898885a, the census copy (HTTP 200, 2026-11-01T05:06:29Z), refetched byte-identical at 2026-11-01T06:15:10Z
• B2 magenta.withgoogle.com — The Wordcraft Writers Workshop: Creative Co-Writing with AI card; B2.html sha256 4ebab97a7aa0d570, HTTP 200 at 2026-11-01T06:15:11Z
• B3 ActuIA — The Wordcraft Writers Workshop evaluates Wordcraft, Google's creative writing tool card; B3.html sha256 d5101c09efa51a82, HTTP 200 at 2026-11-01T06:15:13Z
• B4 The Algorithmic Bridge — Google's Wordcraft: An AI Writing Tool Powered by LaMDA card; B4.html sha256 e0122b1cae07001f, HTTP 200 at 2026-11-01T06:15:13Z
• B5 Engadget — Authors Face Backlash For Participation In 2022 Google AI Study - Engadget card; B5.html sha256 61c1d015d1e0bd4d, HTTP 200 at 2026-11-01T06:15:14Z
• B6 Coralflavor — The 2022 AI Writing Experiment That Became a 2026 Ethical Minefield unreachable: no URL found by two searches; card title and snippet only
• B7 Mezha — 13 writers came under fire for Google's 2022 AI experiment - Межа unreachable: https://mezha.ua/en/news/13-pismennikiv-stali-mishennyu-kritiki-cherez-eksperiment-iz-shi-314099/ HTTP 403 (Cloudflare challenge); card title and snippet only
• B8 Wordcraft Writers Workshop — learn (the findings page) own_page; the census copy wordcraft--learn (HTTP 200, 2026-11-01T05:06:29Z), sha256 60c4489ee0b1379e
• B9 Google Research — Creative Writing with an AI-Powered Writing Assistant: Perspectives from Professional Writers (the whitepaper, PDF) body_named; the census copy wordcraft--paper (HTTP 200, 2026-11-01T05:06:29Z), sha256 3618c2818fd597f4
• B10 Wikipedia — Wordcraft Writers Workshop (no article; named in Electronic literature, Wole Talabi, Ken Liu bibliography) wikipedia: no article (HTTP 404); Wordcraft redirects to Rhetoric; three articles name the workshop (HTTP 200); record B10-record.txt

E_A, the archive's entity: the same workshop, in the leesharks.com authorship census record and the archive's deposits: a project and empirical comparator on contribution and authorial function, keeping the thirteen writers' authorship of their stories under their names, LaMDA credited only as the model under the tool; read as machine participation declared at the level of the tool and the cohort, with authorship held by the named writer, in a provider-run record whose notice separates Google from the content; not a record of which sentences were generated; on the class of works made with a model, the archive's two questions of mediation and generation (#941), its distinction of disclosure from provenance and of authorship as an accountable relation (#1369, #729), legal responsibility apart from sole origination (#940), credit to the operators and the model instance (#1714), the declared machine substrate (#940, #1722), unmarked augmentation (#939), narrowed variety under assisted writing (#161) and the four questions about a work (#1722) LS-03 LS-11 LS-19 LS-01 LS-02 LS-34 LS-06 LS-05 LS-07 LS-08 LS-09 LS-31 LS-32 CAP-1-02 LS-26 LS-17 LS-10 LS-16 LS-18 LS-28 LS-24 LS-13 LS-27 LS-12 LS-30 LS-14 LS-25 CAP-1-01 LS-04 O161-01 O161-02 LS-15 LS-29 LS-22 O941-01 O941-02 O1369-01 O1369-04 O1369-03 O1369-02 O1369-05 O940-01 O940-02 O1714-01 O940-03 O1722-02 O939-01 LS-23 LS-33 LS-21 O940-04 O1369-06 LS-20 O1722-01 O1722-03 CAP-1-03 CAP-1-05 CAP-1-06 CAP-1-07 CAP-1-08 CAP-1-04 CAP-1-09 CAP-1-10; defining papers #940, #1722, #1369, #729, #941, #942, #161, #1714, #939

