{"id":"fc673c29b5f592106b6ac8d334bb728d9656635296995a82ee4128a06554b748","title":"Microsoft Research Podcast: Jennifer Neville on AI’s surprising failures","argument":"The Microsoft Research Podcast episode was published on October 6, 2026. Jennifer Neville argues that AI systems can fail in ways simple benchmarks miss, especially when users clarify requests over multiple turns or models work through long sequences of edits. In her team’s evaluations, performance that is high on single-turn tasks degrades significantly when the same task is specified across multiple turns; errors in longer workflows can also accumulate and cause further confusion.\n\nNeville says practical evaluation should reflect real user tasks, helping researchers find performance gaps and improve systems. She advises users to check answers rather than delegate everything: if a model gives a poor answer, trying again or asking differently may help. With oversight and verification, she says, current systems can be useful for productivity and speed. She also says users’ detailed feedback about what went wrong can help improve models, while noting that her team analyzes user data in privacy-preserving ways and does not fine-tune on users’ data or responses.\n\nNeville’s larger aim is for people and AI to work together in organizations. How much the current transformer architecture’s limitations can be addressed with wrappers, and what requires a different architecture, remains an open research question.","topicIds":["artificial-general-intelligence"],"domain":"other","claimType":"attributed reading","limits":"Text as retrieved on October 23, 2026; originally published October 6, 2026. The page identifies Jennifer Neville and Chad Atalla as speakers; the episode does not provide a transcript date within the text. Claims about privacy-preserving data analysis and feedback reflect Neville’s account, not independently supplied evidence.","speaker":{"name":"Microsoft Research","roleAtSource":"Publisher"},"source":{"title":"What AI gets wrong and what failure teaches us","publisher":"Microsoft Research","format":"Article","url":"https://www.microsoft.com/en-us/research/podcast/what-ai-gets-wrong-and-what-failure-teaches-us/","publishedAt":"2026-10-06T16:19:06.000Z","dateBasis":"Date supplied by the publisher’s feed."},"evidence":[{"text":"Jennifer Neville says that performance measured in single-turn scenarios can degrade significantly when tasks are specified across multiple turns.","excerpt":"the performance that we see in single-turn scenarios—which is very high because, of course, we’re optimizing to that as model builders—actually degrades significantly over multiple turns","url":"https://www.microsoft.com/en-us/research/podcast/what-ai-gets-wrong-and-what-failure-teaches-us/","relationship":"Source-stated"},{"text":"Neville says errors in long workflows can accumulate and further confuse AI systems.","excerpt":"we do have some work showing that again we get surprising failures as we get longer and longer into these workflows because errors, if you don’t catch them, can accumulate","url":"https://www.microsoft.com/en-us/research/podcast/what-ai-gets-wrong-and-what-failure-teaches-us/","relationship":"Source-stated"},{"text":"Neville says users should check AI answers and not expect systems to be fully delegated; she sees potential for usefulness with oversight and verification.","excerpt":"they’re not ready for things to be 100% delegated to them. But if you can figure out how to use them in a workflow with you, with oversight and verification, I think they can be very useful","url":"https://www.microsoft.com/en-us/research/podcast/what-ai-gets-wrong-and-what-failure-teaches-us/","relationship":"Source-stated"},{"text":"Neville says an open question is how much wrappers can address transformer architecture limitations versus what requires a different underlying architecture.","excerpt":"how much can we deal with the limitations of the transformer architecture with this, sort of, wrapper foundation around it and what needs to be addressed with, sort of, a fundamentally different","url":"https://www.microsoft.com/en-us/research/podcast/what-ai-gets-wrong-and-what-failure-teaches-us/","relationship":"Source-stated"}],"sourceMedia":{"status":"unchecked","sourceUrl":"https://www.microsoft.com/en-us/research/podcast/what-ai-gets-wrong-and-what-failure-teaches-us/","checkedAt":null,"items":[]},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How much can wrappers around transformers address the architecture’s limitations, and what requires a different underlying architecture?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Microsoft Research Podcast: Jennifer Neville on AI’s surprising failures","summary":"The Microsoft Research Podcast episode was published on October 6, 2026. Jennifer Neville argues that AI systems can fail in ways simple benchmarks miss, especially when users clarify requests over multiple turns or models work through long sequences of edits. In her team’s evaluations, performance that is high on single-turn tasks degrades significantly when the same task is specified across multiple turns; errors in longer workflows can also accumulate and cause further confusion.