{"id":"d72b58ce60753df0df68904388775e2ebdcdcf6d377e242d30d5a30ecd451103","title":"arXiv: Does generative AI narrow education-based productivity gaps? Evidence from a randomized experiment","argument":"The authors report that generative AI narrowed the performance gap between adults with more and less education in a randomized online experiment. In a workplace-style problem-solving task, participants worked with or without an AI assistant, then completed an unassisted module. Without AI, the higher-education group outperformed the lower-education group by 0.548 standard deviations; with AI, the gap fell to 0.139, closing about three-quarters of the initial difference. The authors say lower-education participants received substantial assistance, while higher-education participants used AI more effectively. The gap narrowed in task execution but did not disappear: after AI was removed, lower-education participants retained some improvement and a sizable gap re-emerged. Intensive AI use raised assisted performance regardless of participants’ own effort, but follow-up performance improved only when intensive use was combined with sustained effort. The authors distinguish help with a task from improvement that persists without the tool: education continued to shape unassisted performance and how participants used AI.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The evidence presented here is the abstract’s account of a randomized online experiment involving 1,174 adults aged 25–45 and a workplace-style problem-solving task; it does not establish whether the reported pattern holds in other tasks or populations.","speaker":{"name":"Cruces, Guillermo, Meijide, Diego Fernandez, Galiani, Sebastian, Galvez, Ramiro, Lombardi, Maria","roleAtSource":"Article author"},"source":{"title":"Does generative AI narrow education-based productivity gaps? Evidence from a randomized experiment","publisher":"arXiv","format":"Article","url":"https://arxiv.org/abs/2608.04198","publishedAt":"2026-08-04","dateBasis":"Date basis not provided."},"evidence":[{"text":"The study reports that generative AI improved performance for all participants, with larger gains among those with less education.","excerpt":"AI improves performance for all participants, but gains are larger among those with less education.","url":"https://arxiv.org/abs/2608.04198","relationship":"Source-stated"},{"text":"In the experiment, the education-based performance gap fell from 0.548 standard deviations without AI to 0.139 with AI, closing about three-quarters of the initial difference.","excerpt":"Without AI, higher-education participants outperform lower-education participants by 0.548 standard deviations; with AI, the gap falls to 0.139, closing about three-quarters of the initial difference.","url":"https://arxiv.org/abs/2608.04198","relationship":"Source-stated"},{"text":"The authors report that lower-education participants received substantial assistance, while higher-education participants used AI more effectively.","excerpt":"Chat logs show that lower-education participants obtain substantial assistance, while higher-education participants use AI more effectively.","url":"https://arxiv.org/abs/2608.04198","relationship":"Source-stated"},{"text":"The authors report that some improvement persisted for lower-education participants after AI was removed, although a sizable gap re-emerged.","excerpt":"lower-education participants retain part of their improvement, although a sizable gap re-emerges.","url":"https://arxiv.org/abs/2608.04198","relationship":"Source-stated"},{"text":"The study reports that intensive AI use improved follow-up performance only when paired with sustained effort.","excerpt":"follow-up performance improves only when intensive use is combined with sustained effort.","url":"https://arxiv.org/abs/2608.04198","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2608.04198","checkedAt":"2026-11-20T20:58:09.902Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"Would the reported pattern hold beyond this workplace-style problem-solving task and the study’s adult participants?