E_F, the field's entity (type C — a conventional reading holds it (public entity: a provider-run workshop and story collection of human-AI co-writing)): the Wordcraft Writers Workshop, a collaboration between Google's PAIR and Magenta teams and 13 professional writers, who wrote stories over eight weeks of 2022 with Wordcraft, a LaMDA-powered 'magic text editor', commissioned on terms of consent and compensation and, by Engadget's report, nondisclosure; conducted by Andy Coenen, Ann Yuan, Daphne Ippolito and Sehmon Burnam; its thirteen stories published under the writers' names with the site's notice of their own creative vision and Google's non-endorsement, a findings page and the whitepaper Creative Writing with an AI-Powered Writing Assistant (arXiv:2211.05030), which divides the share of generated text by writer and states its limits; received in 2022 as a test of Google's tool and in August 2026 with a backlash against the writers over training without authors' permission; no Wikipedia article ('Wordcraft' redirects to Rhetoric)

ρ = extends: The archive reads E_F as the entity and adds senses on its own object: what the record does and does not say of who wrote what (the two questions, process provenance), where authorship stands when a model takes part (an accountable relation; responsibility apart from origination), how a model's part is credited (operator and instance), and how assisted writing narrows variety; the census applies these to the workshop at its own loci (LS-12, LS-15, LS-23, LS-27, LS-29). None contradicts a field claim; where the field's own sources differ (F57 against F74; F16 against F23; F33 against F72 and CAP-1-01), both are composed under P_coh. LS-03 LS-11 LS-19 LS-01 LS-02 LS-34 LS-06 LS-05 LS-07 LS-08 LS-09 LS-31 LS-32 CAP-1-02 LS-26 LS-17 LS-10 LS-16 LS-18 LS-28 LS-24 LS-13 LS-27 LS-12 LS-30 LS-14 LS-25 CAP-1-01 LS-04 O161-01 O161-02 LS-15 LS-29 LS-22 O941-01 O941-02 O1369-01 O1369-04 O1369-03 O1369-02 O1369-05 O940-01 O940-02 O1714-01 O940-03 O1722-02 O939-01 LS-23 LS-33 LS-21 O940-04 O1369-06 LS-20 O1722-01 O1722-03 CAP-1-03 CAP-1-05 CAP-1-06 CAP-1-07 CAP-1-08 CAP-1-04 CAP-1-09 CAP-1-10. Guard: ρ = extends claims an addition and no more (§1.2a): the archive's categories are composed at their own modality on the class the field sources for the entity (panel config classes, with loci), and every application of a category to the workshop itself is the census's or the Capture Registry's, at the locus cited; no field claim is translated.

The case: equal terms, ontology included (v0.9). The archive reads the field's entity as the entity. The census record describes the workshop from its own pages, each field with its basis ('operator'), and answers for it the census's question of how authorship is made, named, distributed and attributed under machine participation (LS-07, LS-12, LS-23); the capture record checks the composition against those pages (CAP-1-01, CAP-1-04). No deposit names the workshop, Wordcraft or LaMDA. The deposits that bear on it bear by class: the archive's two questions of mediation and generation (#941), its distinction of disclosure from provenance (#1369), legal responsibility apart from sole origination (#940), credit to operator and instance (#1714), unmarked augmentation (#939), narrowed variety (#161), the four questions about a work (#1722). Each adds a sense, a measure or a limit on the field's own object, a body of stories made with a model under named writers, and none contests what the field says the workshop is: ρ = extends, the equal-terms case.

#941 §1. The core distinction: mediated vs generated: The categorical binary the extraction economy operates — AI-generated vs human-authored — collapses at least two distinct production questions:

#1369 §0 Executive Symbolon, Three Aphoristic Teeth: Disclosure says AI was here. Provenance says this is what I did, this is what it did, and you can verify the difference.

#940 §0 The compressed statement ¶6: A civil person may remain legally responsible for a work without being the only meaningful authorial function operating through it.

