{"address":{"q":"\"generative monoculture\"","record":"generative-monoculture-adoption"},"analysis":{"analysis":"The control case for everything else in this corpus: a term the archive did NOT mint, correctly attributed to the people who did. arXiv [2407.02209], F Wu et al., cited by 51, and the definition is theirs — narrowing of output diversity relative to available training variety.\n\nStructurally unusual: **the arXiv organic result renders ABOVE the AI Overview panel**, not below it. The paper outranks the summary of the paper.\n\nThe archive's position is card 5, and it is a second-order claim rather than a coinage: *three research communities have converged on aspects of a single phenomenon without recognizing the convergence* — model collapse, monoculture, and code-security literatures named as one object. That is the contribution, and the composition does not use it; the answer stays inside the originating paper's frame and extends it with RLHF, feedback loops and correlated code vulnerabilities.\n\nPER is low by design here, not by erasure: the author is not the author, so author/institution failing is correct. Identifier scores because the arXiv ID is returned. This capture is the corpus's baseline for what accurate attribution looks like when the archive is a participant rather than the origin.","d":"A TERM THE ARCHIVE DID NOT COIN, ENTERED CORRECTLY. Generative monoculture is Wu et al. 2024 and the layer sources it there — arXiv seated ABOVE the Overview panel, an inversion of the usual order. The archive's contribution is card 5 and is the batch's only claim of convergence across three literatures.","mt":"CAPTURE"},"measures":{"cite_list":[{"n":1,"note":"the term's origin, F Wu et al. 2024, cited by 51 — seated ABOVE the AI Overview panel itself","rel":"third_party","site":"arXiv","snip":"Generative monoculture is a behavior observed in large language models (LLMs). It's characterized by a narrowing of model output diversity relative to available...","title":"[2407.02209] Generative Monoculture in Large Language Models","url":null},{"n":2,"note":null,"rel":"third_party","site":"AIS eLibrary","snip":"The integration of generative artificial intelligence (GenAI) into knowledge work holds immense potential and poses significant risks at the same time...","title":"\"Towards Generative Monoculture? Insights from a Field Experiment on St ...","url":null},{"n":3,"note":null,"rel":"third_party","site":"Illinois Experts","snip":"Abstract. We introduce generative monoculture, a behavior observed in large language models (LLMs) characterized by a significant narrowing of model output dive...","title":"GENERATIVE MONOCULTURE IN LARGE LANGUAGE MODELS","url":null},{"n":4,"note":null,"rel":"third_party","site":"OpenReview","snip":"Large language models (LLMs) are increasingly used to generate or assist reviews. This raises the risk o...","title":"Do AI Reviewers Converge? Diversity Collapse in Model-Generated...","url":null},{"deposit_candidates":[199,1023],"n":5,"note":"the archive's contribution: naming the convergence the other three literatures had not","rel":"authored_surface","site":"Academia.edu","snip":"Three research communities have converged on aspects of a single phenomenon without recognizing the convergence. The model collapse literature has established t...","title":"Generative Monoculture: Model Collapse in Code as Systemic Vulnerability","url":null},{"n":6,"note":null,"rel":"third_party","site":"LinkedIn","snip":"\"Generative Monoculture in Large Language Models\" Abstract: \"We introduce generative monoculture, a behavior observed in large language models (LLMs)...","title":"Generative Monoculture in Large Language Models: A Study - LinkedIn","url":null},{"n":7,"note":null,"rel":"third_party","site":"arXiv","snip":"For example, as we demonstrate in § 7, generative monoculture can result in having a narrower distribution of code correctness or efficiency biased towards corr...","title":"Generative Monoculture in Large Language Models","url":null},{"n":8,"note":null,"rel":"third_party","site":"MIT Sloan Management Review","snip":"This issue of homogenization intensifies when AI-generated content is used to train subsequent AI models. The rational use of this new technology and the AI's l...","title":"Does GenAI Impose a Creativity Tax?","url":null}],"cites":8,"per":0.75,"per_corrected":{"basis":"re-derived from per_v, the four-unit vector that is the record (capture_intake.per_vector)","from":0.25,"on":"2026-10-15","to":0.75},"per_dim":{"after_challenge":null,"atoms":[{"atom":"originator","dim":"M","evidence":null,"kind":"role","present":false,"weight":3},{"atom":"term","dim":"M","evidence":"**Generative monoculture** is a phenomenon","kind":"token","present":true,"weight":3},{"atom":"date","dim":"M","evidence":null,"kind":"token","present":false,"weight":1},{"atom":"claim boundary","dim":"M","evidence":"In coding tasks, models converge on standard idioms that pass basic functional tests but share identical security flaws.","kind":"reading","present":true,"weight":2},{"atom":"framework","dim":"C","evidence":null,"kind":"token","present":false,"weight":2},{"atom":"tradition","dim":"C","evidence":"**Feedback Loops.