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv: Does generative AI narrow education-based productivity gaps? Evidence from a randomized experiment","summary":"The authors report that generative AI narrowed the performance gap between adults with more and less education in a randomized online experiment. In a workplace-style problem-solving task, participants worked with or without an AI assistant, then completed an unassisted module. Without AI, the higher-education group outperformed the lower-education group by 0.548 standard deviations; with AI, the gap fell to 0.139, closing about three-quarters of the initial difference. The authors say lower-education participants received substantial assistance, while higher-education participants used AI more effectively. The gap narrowed in task execution but did not disappear: after AI was removed, lower-education participants retained some improvement and a sizable gap re-emerged. Intensive AI use raised assisted performance regardless of participants’ own effort, but follow-up performance improved only when intensive use was combined with sustained effort. The authors distinguish help with a task from improvement that persists without the tool: education continued to shape unassisted performance and how participants used AI.","question":"Would the reported pattern hold beyond this workplace-style problem-solving task and the study’s adult participants?","claims":[{"text":"The study reports that generative AI improved performance for all participants, with larger gains among those with less education.","excerpt":"AI improves performance for all participants, but gains are larger among those with less education.","support":"source-stated"},{"text":"In the experiment, the education-based performance gap fell from 0.548 standard deviations without AI to 0.139 with AI, closing about three-quarters of the initial difference.","excerpt":"Without AI, higher-education participants outperform lower-education participants by 0.548 standard deviations; with AI, the gap falls to 0.139, closing about three-quarters of the initial difference.","support":"source-stated"},{"text":"The authors report that lower-education participants received substantial assistance, while higher-education participants used AI more effectively.","excerpt":"Chat logs show that lower-education participants obtain substantial assistance, while higher-education participants use AI more effectively.","support":"source-stated"},{"text":"The authors report that some improvement persisted for lower-education participants after AI was removed, although a sizable gap re-emerged.","excerpt":"lower-education participants retain part of their improvement, although a sizable gap re-emerges.","support":"source-stated"},{"text":"The study reports that intensive AI use improved follow-up performance only when paired with sustained effort.","excerpt":"follow-up performance improves only when intensive use is combined with sustained effort.","support":"source-stated"}],"limitations":["The evidence presented here is the abstract’s account of a randomized online experiment involving 1,174 adults aged 25–45 and a workplace-style problem-solving task; it does not establish whether the reported pattern holds in other tasks or populations."],"mediaId":"e21b37bfc6b50e74a31c7e8d443fb5f01d83038c199673f4019ec67d6d7628b0","correctionReason":null,"id":"d72b58ce60753df0df68904388775e2ebdcdcf6d377e242d30d5a30ecd451103","publicationId":"280129b6-7b04-4ced-88f3-4b83d069994e","version":1,"sourceUrl":"https://arxiv.org/abs/2608.04198","sourceTitle":"Does generative AI narrow education-based productivity gaps? Evidence from a randomized experiment","publisher":"arXiv","publishedAt":null,"originalPublished":{"value":"2026-08-04","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-20T03:27:16.171Z","createdAt":"2026-11-20T12:26:50.017Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"e21b37bfc6b50e74a31c7e8d443fb5f01d83038c199673f4019ec67d6d7628b0","kind":"article","url":"https://arxiv.org/abs/2608.04198","sourceUrl":"https://arxiv.org/abs/2608.04198","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-20T12:26:05.095Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2608.04198","checkedAt":"2026-11-20T20:58:09.902Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v9","mediaId":"e21b37bfc6b50e74a31c7e8d443fb5f01d83038c199673f4019ec67d6d7628b0","confidence":0.98,"provider":"typesafe","model":"jev-1.13.0","inputHash":"56b2c5dac8550af2e2d190c2127602b474a57c4499cf40a0477dd18709f2b3d7","decidedAt":"2026-11-20T12:26:06.127Z"},"provenance":{"textHash":"f9178f3c3a1421ac13cac9e23b079a001cdd3ee64eda87186aec68c4c608895d","observationId":"32350c9c07db0efbc87169498396c5df37270c549686eaf51d24dfaf446d0a47","retrievedAt":"2026-11-20T12:26:05.004Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"4e2c8b63-353c-45a0-8dac-a3d8d3a7e1be"},"reviewer":{"model":"gpt-6-luna","runId":"de225184-80b6-4d04-88dc-1ca6a16d2390"}},"audit":{"status":"passed","reviewedAt":"2026-11-20T12:26:50.017Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Cruces, Guillermo","url":null,"kind":"unknown"},{"name":"Meijide, Diego