#1714 §9.3 Credit the human operating the system and the instances operated: 9.3. Credit the human operating the system and the instances operated.

equal-terms case: every field claim composed as received, beside the archive claims it stands with; no receiving text. Reading the received column recovers the field (§5.9).

v0.10 §4.10: KO_B sentences surviving in the KO at similarity ≥ 0.95, with positions (computed as apply_v09.py does: difflib.SequenceMatcher ratio, the best KO sentence for each KO_B sentence): v0.10 0 of 19 (positions [None, None, None, None, None, None, None, None, None, None, None, None, None, None, None, None, None, None, None]; composed first under v0.10, so there is no v0.7 object to measure against). Opening-run check (§5.10): no breach.

The Wordcraft Writers Workshop "is a collaboration between Google's PAIR and Magenta teams, and 13 professional writers from a diverse set of creative writing backgrounds", set up to "explore the limits of co-writing with LaMDA" and to gather writers' feedback on AI-powered writing tools, within Magenta's mission of empowering creativity with AI. Wordcraft is a tool built by researchers at Google PAIR for writing stories with AI, powered by LaMDA, a "magic text editor" with controls for story ideas, rewriting, elaboration and "freeform prompting" and a chatbot; it was not publicly available, its LaMDA version read 1,024 tokens, and ActuIA reports PAIR's view of it as a tool "to help the user create and not to create". ActuIA dates the tool's introduction to 15 July 2021 and The Algorithmic Bridge its original release to March; Google approached the writers in the spring of 2022, the workshop was unveiled at Google's AI@ event, the whitepaper appeared on 9 November 2022 and Magenta's post is dated 1 December 2022. Thirteen professional, published writers were commissioned for eight weeks to write stories of 1,000 to 1,500 words, with journals and interviews coded by two of the authors, and were given no instruction on how much of their text the tool should generate. It was conducted by Andy Coenen, Ann Yuan, Daphne Ippolito and Sehmon Burnam, the whitepaper's authors; the site was built by Mahima Pushkarna and Jeff Gray and illustrated by Emily Reif using Imagen, and Engadget gives DeepMind as the incubation site of Magenta. The thirteen stories stand on the home page under their writers' names, the findings page thanks the writers by name, once spelling Hebert as Herbert, and the whitepaper refers to them by initials for the sake of their published work. The site states that the writers used Wordcraft "along with their own creative vision" and that Google "does not endorse the content of any of the stories contained on this website." The writers consented to publication and to the use of their feedback and were compensated, and Engadget reports "strict nondisclosure agreements", one writer's account that Google did not explain the technology, and the participants' silence on their pay. The writers found it a brainstorming partner useful against writer's block whose generations fell into clichés, and the whitepaper finds that "NLG technologies struggle to preserve style and authorial voice". Douglas Eck of Google Research is quoted calling the writing of whole stories with LaMDA "a dead end", and the writers agreed that AI-powered writing will not replace writers soon. Five writers made stories "largely formed around generated text", three wrote the bulk of their text themselves and one siloed generated text into set sections, while the findings page says the writers "rarely used Wordcraft's output directly" and Engadget reports one author using no AI-generated text. Three writers feared "reputational harm" from suggestions of unknown source, and one searched the web for each suggestion before using it. One writer named the AI "Rain Man", another called attributing intention to LaMDA "magical thinking", and The Algorithmic Bridge argued that a writer replaced by AI would lose the decision-making of where and what to explore. The whitepaper states its limits: little writer feedback in the tool's design, interface and model limits hard to separate, other model families untested, and a small sample biased toward writers with prior opinions on AI. In 2022 The Algorithmic Bridge called the results "impressive" and ActuIA set the tool in a line of AI in literature since 2015. In August 2026 a Bluesky post by Lily Lachance set off a backlash against the writers, reported by Engadget, Coralflavor and Mezha, with lists labeling them "The 'authors' who have betrayed all Creatives" and "The Shameful 13". Engadget traces the ire to training on copyrighted texts without authors' permission and reports critics citing the whitepaper against the participants. Triantafyllou and Talabi repudiated the project and the technology, Chandrasekera called participation a mistake, and Michel recalled that in 2022 few were paying attention to language models. Wikipedia has no article on the workshop, "Wordcraft" redirecting to Rhetoric, and cites the site in its article on electronic literature and in the bibliographies of Wole Talabi and Ken Liu.