\n\nNeville says practical evaluation should reflect real user tasks, helping researchers find performance gaps and improve systems. She advises users to check answers rather than delegate everything: if a model gives a poor answer, trying again or asking differently may help. With oversight and verification, she says, current systems can be useful for productivity and speed. She also says users’ detailed feedback about what went wrong can help improve models, while noting that her team analyzes user data in privacy-preserving ways and does not fine-tune on users’ data or responses.\n\nNeville’s larger aim is for people and AI to work together in organizations. How much the current transformer architecture’s limitations can be addressed with wrappers, and what requires a different architecture, remains an open research question.","question":"How much can wrappers around transformers address the architecture’s limitations, and what requires a different underlying architecture?","claims":[{"text":"Jennifer Neville says that performance measured in single-turn scenarios can degrade significantly when tasks are specified across multiple turns.","excerpt":"the performance that we see in single-turn scenarios—which is very high because, of course, we’re optimizing to that as model builders—actually degrades significantly over multiple turns","support":"source-stated"},{"text":"Neville says errors in long workflows can accumulate and further confuse AI systems.","excerpt":"we do have some work showing that again we get surprising failures as we get longer and longer into these workflows because errors, if you don’t catch them, can accumulate","support":"source-stated"},{"text":"Neville says users should check AI answers and not expect systems to be fully delegated; she sees potential for usefulness with oversight and verification.","excerpt":"they’re not ready for things to be 100% delegated to them. But if you can figure out how to use them in a workflow with you, with oversight and verification, I think they can be very useful","support":"source-stated"},{"text":"Neville says an open question is how much wrappers can address transformer architecture limitations versus what requires a different underlying architecture.","excerpt":"how much can we deal with the limitations of the transformer architecture with this, sort of, wrapper foundation around it and what needs to be addressed with, sort of, a fundamentally different","support":"source-stated"}],"limitations":["Text as retrieved on October 23, 2026; originally published October 6, 2026. The page identifies Jennifer Neville and Chad Atalla as speakers; the episode does not provide a transcript date within the text. Claims about privacy-preserving data analysis and feedback reflect Neville’s account, not independently supplied evidence."],"mediaId":null,"correctionReason":null,"id":"fc673c29b5f592106b6ac8d334bb728d9656635296995a82ee4128a06554b748","publicationId":"0143f098-7706-1c3b-9958-057a43ebfbe7","version":1,"sourceUrl":"https://www.microsoft.com/en-us/research/podcast/what-ai-gets-wrong-and-what-failure-teaches-us/","sourceTitle":"What AI gets wrong and what failure teaches us","publisher":"Microsoft Research","publishedAt":"2026-10-06T16:19:06.000Z","originalPublished":{"value":"2026-10-06T16:19:06.000Z","precision":"instant","basis":"backfill:feed_published"},"firstObservedAt":"2026-10-18T02:35:41.193Z","createdAt":"2026-11-05T04:33:23.456Z","historical":true,"entityIds":["microsoft-research"],"topicIds":["artificial-general-intelligence"],"media":[],"sourceMedia":{"status":"unchecked","sourceUrl":"https://www.microsoft.com/en-us/research/podcast/what-ai-gets-wrong-and-what-failure-teaches-us/","checkedAt":null,"items":[]},"editorialDecision":{"outcome":"accept","mediaId":null,"inputHash":"89cc31efb0be5008d24a7d9479a8c3c42582f9066c15230a20d9d11d90fa5299","rubricVersion":"unrecorded","provider":"vercel","model":"typesafe-ai/jev","reasonCodes":["bridge_admitted_by_model"],"confidence":0.59,"decidedAt":"2026-10-23T17:53:54.434Z"},"provenance":{"textHash":"625308b614fe6f33df07751a283593edadbf5ef96dc456b9c4a14eac7d226fa0","observationId":"762b78d22e87810558f6191c6b9e5b31197efb562a506b8acf68c8c873e4c11c","retrievedAt":"2026-10-23T17:55:51.918Z","kind":"publisher-page","preparer":{"model":"openai/gpt-6-luna","runId":"58675105-dc5f-efe8-451d-9874eba27ad3"},"reviewer":{"model":"anthropic/claude-opus-5.5","runId":"d71aa09f-5579-2ddb-21ab-0066b1ec2364"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-05T04:28:24.913Z","model":"anthropic/claude-opus-5.5"}},"canonicalUrl":"https://escapemeridian.com/readings/microsoft-research-podcast-jennifer-neville-on-ai-s-surprising-fc673c29","formats":{"html":"/readings/microsoft-research-podcast-jennifer-neville-on-ai-s-surprising-fc673c29","markdown":"/readings/microsoft-research-podcast-jennifer-neville-on-ai-s-surprising-fc673c29.md","json":"/readings/microsoft-research-podcast-jennifer-neville-on-ai-s-surprising-fc673c29.