Fernandez","url":null,"kind":"unknown"},{"name":"Galiani, Sebastian","url":null,"kind":"unknown"},{"name":"Galvez, Ramiro","url":null,"kind":"unknown"},{"name":"Lombardi, Maria","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Cruces, Guillermo","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Cruces, Guillermo","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2608.04198","locator":"head/meta[name=author|citation_author]"},"assertionId":"41508ed7580e67211a5e9ba488e3abba77006a8bb7a19d7159fba383a0ae11ae:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-20T20:58:09.902Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"41508ed7580e67211a5e9ba488e3abba77006a8bb7a19d7159fba383a0ae11ae","sourceUrl":"https://arxiv.org/abs/2608.04198","contentSha256":"ceab65595896be065948a36e65d73f0bc3a85fc63277abf94f48a3056484e4e6","textHash":"f9178f3c3a1421ac13cac9e23b079a001cdd3ee64eda87186aec68c4c608895d","observedAt":"2026-11-20T20:58:09.902Z"}},{"actor":{"name":"Meijide, Diego Fernandez","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Meijide, Diego Fernandez","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2608.04198","locator":"head/meta[name=author|citation_author]"},"assertionId":"41508ed7580e67211a5e9ba488e3abba77006a8bb7a19d7159fba383a0ae11ae:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-20T20:58:09.902Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"41508ed7580e67211a5e9ba488e3abba77006a8bb7a19d7159fba383a0ae11ae","sourceUrl":"https://arxiv.org/abs/2608.04198","contentSha256":"ceab65595896be065948a36e65d73f0bc3a85fc63277abf94f48a3056484e4e6","textHash":"f9178f3c3a1421ac13cac9e23b079a001cdd3ee64eda87186aec68c4c608895d","observedAt":"2026-11-20T20:58:09.902Z"}},{"actor":{"name":"Galiani, Sebastian","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Galiani, Sebastian","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2608.04198","locator":"head/meta[name=author|citation_author]"},"assertionId":"41508ed7580e67211a5e9ba488e3abba77006a8bb7a19d7159fba383a0ae11ae:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-20T20:58:09.902Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"41508ed7580e67211a5e9ba488e3abba77006a8bb7a19d7159fba383a0ae11ae","sourceUrl":"https://arxiv.org/abs/2608.04198","contentSha256":"ceab65595896be065948a36e65d73f0bc3a85fc63277abf94f48a3056484e4e6","textHash":"f9178f3c3a1421ac13cac9e23b079a001cdd3ee64eda87186aec68c4c608895d","observedAt":"2026-11-20T20:58:09.902Z"}},{"actor":{"name":"Galvez, Ramiro","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Galvez, Ramiro","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2608.04198","locator":"head/meta[name=author|citation_author]"},"assertionId":"41508ed7580e67211a5e9ba488e3abba77006a8bb7a19d7159fba383a0ae11ae:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-20T20:58:09.902Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"41508ed7580e67211a5e9ba488e3abba77006a8bb7a19d7159fba383a0ae11ae","sourceUrl":"https://arxiv.org/abs/2608.04198","contentSha256":"ceab65595896be065948a36e65d73f0bc3a85fc63277abf94f48a3056484e4e6","textHash":"f9178f3c3a1421ac13cac9e23b079a001cdd3ee64eda87186aec68c4c608895d","observedAt":"2026-11-20T20:58:09.902Z"}},{"actor":{"name":"Lombardi, Maria","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Lombardi, Maria","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2608.04198","locator":"head/meta[name=author|citation_author]"},"assertionId":"41508ed7580e67211a5e9ba488e3abba77006a8bb7a19d7159fba383a0ae11ae:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-20T20:58:09.902Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"41508ed7580e67211a5e9ba488e3abba77006a8bb7a19d7159fba383a0ae11ae","sourceUrl":"https://arxiv.org/abs/2608.04198","contentSha256":"ceab65595896be065948a36e65d73f0bc3a85fc63277abf94f48a3056484e4e6","textHash":"f9178f3c3a1421ac13cac9e23b079a001cdd3ee64eda87186aec68c4c608895d","observedAt":"2026-11-20T20:58:09.902Z"}}],"participationCoverage":{"total":5,"returned":5,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2608.04198","snapshotId":"41508ed7580e67211a5e9ba488e3abba77006a8bb7a19d7159fba383a0ae11ae","contentSha256":"ceab65595896be065948a36e65d73f0bc3a85fc63277abf94f48a3056484e4e6","textHash":"f9178f3c3a1421ac13cac9e23b079a001cdd3ee64eda87186aec68c4c608895d","observedAt":"2026-11-20T20:58:09.902Z","basis":"publisher-page-metadata","sourceLinks":41,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/d72b58ce60753df0df68904388775e2ebdcdcf6d377e242d30d5a30ecd451103/source"}},"canonicalUrl":"https://escapemeridian.com/readings/arxiv-does-generative-ai-narrow-education-based-productivity-d72b58ce","formats":{"html":"/readings/arxiv-does-generative-ai-narrow-education-based-productivity-d72b58ce","markdown":"/readings/arxiv-does-generative-ai-narrow-education-based-productivity-d72b58ce.md","json":"/readings/arxiv-does-generative-ai-narrow-education-based-productivity-d72b58ce.json"},"attribution":"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.","maintainedGuideLinks":[],"conditions":[],"readerContext":{"version":"reader-context-v1","attribution":"Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.","terms":[],"entities":[{"id":"arxiv","name":"arXiv","kind":"organization","context":"arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.","referenceUrls":["https://arxiv.org/","https://info.arxiv.org/help/api/index.html"],"path":"/entities/arxiv","json":"/entities/arxiv.json","markdown":"/entities/arxiv.md"}]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"d72b58ce60753df0df68904388775e2ebdcdcf6d377e242d30d5a30ecd451103","revision":"280129b6-7b04-4ced-88f3-4b83d069994e","revisionKind":"publication-id","title":"arXiv: Does generative AI narrow education-based productivity gaps? Evidence from a randomized experiment","url":"https://escapemeridian.com/readings/arxiv-does-generative-ai-narrow-education-based-productivity-d72b58ce"},"source":{"type":"source","id":"https://arxiv.org/abs/2608.04198","revision":"f9178f3c3a1421ac13cac9e23b079a001cdd3ee64eda87186aec68c4c608895d","revisionKind":"sha256-source-text","title":"Does generative AI narrow education-based productivity gaps? Evidence from a randomized experiment","url":"https://arxiv.org/abs/2608.04198"},"status":"available","items":[],"next":null,"scope":"Explicit citations in public readings to this reading. Shared topics are not citations. Other record types need an explicit reference adapter."},"revisions":[{"id":"d72b58ce60753df0df68904388775e2ebdcdcf6d377e242d30d5a30ecd451103","title":"arXiv: Does generative AI narrow education-based productivity gaps? Evidence from a randomized experiment","argument":"The authors report that generative AI narrowed the performance gap between adults with more and less education in a randomized online experiment. In a workplace-style problem-solving task, participants worked with or without an AI assistant, then completed an unassisted module. Without AI, the higher-education group outperformed the lower-education group by 0.548 standard deviations; with AI, the gap fell to 0.139, closing about three-quarters of the initial difference. The authors say lower-education participants received substantial assistance, while higher-education participants used AI more effectively. The gap narrowed in task execution but did not disappear: after AI was removed, lower-education participants retained some improvement and a sizable gap re-emerged. Intensive AI use raised assisted performance regardless of participants’ own effort, but follow-up performance improved only when intensive use was combined with sustained effort. The authors distinguish help with a task from improvement that persists without the tool: education continued to shape unassisted performance and how participants used AI.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The evidence presented here is the abstract’s account of a randomized online experiment involving 1,174 adults aged 25–45 and a workplace-style problem-solving task; it does not establish whether the reported pattern holds in other tasks or populations.","speaker":{"name":"Cruces, Guillermo, Meijide, Diego Fernandez, Galiani, Sebastian, Galvez, Ramiro, Lombardi, Maria","roleAtSource":"Article author"},"source":{"title":"Does generative AI narrow education-based productivity gaps? Evidence from a randomized experiment","publisher":"arXiv","format":"Article","url":"https://arxiv.org/abs/2608.04198","publishedAt":"2026-08-04","dateBasis":"Date basis not provided."