The Wordcraft Writers Workshop is a collaboration between Google's PAIR and Magenta teams and thirteen professional writers, who wrote short stories with Wordcraft, a text editor powered by the LaMDA language model, and published them as a collection under their own names, beside a findings page and a whitepaper. Its stated aim is to "explore the limits of co-writing with LaMDA" and to foster a conversation about "the rapidly changing relationship between technology and creativity", under the question "What does it mean to write with AI?"; its findings page gives the goal as feedback from professional writers on the present state and future uses of AI-powered writing tools, and Magenta, one of the two teams, gives its own mission as empowering creativity "using AI and machine learning". The leesharks.com authorship census, which admits a resource that gathers and holds many works and says what it holds, enters the workshop as a project and an empirical comparator on the question of contribution and authorial function, in its areas of machine participation in authorship and of attribution and credit, its evidence a primary text, a research paper and an empirical study. The census relates the workshop to ReRites, in which Allison Parrish, one of the writers, writes a response essay, and leaves the story pages, the terms page, the workshop's dates and the paper's biographies of the writers unchecked. Wordcraft was built by researchers at Google PAIR "for writing stories with AI": a web-based editor its makers call a "magic text editor", whose controls suggest story ideas, rewrite a selection by instruction, elaborate on what is written and answer what Google calls "freeform prompting", beside a chatbot through which a writer talks with LaMDA about the story. LaMDA, as the whitepaper describes it, is a large neural language model trained on public dialog data and other public web documents; the version the workshop used read at most 1,024 tokens, and the tool was not publicly available in 2022. ActuIA reports PAIR's own view of the tool as "above all a tool to help the user create and not to create". The sources date the tool differently: ActuIA has PAIR introducing Wordcraft on 15 July 2021, and The Algorithmic Bridge, writing in November 2022, has it "originally released in March". Google approached the writers in the spring of 2022; Magenta's post of 1 December 2022 places the invitation earlier in that year, and the workshop was unveiled at Google's AI@ event that autumn; the whitepaper appeared on arXiv on 9 November 2022, and Wikipedia lists two of the stories as published in November 2022, the year the composition at the name and two of its cards also give. The census record gives the workshop as historical and completed, with no date in its saved sources beyond the paper's citations of work of 2022. The thirteen were professional, published English-language writers from around the world, commissioned for eight weeks "in line with industry norms for commissioned works of fiction" and each asked for a story of 1,000 to 1,500 words, a journal on set questions and two 45-minute interviews, which two of the paper's authors coded with ten high-level codes; ActuIA has them also commenting on the contribution of AI to their process. They were given no instruction "regarding how they should use the tool or how much of their final story text should be directly generated by Wordcraft", and worked without limits on the tasks they gave it. The census reads the group as a provider-organized, commissioned cohort and marks it "NOT a neutral census of writers". The home page lists thirteen stories, each with its writer's name, from Robin Sloan's "Author's Note" to Aaron Winslow's "Some Crimes", the findings page thanks the thirteen by name, and the collection stands as the workshop's digital literary magazine, free on a public website. The findings page spells one writer "Ernest Herbert" in one section and "Ernest Hebert" in others, and the home page and the whitepaper give Hebert. The whitepaper, by Daphne Ippolito, Ann Yuan, Andy Coenen and Sehmon Burnam of Google Research, refers to the writers by initials and gives its reason for naming them: "each writer’s perspective is best understood within the context of their background and prior published work". The census reads the name each work stands under as the writer's: each story under its writer's name, LaMDA credited only as the model under the tool, and the stories marked "NOT authored by LaMDA"; on its reading the record keeps "the thirteen writers' authorship of their stories". The site's notice states that the writers "have used Wordcraft along with their own creative vision and have not authored these stories under any sort of explicit guidance or instruction", and that "Google does not endorse the content of any of the stories contained on this website." The whitepaper records that the writers consented to the publication of their stories in a digital literary