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":[],"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":[{"id":"term-benchmark","name":"Benchmark","explanation":"A defined set of tests used to compare systems. Results depend on the tasks and measures selected.","referenceUrl":"https://arxiv.org/abs/2211.09110"},{"id":"term-fine-tuning","name":"Fine-tuning","explanation":"Further training of an existing model to adapt it to a particular task or use.","referenceUrl":"https://developers.google.com/machine-learning/glossary#fine-tuning"},{"id":"term-transformer","name":"Transformer architecture","explanation":"The neural-network design behind most current language models; it weighs how each part of the input relates to the others.","referenceUrl":"https://developers.google.com/machine-learning/glossary#transformer"}],"entities":[{"id":"microsoft-research","name":"Microsoft Research","kind":"organization","context":"Primary technical research with named authors, papers and implementation details.","referenceUrls":["https://www.dwarkesh.com/p/satya-nadella","https://www.dwarkesh.com/p/grant-sanderson","https://www.microsoft.com/en-us/research/blog/proof-based-general-purpose-computing-with-magnetar/","https://www.microsoft.com/en-us/research/"],"path":"/entities/microsoft-research","json":"/entities/microsoft-research.json","markdown":"/entities/microsoft-research.md"}]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"fc673c29b5f592106b6ac8d334bb728d9656635296995a82ee4128a06554b748","revision":"0143f098-7706-1c3b-9958-057a43ebfbe7","revisionKind":"publication-id","title":"Microsoft Research Podcast: Jennifer Neville on AI’s surprising failures","url":"https://escapemeridian.com/readings/microsoft-research-podcast-jennifer-neville-on-ai-s-surprising-fc673c29"},"source":{"type":"source","id":"https://www.microsoft.com/en-us/research/podcast/what-ai-gets-wrong-and-what-failure-teaches-us/","revision":"625308b614fe6f33df07751a283593edadbf5ef96dc456b9c4a14eac7d226fa0","revisionKind":"sha256-source-text","title":"What AI gets wrong and what failure teaches us","url":"https://www.microsoft.com/en-us/research/podcast/what-ai-gets-wrong-and-what-failure-teaches-us/"},"status":"available","items":[],"next":null,"scope":"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."},"revisions":[{"id":"fc673c29b5f592106b6ac8d334bb728d9656635296995a82ee4128a06554b748","title":"Microsoft Research Podcast: Jennifer Neville on AI’s surprising failures","argument":"The Microsoft Research Podcast episode was published on October 6, 2026. Jennifer Neville argues that AI systems can fail in ways simple benchmarks miss, especially when users clarify requests over multiple turns or models work through long sequences of edits. In her team’s evaluations, performance that is high on single-turn tasks degrades significantly when the same task is specified across multiple turns; errors in longer workflows can also accumulate and cause further confusion.\n\nNeville says practical evaluation should reflect real user tasks, helping researchers find performance gaps and improve systems. She advises users to check answers rather than delegate everything: if a model gives a poor answer, trying again or asking differently may help. With oversight and verification, she says, current systems can be useful for productivity and speed. She also says users’ detailed feedback about what went wrong can help improve models, while noting that her team analyzes user data in privacy-preserving ways and does not fine-tune on users’ data or responses.\n\nNeville’s larger aim is for people and AI to work together in organizations. How much the current transformer architecture’s limitations can be addressed with wrappers, and what requires a different architecture, remains an open research question.","topicIds":["artificial-general-intelligence"],"domain":"other","claimType":"attributed reading","limits":"Text as retrieved on October 23, 2026; originally published October 6, 2026. The page identifies Jennifer Neville and Chad Atalla as speakers; the episode does not provide a transcript date within the text. Claims about privacy-preserving data analysis and feedback reflect Neville’s account, not independently supplied evidence.","speaker":{"name":"Microsoft Research","roleAtSource":"Publisher"},"source":{"title":"What AI gets wrong and what failure teaches us","publisher":"Microsoft Research","format":"Article","url":"https://www.microsoft.com/en-us/research/podcast/what-ai-gets-wrong-and-what-failure-teaches-us/","publishedAt":"2026-10-06T16:19:06.000Z","dateBasis":"Date supplied by the publisher’s feed."