# Campbell Lund et al. argue in Cohere that exposure scores need policy coordination [Canonical reading](https://escapemeridian.com/readings/campbell-lund-et-al-argue-in-cohere-that-exposure-scores-need-7a6159bc) Cohere authors Campbell Lund, Thomas Euyang, Zanele Munyikwa and Marzieh Fadaee published this paper on June 10, 2026. They describe GPTs are GPTs scores as measuring the share of occupational tasks a large language model tool can assist with. The scores capture particular AI systems, an occupational taxonomy and a specific moment, the authors say; their influence in policy analysis has outlasted those limits. They argue that policy-facing work continues to use static scores without engaging with methodological updates that could make assessments more reliable. The authors survey research responding to static scoring’s limits, including worker-centered metrics and adoption and usage data. But better measurement alone, they argue, will not close the deeper coordination gap between researchers and policymakers. They call on policymakers to widen the evidence base, engage workers as epistemic partners and shift from prediction to preparedness. Researchers, they say, should build data infrastructure, use interdisciplinary and participatory methods, and produce work with policymakers’ needs in mind. The authors conclude that closing the research-policy gap is a shared task, with distinct but parallel responsibilities. The abstract does not specify how these changes would affect particular policy decisions or outcomes. Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source Cohere · Article Original source date: Jun 10, 2026 (2026-06-10T00:00:00.000Z). Date basis: Date stated on the publisher’s page. Added to Escape Meridian: 2026-11-06T17:08:23.378Z. First observed: 2026-11-05T12:27:41.293Z. Retrieved: 2026-11-05T12:29:38.786Z. - [AI Exposure Scores: What they measure, what they miss, and what comes next](https://cohere.com/research/papers/ai-exposure-scores-what-they-measure-what-they-miss-and-what-comes-next-2026-06-10) ## Limits Text as retrieved on November 5, 2026; originally published June 10, 2026. The page provides the paper’s abstract rather than its full text, so the five research families and proposed responsibilities are described only at abstract level. The abstract does not specify how the proposed changes would affect particular policy decisions or outcomes. ## Open question How might policymakers and researchers carry out these distinct but parallel responsibilities, and what evidence would show that they improve policy decisions? ## Source-stated The authors describe GPTs are GPTs scores as measuring the percentage of occupational tasks a large language model tool can assist with. > they define exposure as the percentage of occupational tasks a large language model (LLM) tool can assist with - [Supporting source](https://cohere.com/research/papers/ai-exposure-scores-what-they-measure-what-they-miss-and-what-comes-next-2026-06-10) ## Source-stated The authors argue that static exposure scores have specific temporal, geographic and ontological limits, and that policy analyses have not kept pace with methodological updates. > The policy-relevant work—analyses which ask who is harmed, who benefits, how, and when—continues to reference the static GPTs are GPTs scores without engagement with the methodological updates - [Supporting source](https://cohere.com/research/papers/ai-exposure-scores-what-they-measure-what-they-miss-and-what-comes-next-2026-06-10) ## Source-stated The authors argue that researchers and policymakers share responsibility for closing the gap, with distinct but parallel roles. > We argue that closing the gap between research and policy is a shared task, with distinct but parallel responsibilities. - [Supporting source](https://cohere.com/research/papers/ai-exposure-scores-what-they-measure-what-they-miss-and-what-comes-next-2026-06-10) ## Source-stated The authors say better measurement alone will not close the research-policy coordination gap. > Better measurement matters, but it will not close the second gap alone. - [Supporting source](https://cohere.com/research/papers/ai-exposure-scores-what-they-measure-what-they-miss-and-what-comes-next-2026-06-10) ## Source and contributor context Cohere · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media retrieval status: unchecked. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Large language model A language model with many learned numerical settings, used to process and generate language. - [Definition reference](https://developers.google.com/machine-learning/glossary#large-language-model) ## Editorial provenance Version 1. Publication ID: f14e1ae0-eedf-8637-7f89-6de86b1509be. Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5. Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-06T17:03:48.309Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: bridge\_admitted\_by\_model. Observation: b3796f0b7761d08f97a012432954daeaf450f97e8b973eb0779e35727d10992c. Source text hash: 79aac34fc6c4ce2605719e42739f2da075341ea2b7ff1161ffa43ed5e162583d. ## Revision and source-check history ### Version 1 — 2026-11-06T17:08:23.378Z Cohere authors Campbell Lund, Thomas Euyang, Zanele Munyikwa and Marzieh Fadaee published this paper on June 10, 2026. They describe GPTs are GPTs scores as measuring the share of occupational tasks a large language model tool can assist with. The scores capture particular AI systems, an occupational taxonomy and a specific moment, the authors say; their influence in policy analysis has outlasted those limits. They argue that policy-facing work continues to use static scores without engaging with methodological updates that could make assessments more reliable. The authors survey research responding to static scoring’s limits, including worker-centered metrics and adoption and usage data. But better measurement alone, they argue, will not close the deeper coordination gap between researchers and policymakers. They call on policymakers to widen the evidence base, engage workers as epistemic partners and shift