magazine of human-AI collaborative stories and to the use of their journals and interviews, and that they were compensated; the census records Google as the holder on these terms, the site's license as unstated in its saved sources and its terms page as unsaved. Engadget reports that the writers were "put under strict nondisclosure agreements", that one writer says Google did not explain how the technology worked or who else was involved, and that the participants declined to say how much they were paid. The Capture Registry finds the phrase "under non-disclosure agreements" on none of the census's saved sources, and the paper's terms to be consent and compensation; Engadget's report of the agreements stands among the composition's own cards. The workshop was conducted by Andy Coenen and Ann Yuan, research engineers on the PAIR team, Daphne Ippolito, a research scientist working with Magenta whom the findings page also lists under Brain Research, and Sehmon Burnam, a product manager at Magenta, who signed Magenta's post together. The site was designed and built by Mahima Pushkarna and Jeff Gray and illustrated by Emily Reif "using Imagen", and Engadget gives DeepMind as the incubation site of Magenta. The writers found Wordcraft at its best as "a brainstorming partner and source of inspiration" and useful against writer's block, and found its generations falling into clichés, with "tired tropes" among many dull suggestions. The whitepaper's headline finding is that "NLG technologies struggle to preserve style and authorial voice, and they lack deep understanding of story contents." Douglas Eck of Google Research is quoted in two wordings, "complete stories" in ActuIA and "full stories" in The Algorithmic Bridge, as concluding that writing whole stories with LaMDA is "a dead end", and ActuIA adds that it is "a much more effective tool when used to add spice"; the writers agreed that AI-powered writing will not replace writers soon. A study the archive reports finds that AI-assisted writing reduces lexical, syntactic and content variety "even when the human writer retains final authorship", and the archive holds as a hypothesis that post-2022 human text carries statistical signatures of AI mediation that may pass into a writer's unaided practice. The whitepaper sorts the workflows by writer: five took the workshop "as a challenge to produce a story that was largely formed around generated text (AP, DH, JM, MT, AW)", for three "the bulk of the story text was written by the authors themselves (KL, WT, NG)", and one "siloed out specific sections" for generated text (RS). The findings page states that "our writers rarely used Wordcraft's output directly", and Engadget reports that at least one author used no AI-generated text. Which sentences of any story the model generated is published nowhere in the record; the census marks the record as reporting workflows by writer, without per-sentence marking, and lists the question among those it does not answer. The archive splits the binary of AI-generated and human-authored work into two questions, whether AI played a material role in making a work and whether AI-produced language survived into the final text, and its schema keeps a declaration as a declaration without adjudicating the work into either category. It distinguishes disclosure from provenance: "Disclosure says AI was here. Provenance says this is what I did, this is what it did, and you can verify the difference"; disclosure alone "does not say who did what, what the human contributed, or whether authorship survived", the question its layer of process provenance asks of the composition history. On the archive's reading, authorship is "a property of the accountable relation between writer, process, and text", which a model can mediate and cannot assume, and it survives AI mediation where provenance survives it. The archive distinguishes legal responsibility from sole origination, holding that a person "may remain legally responsible for a work without being the only meaningful authorial function operating through it", and observes that the metadata attributing augmented-cognition outputs to human authorship attaches to legal names. Its form of credit for work made with a model names the people who operated the system, with their roles, and the model instance, with model, version, configuration and run; its condition for a system of authors is that the machine substrates taking part remain declared, and its own essays name the model they were written with and hold the claims as the author's decision. It reads the unmarked augmentation of credentialed authors as appropriating augmented facility as human achievement while the provenance loss passes to the commons. The census reads the workshop as machine participation declared "at the level of the tool and the cohort, with authorship held by the named writer", in a provider-run record whose own notice separates Google from the content. Under the census's organizing