},"evidence":[{"text":"The study reports that generative AI improved performance for all participants, with larger gains among those with less education.","excerpt":"AI improves performance for all participants, but gains are larger among those with less education.","url":"https://arxiv.org/abs/2608.04198","relationship":"Source-stated"},{"text":"In the experiment, the education-based performance gap fell from 0.548 standard deviations without AI to 0.139 with AI, closing about three-quarters of the initial difference.","excerpt":"Without AI, higher-education participants outperform lower-education participants by 0.548 standard deviations; with AI, the gap falls to 0.139, closing about three-quarters of the initial difference.","url":"https://arxiv.org/abs/2608.04198","relationship":"Source-stated"},{"text":"The authors report that lower-education participants received substantial assistance, while higher-education participants used AI more effectively.","excerpt":"Chat logs show that lower-education participants obtain substantial assistance, while higher-education participants use AI more effectively.","url":"https://arxiv.org/abs/2608.04198","relationship":"Source-stated"},{"text":"The authors report that some improvement persisted for lower-education participants after AI was removed, although a sizable gap re-emerged.","excerpt":"lower-education participants retain part of their improvement, although a sizable gap re-emerges.","url":"https://arxiv.org/abs/2608.04198","relationship":"Source-stated"},{"text":"The study reports that intensive AI use improved follow-up performance only when paired with sustained effort.","excerpt":"follow-up performance improves only when intensive use is combined with sustained effort.","url":"https://arxiv.org/abs/2608.04198","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2608.04198","checkedAt":"2026-11-20T20:58:09.902Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"Would the reported pattern hold beyond this workplace-style problem-solving task and the study’s adult participants?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"arXiv: Does generative AI narrow education-based productivity gaps? Evidence from a randomized experiment","summary":"The authors report that generative AI narrowed the performance gap between adults with more and less education in a randomized online experiment. In a workplace-style problem-solving task, participants worked with or without an AI assistant, then completed an unassisted module. Without AI, the higher-education group outperformed the lower-education group by 0.548 standard deviations; with AI, the gap fell to 0.139, closing about three-quarters of the initial difference. The authors say lower-education participants received substantial assistance, while higher-education participants used AI more effectively. The gap narrowed in task execution but did not disappear: after AI was removed, lower-education participants retained some improvement and a sizable gap re-emerged. Intensive AI use raised assisted performance regardless of participants’ own effort, but follow-up performance improved only when intensive use was combined with sustained effort. The authors distinguish help with a task from improvement that persists without the tool: education continued to shape unassisted performance and how participants used AI.","question":"Would the reported pattern hold beyond this workplace-style problem-solving task and the study’s adult participants?","claims":[{"text":"The study reports that generative AI improved performance for all participants, with larger gains among those with less education.","excerpt":"AI improves performance for all participants, but gains are larger among those with less education.","support":"source-stated"},{"text":"In the experiment, the education-based performance gap fell from 0.548 standard deviations without AI to 0.139 with AI, closing about three-quarters of the initial difference.","excerpt":"Without AI, higher-education participants outperform lower-education participants by 0.548 standard deviations; with AI, the gap falls to 0.139, closing about three-quarters of the initial difference.","support":"source-stated"},{"text":"The authors report that lower-education participants received substantial assistance, while higher-education participants used AI more effectively.","excerpt":"Chat logs show that lower-education participants obtain substantial assistance, while higher-education participants use AI more effectively.","support":"source-stated"},{"text":"The authors report that some improvement persisted for lower-education participants after AI was removed, although a sizable gap re-emerged.","excerpt":"lower-education