},"evidence":[{"text":"Jennifer Neville says that performance measured in single-turn scenarios can degrade significantly when tasks are specified across multiple turns.","excerpt":"the performance that we see in single-turn scenarios—which is very high because, of course, we’re optimizing to that as model builders—actually degrades significantly over multiple turns","url":"https://www.microsoft.com/en-us/research/podcast/what-ai-gets-wrong-and-what-failure-teaches-us/","relationship":"Source-stated"},{"text":"Neville says errors in long workflows can accumulate and further confuse AI systems.","excerpt":"we do have some work showing that again we get surprising failures as we get longer and longer into these workflows because errors, if you don’t catch them, can accumulate","url":"https://www.microsoft.com/en-us/research/podcast/what-ai-gets-wrong-and-what-failure-teaches-us/","relationship":"Source-stated"},{"text":"Neville says users should check AI answers and not expect systems to be fully delegated; she sees potential for usefulness with oversight and verification.","excerpt":"they’re not ready for things to be 100% delegated to them. But if you can figure out how to use them in a workflow with you, with oversight and verification, I think they can be very useful","url":"https://www.microsoft.com/en-us/research/podcast/what-ai-gets-wrong-and-what-failure-teaches-us/","relationship":"Source-stated"},{"text":"Neville says an open question is how much wrappers can address transformer architecture limitations versus what requires a different underlying architecture.","excerpt":"how much can we deal with the limitations of the transformer architecture with this, sort of, wrapper foundation around it and what needs to be addressed with, sort of, a fundamentally different","url":"https://www.microsoft.com/en-us/research/podcast/what-ai-gets-wrong-and-what-failure-teaches-us/","relationship":"Source-stated"}],"sourceMedia":{"status":"unchecked","sourceUrl":"https://www.microsoft.com/en-us/research/podcast/what-ai-gets-wrong-and-what-failure-teaches-us/","checkedAt":null,"items":[]},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How much can wrappers around transformers address the architecture’s limitations, and what requires a different underlying architecture?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Microsoft Research Podcast: Jennifer Neville on AI’s surprising failures","summary":"The Microsoft Research Podcast episode was published on October 6, 2026. Jennifer Neville argues that AI systems can fail in ways simple benchmarks miss, especially when users clarify requests over multiple turns or models work through long sequences of edits. In her team’s evaluations, performance that is high on single-turn tasks degrades significantly when the same task is specified across multiple turns; errors in longer workflows can also accumulate and cause further confusion.\n\nNeville says practical evaluation should reflect real user tasks, helping researchers find performance gaps and improve systems. She advises users to check answers rather than delegate everything: if a model gives a poor answer, trying again or asking differently may help. With oversight and verification, she says, current systems can be useful for productivity and speed. She also says users’ detailed feedback about what went wrong can help improve models, while noting that her team analyzes user data in privacy-preserving ways and does not fine-tune on users’ data or responses.\n\nNeville’s larger aim is for people and AI to work together in organizations. How much the current transformer architecture’s limitations can be addressed with wrappers, and what requires a different architecture, remains an open research question.","question":"How much can wrappers around transformers address the architecture’s limitations, and what requires a different underlying architecture?","claims":[{"text":"Jennifer Neville says that performance measured in single-turn scenarios can degrade significantly when tasks are specified across multiple turns.","excerpt":"the performance that we see in single-turn scenarios—which is very high because, of course, we’re optimizing to that as model builders—actually degrades significantly over multiple turns","support":"source-stated"},{"text":"Neville says errors in long workflows can accumulate and further confuse AI systems.","excerpt":"we do have some work showing that again we get surprising failures as we get longer and longer into these workflows because errors, if you don’t catch them, can accumulate","support":"source-stated"},{"text":"Neville says users should check AI answers and not expect systems to be fully delegated; she sees potential for usefulness with oversight and verification.","excerpt":"they’re not ready for things to be 100% delegated to them. But if you can figure out how to use them in a workflow with you, with oversight and verification, I think they can be very useful","support":"source-stated"},{"text":"Neville says an open question is how much wrappers can address transformer