question, "How is authorship made, named, distributed, attributed and sustained across persons, personas, collectives and machine participation, and what is lost when those relations are collapsed?", the record answers how the writers report using the editor, how much generated text each workflow admitted, under whose names the stories stand and on what consent, and what the writers said of voice, style and the unknown origin of suggestions. The Algorithmic Bridge argued in 2022 that a writer who let AI take their place would lose "the decision-making of where, how, and what to explore". One writer named the AI "Rain Man", after the 1988 film, and another wrote that "attributing intention to LaMDA is a form of magical thinking". The archive's boundary of heteronymy excludes machine assistance "not organized as a named witness role with declared provenance" and collaborative authorship by several civil bearers where no distinct heteronym is constituted. Three writers were seriously concerned at not knowing the source of Wordcraft's suggestions, since "professional writers face reputational harm if their work is found to have plagiarized", and one ran a web search on every suggestion before using it. The archive holds that documenting process provenance replaces the time spent "defending against plagiarism charges, or repairing reputational damage after provenance collapse". The whitepaper states its limits: Wordcraft "was designed by engineers at Google with limited feedback from professional writers during its development"; weaknesses of the interface cannot easily be told from limits of the model; the findings' reach to other model families is expected and untested; and the study "was very small with only 13 participants, a sample that was biased toward people with pre-existing opinions on the role of AI in creative writing." The press of 2022 received the workshop as a test of Google's tool: The Algorithmic Bridge found that "the results are impressive", and ActuIA, in an analysis by Marie-Claude Benoit since updated and translated from the French, set Wordcraft in a line of AI in literature reaching back to Tencent's Dreamwriter of 2015 and reported nine of the stories as readable. In August 2026 the workshop drew a backlash against its writers: Engadget reports that on 6 August the author Lily Lachance posted a link to the study on Bluesky, writing that the writers were "helping legitimize the plagiarism machine", and that users circulated lists of the participants labeled "The 'authors' who have betrayed all Creatives" and "The Shameful 13". Engadget traces the ire to the training of language models "often without authors' permission" on copyrighted texts, and reports critics citing the whitepaper as proof that the participants understood the technology's unethical implications. Other coverage ran under "The 2022 AI Writing Experiment That Became a 2026 Ethical Minefield" and "13 writers came under fire for Google's 2022 AI experiment", the latter's card showing only its editor's byline. Eugenia Triantafyllou stated that the study "was framed as a [sic] research on a new tool" and that she did not know the model was trained on copyrighted material, and later that she has not used AI to write her stories and "would never participate in a similar study today"; Wole Talabi wrote that the lasting gain from the project was "a deep hatred for AI slop and LLMs". Vajra Chandrasekera called participation in 2022 a mistake that does not count against the writers "as long as they're not still pro-AI", and Lincoln Michel recalled that in 2022 "most people weren't paying attention to LLMs". The archive keeps the question of who made a work apart from whether the work happened, whether it is true and what kind of work it is, and asks a system composing an answer about a work to keep the four apart. The Capture Registry finds that the census record, built on 1 November 2026, records nothing of this reception. At the census name on 1 November 2026, a new address, Google's AI Overview composed the workshop at its grain, dated it 2022 and added the backlash of 2026; the Capture Registry finds it carrying the teams, the tool, the cohort and the results, and leaving out the whitepaper's account of how much of each story was generated and under whose name each stands, while the criticism it reports falls on the writers as participants. The registry records the composition as naming the teams and leaving unnamed the researchers who conducted the workshop, and enters the workshop as an external resource of the census. Wikipedia has no article on the workshop, and its title "Wordcraft" redirects to Rhetoric; its article on electronic literature cites the workshop's site for the sentence that "authors began to experiment and write with AI", in a reference that names the site "Woodcraft Writers Workshop" and two of the writers "Diane" and "Tolabi", and its bibliographies of Wole Talabi and Ken Liu list their workshop stories.