participants retain part of their improvement, although a sizable gap re-emerges.","support":"source-stated"},{"text":"The study reports that intensive AI use improved follow-up performance only when paired with sustained effort.","excerpt":"follow-up performance improves only when intensive use is combined with sustained effort.","support":"source-stated"}],"limitations":["The evidence presented here is the abstract’s account of a randomized online experiment involving 1,174 adults aged 25–45 and a workplace-style problem-solving task; it does not establish whether the reported pattern holds in other tasks or populations."],"mediaId":"e21b37bfc6b50e74a31c7e8d443fb5f01d83038c199673f4019ec67d6d7628b0","correctionReason":null,"id":"d72b58ce60753df0df68904388775e2ebdcdcf6d377e242d30d5a30ecd451103","publicationId":"280129b6-7b04-4ced-88f3-4b83d069994e","version":1,"sourceUrl":"https://arxiv.org/abs/2608.04198","sourceTitle":"Does generative AI narrow education-based productivity gaps? Evidence from a randomized experiment","publisher":"arXiv","publishedAt":null,"originalPublished":{"value":"2026-08-04","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-20T03:27:16.171Z","createdAt":"2026-11-20T12:26:50.017Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"e21b37bfc6b50e74a31c7e8d443fb5f01d83038c199673f4019ec67d6d7628b0","kind":"article","url":"https://arxiv.org/abs/2608.04198","sourceUrl":"https://arxiv.org/abs/2608.04198","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-20T12:26:05.095Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2608.04198","checkedAt":"2026-11-20T20:58:09.902Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v9","mediaId":"e21b37bfc6b50e74a31c7e8d443fb5f01d83038c199673f4019ec67d6d7628b0","confidence":0.98,"provider":"typesafe","model":"jev-1.13.0","inputHash":"56b2c5dac8550af2e2d190c2127602b474a57c4499cf40a0477dd18709f2b3d7","decidedAt":"2026-11-20T12:26:06.127Z"},"provenance":{"textHash":"f9178f3c3a1421ac13cac9e23b079a001cdd3ee64eda87186aec68c4c608895d","observationId":"32350c9c07db0efbc87169498396c5df37270c549686eaf51d24dfaf446d0a47","retrievedAt":"2026-11-20T12:26:05.004Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"4e2c8b63-353c-45a0-8dac-a3d8d3a7e1be"},"reviewer":{"model":"gpt-6-luna","runId":"de225184-80b6-4d04-88dc-1ca6a16d2390"}},"audit":{"status":"passed","reviewedAt":"2026-11-20T12:26:50.017Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Cruces, Guillermo","url":null,"kind":"unknown"},{"name":"Meijide, Diego Fernandez","url":null,"kind":"unknown"},{"name":"Galiani, Sebastian","url":null,"kind":"unknown"},{"name":"Galvez, Ramiro","url":null,"kind":"unknown"},{"name":"Lombardi, Maria","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Cruces, Guillermo","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Cruces, Guillermo","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2608.04198","locator":"head/meta[name=author|citation_author]"},"assertionId":"41508ed7580e67211a5e9ba488e3abba77006a8bb7a19d7159fba383a0ae11ae:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-20T20:58:09.902Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"41508ed7580e67211a5e9ba488e3abba77006a8bb7a19d7159fba383a0ae11ae","sourceUrl":"https://arxiv.org/abs/2608.04198","contentSha256":"ceab65595896be065948a36e65d73f0bc3a85fc63277abf94f48a3056484e4e6","textHash":"f9178f3c3a1421ac13cac9e23b079a001cdd3ee64eda87186aec68c4c608895d","observedAt":"2026-11-20T20:58:09.902Z"}},{"actor":{"name":"Meijide, Diego Fernandez","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Meijide, Diego Fernandez","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2608.04198","locator":"head/meta[name=author|citation_author]"},"assertionId":"41508ed7580e67211a5e9ba488e3abba77006a8bb7a19d7159fba383a0ae11ae:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-20T20:58:09.902Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"41508ed7580e67211a5e9ba488e3abba77006a8bb7a19d7159fba383a0ae11ae","sourceUrl":"https://arxiv.org/abs/2608.04198","contentSha256":"ceab65595896be065948a36e65d73f0bc3a85fc63277abf94f48a3056484e4e6","textHash":"f9178f3c3a1421ac13cac9e23b079a001cdd3ee64eda87186aec68c4c608895d","observedAt":"2026-11-20T20:58:09.902Z"}},{"actor":{"name":"Galiani, Sebastian","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Galiani, Sebastian","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2608.04198","locator":"head/meta[name=author|citation_author]"},"assertionId":"41508ed7580e67211a5e9ba488e3abba