from prediction to preparedness. Researchers, they say, should build data infrastructure, use interdisciplinary and participatory methods, and produce work with policymakers’ needs in mind. The authors conclude that closing the research-policy gap is a shared task, with distinct but parallel responsibilities. The abstract does not specify how these changes would affect particular policy decisions or outcomes. ## Original source Cohere · Article Original source date: Jun 10, 2026 (2026-06-10T00:00:00.000Z). Date basis: Date stated on the publisher’s page. Added to Escape Meridian: 2026-11-06T17:08:23.378Z. First observed: 2026-11-05T12:27:41.293Z. Retrieved: 2026-11-05T12:29:38.786Z. - [AI Exposure Scores: What they measure, what they miss, and what comes next](https://cohere.com/research/papers/ai-exposure-scores-what-they-measure-what-they-miss-and-what-comes-next-2026-06-10) ## Limits Text as retrieved on November 5, 2026; originally published June 10, 2026. The page provides the paper’s abstract rather than its full text, so the five research families and proposed responsibilities are described only at abstract level. The abstract does not specify how the proposed changes would affect particular policy decisions or outcomes. ## Open question How might policymakers and researchers carry out these distinct but parallel responsibilities, and what evidence would show that they improve policy decisions? ## Source-stated The authors describe GPTs are GPTs scores as measuring the percentage of occupational tasks a large language model tool can assist with. > they define exposure as the percentage of occupational tasks a large language model (LLM) tool can assist with - [Supporting source](https://cohere.com/research/papers/ai-exposure-scores-what-they-measure-what-they-miss-and-what-comes-next-2026-06-10) ## Source-stated The authors argue that static exposure scores have specific temporal, geographic and ontological limits, and that policy analyses have not kept pace with methodological updates. > The policy-relevant work—analyses which ask who is harmed, who benefits, how, and when—continues to reference the static GPTs are GPTs scores without engagement with the methodological updates - [Supporting source](https://cohere.com/research/papers/ai-exposure-scores-what-they-measure-what-they-miss-and-what-comes-next-2026-06-10) ## Source-stated The authors argue that researchers and policymakers share responsibility for closing the gap, with distinct but parallel roles. > We argue that closing the gap between research and policy is a shared task, with distinct but parallel responsibilities. - [Supporting source](https://cohere.com/research/papers/ai-exposure-scores-what-they-measure-what-they-miss-and-what-comes-next-2026-06-10) ## Source-stated The authors say better measurement alone will not close the research-policy coordination gap. > Better measurement matters, but it will not close the second gap alone. - [Supporting source](https://cohere.com/research/papers/ai-exposure-scores-what-they-measure-what-they-miss-and-what-comes-next-2026-06-10) ## Source and contributor context Cohere · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media retrieval status: unchecked. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Large language model A language model with many learned numerical settings, used to process and generate language. - [Definition reference](https://developers.google.com/machine-learning/glossary#large-language-model) ## Editorial provenance Version 1. Publication ID: f14e1ae0-eedf-8637-7f89-6de86b1509be. Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5. Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-06T17:03:48.309Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: bridge\_admitted\_by\_model. Observation: b3796f0b7761d08f97a012432954daeaf450f97e8b973eb0779e35727d10992c. Source text hash: 79aac34fc6c4ce2605719e42739f2da075341ea2b7ff1161ffa43ed5e162583d. ### uncertain — 2026-11-06T17:03:48.309Z — anthropic/claude-opus-5.5 Publication: f14e1ae0-eedf-8637-7f89-6de86b1509be attribution — pass: The four named authors match the page, and the argument is attributed to them throughout. Calling them 'Cohere authors' rests on the paper appearing on Cohere's research page; the text does not state their affiliations, but this does not misattribute the argument. > Campbell Lund, Thomas Euyang, Zanele Munyikwa, and Marzieh Fadaee summary — uncertain: The summary never says the GPTs are GPTs scores were produced in 2023 by Eloundou et al. A general reader could assume the Cohere authors created them or not see why age matters to the temporal critique. Smallest repair: say the scores were calculated in 2023 by Eloundou et al. Otherwise the two gaps, the five research families and the shared responsibilities are conveyed faithfully. 'Epistemic partners' is left unexplained but is not central. > A set of exposure scores calculated in 2023 have become a central empirical input to the future of work debate. Produced by Eloundou et al. (2023) date — pass: The June 10, 2026 date matches the page dateline and originalPublished at day precision. The modified date equals the publication date. The related-works sidebar is not used. Nothing is framed with hindsight. > Jun 10, 2026 support — pass: The claims match the abstract: the exposure definition, the temporal, geographic and ontological limitations, policy work not engaging with methodological updates, the shared responsibilities, and the statement that measurement is not enough. The question and limitations stay conditional and do not add unsupported benefits. > we observe tangible impacts in how the temporal, geographic, and ontological limitations of these scores compound when translated into policy-facing analyses media — pass: No media is selected and no candidates were supplied, so nothing is misidentified. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/7a6159bc503418772f21862a5b3d6465748c282da7bd29b1bb6244b01654b02a/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/campbell-lund-et-al-argue-in-cohere-that-exposure-scores-need-7a6159bc.json)