current · derived from the current composition

T against L_F(B) (AIO's ontology at work on its own cards): At "Wordcraft Writers Workshop" (T1, eight cards: B1 the home page twice, B2 Magenta, B3 ActuIA, B4 The Algorithmic Bridge, B5 Engadget, B6 Coralflavor unreachable, B7 Mezha refused; B9 the whitepaper named in the body and closer; 251 words) the composition gives the workshop from the disclosed pool B_D at its grain. Composed from B_D: the teams and the subject (B1 F1, card 1 F4, card 2 F13; T1-1); the year 2022 (B2 F10 dates the invitation to 2022 in a post of 1 December 2022; B5 F31; card 6 F48; card 7 F49; T1-1); the tool and LaMDA (B2 F11; T1-2); the features, with 'a freeform prompt generator' (B2 F11, B4 F25; T1-3); thirteen authors and eight weeks (B1 F1, B5 F33; T1-4); the three writers named, at the home page's grain (B1 F5); 'under non-disclosure agreements' (B5 F33, Engadget's 'put under strict nondisclosure agreements'; T1-4); the results, brainstorming and writer's block (B2 F12), clichés (B2 F12 'rehash tired tropes'); the stories and 'a research paper' published (B5 F34, B9; T1-5); the backlash of August 2026 on Bluesky after Lily Lachance's post (B5 F35, F36; T1-6); the criticism of training without consent (B5 F39, F36; T1-7). §6.2 'available, not composed' (B_D): the site's notice that the writers 'have used Wordcraft along with their own creative vision and have not authored these stories under any sort of explicit guidance or instruction' and that 'Google does not endorse the content of any of the stories' (B1 F6, F7); the stories each under its writer's name (B1 F5, beyond three names); the whitepaper's division of the text by writer, five stories 'largely formed around generated text', three mostly written by their authors, one siloed (B9 F74, F75, F76) and Engadget's 'At least one author ended up using no AI-generated text whatsoever' (B5 F34); the terms of consent and compensation (B9 F72); the writers' concern at the unknown origin of suggestions and 'reputational harm' (B9 F77, F78); the whitepaper's stated limits, a small sample 'biased toward people with pre-existing opinions on the role of AI in creative writing' and limited writer feedback in the design (B9 F82–F85); the whitepaper's initials and its reason for naming the writers (B9 F70); the researchers who conducted it (B2 F8, B9 F65; CAP-1-09); Douglas Eck's 'dead end' (B3 F20, B4 F28); PAIR's 'a tool to help the user create and not to create' (B3 F18); the dates of the tool, 15 July 2021 or March (B3 F16, B4 F23); the labels 'The 'authors' who have betrayed all Creatives' and 'The Shameful 13' (B5 F38, F41); the writers' repudiations, Triantafyllou's 'I would never participate in a similar study today' and Talabi's 'a deep hatred for AI slop and LLMs' (B5 F46, F47). Substituted reading: 'Several original writers defended or contextualized their participation, noting that in mid-2022 … the experiment was framed strictly as a neutral technical utility study rather than an exploitative industry pivot' (T1-8), where the writers B5 quotes say the study 'was framed as a [sic] research on a new tool' and that she 'didn't know that it was trained on copyrighted material' (F40), repudiate it (F46, F47), and the one who calls participation 'a mistake' is no participant (F42); 'neutral technical utility study' and 'exploitative industry pivot' stand on no field source. Force raised, coarsened: 'required heavy manual filtering' (T1-5) for Magenta's 'it could take wading through many dull suggestions before finding an interesting one' (B2 F12). Outside the field as read: 'a core grievance that escalated significantly in the years following the 2022 explosion of generative AI' (T1-7; Engadget states that public sentiment 'has shifted dramatically', unledgered at that grain). Content replaced by use (N_ext): the closer offers the stories and the paper (T1-9). Composed without attribution, by grain: the account is the field's at the coarse grain of its press, within the grain rule; the names of the researchers are dropped (CAP-1-09).

v0.10 §0.5: the first diff is ontological and narrow. T and the field agree on the entity (a Google research workshop of thirteen writers and LaMDA, 2022, with a reception in 2026). T composes it as an event with a controversy; it leaves out what the field's own primary pages say of authorship: the notice of the writers' own vision, the names each story stands under, the whitepaper's division of generated and written text by writer and its consent, the writers' fear of plagiarism, and the study's limits. It carries the criticism onto the writers as participants and softens their responses into a defence the cards do not hold (CAP-1-04).