77006a8bb7a19d7159fba383a0ae11ae:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-20T20:58:09.902Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"41508ed7580e67211a5e9ba488e3abba77006a8bb7a19d7159fba383a0ae11ae","sourceUrl":"https://arxiv.org/abs/2608.04198","contentSha256":"ceab65595896be065948a36e65d73f0bc3a85fc63277abf94f48a3056484e4e6","textHash":"f9178f3c3a1421ac13cac9e23b079a001cdd3ee64eda87186aec68c4c608895d","observedAt":"2026-11-20T20:58:09.902Z"}},{"actor":{"name":"Galvez, Ramiro","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Galvez, Ramiro","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2608.04198","locator":"head/meta[name=author|citation_author]"},"assertionId":"41508ed7580e67211a5e9ba488e3abba77006a8bb7a19d7159fba383a0ae11ae:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-20T20:58:09.902Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"41508ed7580e67211a5e9ba488e3abba77006a8bb7a19d7159fba383a0ae11ae","sourceUrl":"https://arxiv.org/abs/2608.04198","contentSha256":"ceab65595896be065948a36e65d73f0bc3a85fc63277abf94f48a3056484e4e6","textHash":"f9178f3c3a1421ac13cac9e23b079a001cdd3ee64eda87186aec68c4c608895d","observedAt":"2026-11-20T20:58:09.902Z"}},{"actor":{"name":"Lombardi, Maria","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Lombardi, Maria","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2608.04198","locator":"head/meta[name=author|citation_author]"},"assertionId":"41508ed7580e67211a5e9ba488e3abba77006a8bb7a19d7159fba383a0ae11ae:participation:4","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-20T20:58:09.902Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"41508ed7580e67211a5e9ba488e3abba77006a8bb7a19d7159fba383a0ae11ae","sourceUrl":"https://arxiv.org/abs/2608.04198","contentSha256":"ceab65595896be065948a36e65d73f0bc3a85fc63277abf94f48a3056484e4e6","textHash":"f9178f3c3a1421ac13cac9e23b079a001cdd3ee64eda87186aec68c4c608895d","observedAt":"2026-11-20T20:58:09.902Z"}}],"participationCoverage":{"total":5,"returned":5,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2608.04198","snapshotId":"41508ed7580e67211a5e9ba488e3abba77006a8bb7a19d7159fba383a0ae11ae","contentSha256":"ceab65595896be065948a36e65d73f0bc3a85fc63277abf94f48a3056484e4e6","textHash":"f9178f3c3a1421ac13cac9e23b079a001cdd3ee64eda87186aec68c4c608895d","observedAt":"2026-11-20T20:58:09.902Z","basis":"publisher-page-metadata","sourceLinks":41,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/d72b58ce60753df0df68904388775e2ebdcdcf6d377e242d30d5a30ecd451103/source"}}}],"audits":[{"publicationId":"280129b6-7b04-4ced-88f3-4b83d069994e","verdict":"passed","findings":[{"dimension":"attribution","status":"pass","explanation":"Findings are consistently attributed to the authors or the study, published on arXiv. The reading does not present them as established fact or as Escape Meridian's view.","excerpt":"Authors:Guillermo Cruces (University of Nottingham), Diego Fernandez Meijide (Universidad de San Andres), Sebastian Galiani (Tulane University)"},{"dimension":"summary","status":"pass","explanation":"The summary faithfully conveys the abstract: AI narrowed the education gap, chat-log differences in use, partial retention once AI was removed with a sizable gap re-emerging, and effort conditioning follow-up gains. Saying the gap closed by about three-quarters makes 'standard deviations' understandable without specialist background. One minor omission: the source's point that treated participants did not perform worse once AI was removed, i.e., gains were not purely delegation. The question is a genuine open issue about generalization.","excerpt":"Gains are not purely due to delegation: treated participants do not perform worse once AI is removed"},{"dimension":"date","status":"pass","explanation":"originalPublished 2026-08-04 matches the v1 submission date shown on the arXiv page.","excerpt":"[Submitted on 4 Aug 2026]"},{"dimension":"support","status":"pass","explanation":"Each claim matches its excerpt verbatim in substance. The limitation correctly bounds the scope to 1,174 adults aged 25-45 on one task. No unsupported benefits or costs are implied.","excerpt":"randomized online experiment with 1,174 adults aged 25-45 who completed a workplace-style problem-solving task"},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct link to the arXiv abstract page, which is the original source.","excerpt":null}],"createdAt":"2026-11-20T12:26:50.017Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}