architecture limitations versus what requires a different underlying architecture.","excerpt":"how much can we deal with the limitations of the transformer architecture with this, sort of, wrapper foundation around it and what needs to be addressed with, sort of, a fundamentally different","support":"source-stated"}],"limitations":["Text as retrieved on October 23, 2026; originally published October 6, 2026. The page identifies Jennifer Neville and Chad Atalla as speakers; the episode does not provide a transcript date within the text. Claims about privacy-preserving data analysis and feedback reflect Neville’s account, not independently supplied evidence."],"mediaId":null,"correctionReason":null,"id":"fc673c29b5f592106b6ac8d334bb728d9656635296995a82ee4128a06554b748","publicationId":"0143f098-7706-1c3b-9958-057a43ebfbe7","version":1,"sourceUrl":"https://www.microsoft.com/en-us/research/podcast/what-ai-gets-wrong-and-what-failure-teaches-us/","sourceTitle":"What AI gets wrong and what failure teaches us","publisher":"Microsoft Research","publishedAt":"2026-10-06T16:19:06.000Z","originalPublished":{"value":"2026-10-06T16:19:06.000Z","precision":"instant","basis":"backfill:feed_published"},"firstObservedAt":"2026-10-18T02:35:41.193Z","createdAt":"2026-11-05T04:33:23.456Z","historical":true,"entityIds":["microsoft-research"],"topicIds":["artificial-general-intelligence"],"media":[],"sourceMedia":{"status":"unchecked","sourceUrl":"https://www.microsoft.com/en-us/research/podcast/what-ai-gets-wrong-and-what-failure-teaches-us/","checkedAt":null,"items":[]},"editorialDecision":{"outcome":"accept","mediaId":null,"inputHash":"89cc31efb0be5008d24a7d9479a8c3c42582f9066c15230a20d9d11d90fa5299","rubricVersion":"unrecorded","provider":"vercel","model":"typesafe-ai/jev","reasonCodes":["bridge_admitted_by_model"],"confidence":0.59,"decidedAt":"2026-10-23T17:53:54.434Z"},"provenance":{"textHash":"625308b614fe6f33df07751a283593edadbf5ef96dc456b9c4a14eac7d226fa0","observationId":"762b78d22e87810558f6191c6b9e5b31197efb562a506b8acf68c8c873e4c11c","retrievedAt":"2026-10-23T17:55:51.918Z","kind":"publisher-page","preparer":{"model":"openai/gpt-6-luna","runId":"58675105-dc5f-efe8-451d-9874eba27ad3"},"reviewer":{"model":"anthropic/claude-opus-5.5","runId":"d71aa09f-5579-2ddb-21ab-0066b1ec2364"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-05T04:28:24.913Z","model":"anthropic/claude-opus-5.5"}}}],"audits":[{"publicationId":"0143f098-7706-1c3b-9958-057a43ebfbe7","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The reading correctly attributes the arguments to Jennifer Neville in a Microsoft Research Podcast conversation hosted by Chad Atalla. It keeps her findings, advice and privacy account as her statements rather than established fact.","excerpt":"Today, I’m joined by Jennifer Neville, a partner research manager at Microsoft Research"},{"dimension":"summary","status":"uncertain","explanation":"The summary is mostly faithful and readable, and transformer architecture is defined in the term library. However, the closing paragraph and the question both depend on 'wrappers', which is neither explained nor defined. In the source, wrappers are the agentic harnesses and back-end reasoning built around the model, and a lay reader cannot follow the open question without that. Smallest repair: add a plain gloss, such as 'wrappers (the agentic harnesses and reasoning layers built around a model)', or add 'wrapper' to the term library.","excerpt":"A lot of those limitations can be solved by the wrapper around the models—the agentic harnesses that we have around the models right now—and the reasoning that goes on, on the back end."},{"dimension":"date","status":"pass","explanation":"The reading's October 6, 2026 date matches the UTC calendar day of originalPublished. The page is a publisher page, not an archive copy, so no archive disclosure is needed. The reading uses no post-publication passages and adds no hindsight framing. The page was modified on October 20, 2026, but every relied-on passage is plainly part of the original episode transcript.","excerpt":null},{"dimension":"support","status":"pass","explanation":"Each claim appears verbatim in the transcript: single-turn performance degrading over multiple turns, errors accumulating in long workflows, the advice to check answers and avoid full delegation, the value of detailed feedback, the privacy account and not fine-tuning on user data, and the open question about transformers. The cited passages are plainly part of the original transcript rather than a later edit. The reading adds no unsupported benefits or costs.","excerpt":"So just to be clear, we’re not fine-tuning on your data or your responses"},{"dimension":"media","status":"pass","explanation":"No media was selected and no verified media candidates were supplied, so there is no media identity or relevance issue.","excerpt":null}],"createdAt":"2026-11-05T04:28:24.913Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}