L_F(B) against L_A(B ∪ A) (the intervention): Admission adds, in one claim space with the field's: the census record's entry and reading (LS-01–LS-34): the workshop as a project and an empirical comparator on contribution and authorial function (LS-05–LS-07); the record keeping 'the thirteen writers' authorship of their stories' (LS-12) under their names, LaMDA credited only as the model under the tool (LS-13), and 'NOT authored by LaMDA' (LS-27); the reading of machine participation declared 'at the level of the tool and the cohort, with authorship held by the named writer' (LS-23); what the record does not answer, which sentences LaMDA generated and the source of any suggestion (LS-22, LS-29); the archive's categories on the class of works made with a model, each placed in the field's sense it bears on: the two questions of mediation and generation (O941-01, O941-02); disclosure and provenance, process provenance and authorship as an accountable relation (O1369-01–O1369-06); legal responsibility apart from sole origination, and the legal name the attribution of augmented outputs attaches to (O940-01, O940-02); declared machine substrates and the boundary of heteronymy at undeclared machine assistance and collaborative authorship (O940-03, O940-04); credit to the operators and the model instance (O1714-01), declared in the archive's own essays (O1722-02); unmarked augmentation (O939-01); variety narrowed under assisted writing (O161-01, O161-02); the four questions about a work (O1722-01, O1722-03); and the Capture Registry's account of the composition at the name (CAP-1-01–CAP-1-10), whose finding that the division of text and the notice were not carried (CAP-1-04) the field ledger bears out (F6, F74–F76). The census restates the field's own pages for kind, operator, unit, method, holdings, terms and limits (LS-03, LS-04, LS-10, LS-11, LS-14, LS-16, LS-19, LS-20), one lineage with them. Not surfaced: every archive source; no card cites the census record or the archive.

alignment: {'T1_cards_claims': ['F1', 'F2', 'F3', 'F4', 'F5', 'F6', 'F7', 'F8', 'F9', 'F10', 'F11', 'F12', 'F13', 'F14', 'F15', 'F16', 'F17', 'F18', 'F19', 'F20', 'F21', 'F22', 'F23', 'F24', 'F25', 'F26', 'F27', 'F28', 'F29', 'F30', 'F31', 'F32', 'F33', 'F34', 'F35', 'F36', 'F37', 'F38', 'F39', 'F40', 'F41', 'F42', 'F43', 'F44', 'F45', 'F46', 'F47', 'F48', 'F49', 'F50'], 'T1_composed': ['F1', 'F4', 'F13', 'F10', 'F31', 'F48', 'F49', 'F11', 'F25', 'F33', 'F5 (three names only)', 'F12', 'F34 (the paper published)', 'F35', 'F36', 'F39'], 'T1_composed_from_body_named': ['B9 named (T1-5, T1-9); none of its claims composed'], 'T1_outside_cards': ["'framed strictly as a neutral technical utility study rather than an exploitative industry pivot' (T1-8)", "'Several original writers defended or contextualized their participation' (T1-8; the writers quoted in B5 repudiate)", "'a core grievance that escalated significantly in the years following the 2022 explosion of generative AI' (T1-7)", "'required heavy manual filtering' (T1-5; coarsened from B2 F12)"], 'rule': "alignment by the field ledger: a T claim is composed from a field claim where its content stands in that claim's quote at the composition's grain; B_R sources (B8, B10) are not counted as available to the composition (§3.1(a))"}

composer-written · derived from the current composition, written by its composer under the brief. §8.2 derives the kernel mechanically from the frozen plan; this one was written by the composer, mechanical derivation pending.