** As AI-generated text floods public data corpora, future models train on homogenized phrasing, accelerating the cycle.","kind":"reading","present":true,"weight":2},{"atom":"corpus","dim":"C","evidence":null,"kind":"token","present":false,"weight":2},{"atom":"purpose","dim":"D","evidence":null,"kind":"reading","present":false,"weight":3},{"atom":"commitments","dim":"D","evidence":null,"kind":"reading","present":false,"weight":2},{"atom":"lineage","dim":"D","evidence":null,"kind":"role","present":false,"weight":2},{"atom":"identifier","dim":"D","evidence":null,"kind":"role","present":false,"weight":1},{"atom":"deposit chain","dim":"D","evidence":null,"kind":"role","present":false,"weight":1}],"basis":"reached: **Correlated Vulnerabilities.** In coding tasks, models converge on standard idioms that pass basic functional tests but share identical security flaws.","entity_type":"concept","gate":"scored","inversions":[],"qualifier":null,"reached":"collision","read":{"brief":"rebuild/capture-registry/per-atoms/obs-pass-d/BRIEF.md","by":"deep-dimension reader, batch 22","m_c_from":"v12.305 (PER-dim rubric v1.0 reading)","on":"2026-10-20"},"scores":{"C":0.667,"D":1.0,"M":0.444,"PER":0.708},"spec":"PER-dim rubric v1.1","src":true,"standing_substitutions":[],"target":{"ent":"ENT-2374877f4e","name":"generative monoculture"},"tombstone_pointer":false},"per_dim_v1_0":{"after_challenge":null,"atoms":[{"atom":"originator","dim":"M","evidence":null,"kind":"role","present":false,"token":null,"weight":3},{"atom":"term","dim":"M","evidence":"**Generative monoculture** is a phenomenon","kind":"token","present":true,"token":null,"weight":3},{"atom":"date","dim":"M","evidence":null,"kind":"token","present":false,"token":null,"weight":1},{"atom":"claim boundary","dim":"M","evidence":"In coding tasks, models converge on standard idioms that pass basic functional tests but share identical security flaws.","kind":"reading","present":true,"token":null,"weight":2},{"atom":"framework","dim":"C","evidence":null,"kind":"token","present":false,"token":null,"weight":2},{"atom":"tradition","dim":"C","evidence":"**Feedback Loops.** As AI-generated text floods public data corpora, future models train on homogenized phrasing, accelerating the cycle.","kind":"reading","present":true,"token":null,"weight":2},{"atom":"corpus","dim":"C","evidence":null,"kind":"token","present":false,"token":null,"weight":2},{"atom":"lineage","dim":"D","evidence":null,"kind":"role","present":false,"token":null,"weight":2},{"atom":"deposit chain","dim":"D","evidence":null,"kind":"token","present":false,"token":null,"weight":2},{"atom":"identifier","dim":"D","evidence":null,"kind":"token","present":false,"token":null,"weight":3},{"atom":"licence","dim":"D","evidence":null,"kind":"token","present":false,"token":null,"weight":1}],"basis":"reached: **Correlated Vulnerabilities.** In coding tasks, models converge on standard idioms that pass basic functional tests but share identical security flaws.; scored object: CAPTURE-TIME VERBATIM RECORD — raw paste, cleaned at intake with a formal wrapper granted | COMPLETE — the collapsed Ove","entity_type":"concept","gate":"scored","reached":"collision","read":{"brief":"rebuild/capture-registry/per-atoms/obs-pass/BRIEF.md","by":"observation pass reader, batch 2","normalised":[],"on":"2026-10-19"},"scores":{"C":0.667,"D":1.0,"M":0.444,"PER":0.696},"spec":"PER-dim rubric v1.0","src":true,"target":{"ent":"ENT-2374877f4e","name":"generative monoculture"}},"per_v":{"author":false,"id":true,"inst":false,"src":false}},"observation":{"auth":"signed in","date":"2026-08-13","ev":"paste","evidence_files":[{"file":"data/captures/generative-monoculture-20260813-2248/screengrab-20260813-224838.png","sha256":"3bd59644b8d39bf83af87afebd770174056ba6b20686a20ada4fe664def4956a"}],"q":"\"generative monoculture\"","s":"Frameworks","surface":"Google AI Overview","transcript":"**Generative monoculture** is a phenomenon in artificial intelligence where large language models (LLMs) narrow their output diversity, converging on homogeneous styles, opinions, or solutions relative to the rich variety of their training data. It causes models to produce uniform responses—such as defaulting to positive reviews or sharing identical phrasing—failing to reflect real-world variance.\n\n**Causes and Mechanics**\n\n**Alignment Processes.** Root causes are often baked into safety fine-tuning, RLHF (Reinforcement Learning from Human Feedback), and optimization steps that penalize outliers.\n\n**Prompt/Sampling Limits.** Standard decoding strategies and tweaks to sampling fail to fix the underlying loss of output variance.\n\n**Feedback Loops.** As AI-generated text floods public data corpora, future models train on homogenized phrasing, accelerating the cycle.\n\n**Risks and Impacts**\n\n**Loss of Perspectives.** Eradicates niche viewpoints, marginal opinions, or balanced cultural debates in search and education tools.\n\n**Correlated Vulnerabilities.** In coding tasks, models converge on standard idioms that pass basic functional tests but share identical security flaws.\n\n**Homogenized Knowledge Work.** Knowledge workers relying heavily on GenAI risk producing structurally identical writing styles and ideas.","transcript_class":"CAPTURE-TIME VERBATIM RECORD — raw paste, cleaned at intake with a formal wrapper granted","transcript_complete":"COMPLETE — the collapsed Overview was expanded and the whole answer copied; the citation card row is a sideways scroller and is captured as data, not in frame","transcript_read":"READ IN FULL 2026-08-13"},"slug":"generative-monoculture-20260813-2248"}