{"id":"01eb1b6c6fec865074612bfe90a6ffdfb9e9cc0d3fcda6dbe4f66408f77453ee","title":"Scott J Mulligan in MIT Technology Review’s The Download on looking inside AI models","argument":"Published November 14, 2024, Scott J Mulligan’s section in MIT Technology Review’s The Download argues that researchers need a better way to inspect what happens inside AI models. He says that without understanding how AI works, a system could be used in a sensitive field such as medicine while critical flaws in its workings remain unknown. Mulligan reports that a Google DeepMind team studying mechanistic interpretability has released a tool to help researchers understand what happens when AI generates an output. He presents this work as part of a broader effort: understanding a model’s internal workings could allow its outputs to be controlled more effectively and lead to better AI systems. The piece describes that outcome as a possibility, not a demonstrated result of the tool. It leaves open what the tool can reveal in practice and whether that understanding will be enough to identify flaws or improve control. The newsletter also previews COP29, reporting that countries need to agree on a new financing goal for developing countries and that the US president-elect’s different approach to climate could shape the talks’ significance.","topicIds":["work-and-business--jobs-and-occupations","job-displacement-and-creation","investment-booms-and-busts"],"domain":"other","claimType":"attributed reading","limits":"Text as retrieved on October 31, 2026; originally published November 14, 2024. The newsletter does not describe what the Google DeepMind tool revealed, how it was evaluated, or whether it identified flaws or improved control. The newsletter previews COP29 but does not report the conference’s outcome.","speaker":{"name":"MIT Technology Review","roleAtSource":"Publisher"},"source":{"title":"The Download: understanding AI, and what to expect from the UN’s climate conference","publisher":"MIT Technology Review","format":"Article","url":"https://www.technologyreview.com/2024/11/14/1106925/the-download-understanding-ai-and-what-to-expect-from-the-uns-climate-conference/","publishedAt":"2024-11-14T13:10:00Z","dateBasis":"Date from the publisher’s own records, through its API."},"evidence":[{"text":"Scott J Mulligan warns that not understanding how AI works could leave critical flaws undiscovered when AI is used in sensitive fields such as medicine.","excerpt":"It could lead us to deploy an AI system in a highly sensitive field like medicine without understanding that it could have critical flaws embedded in its workings.","url":"https://www.technologyreview.com/2024/11/14/1106925/the-download-understanding-ai-and-what-to-expect-from-the-uns-climate-conference/","relationship":"Source-stated"},{"text":"Mulligan reports that a Google DeepMind team studying mechanistic interpretability released a tool to help researchers understand what happens when AI generates an output.","excerpt":"A team at Google DeepMind that studies something called mechanistic interpretability has been working on new ways to let us peer under the hood. It recently released a tool to help researchers understand what is happening when AI is generating an output.","url":"https://www.technologyreview.com/2024/11/14/1106925/the-download-understanding-ai-and-what-to-expect-from-the-uns-climate-conference/","relationship":"Source-stated"},{"text":"Mulligan argues that better understanding of what happens inside an AI model could help control its outputs more effectively and lead to better systems.","excerpt":"If we do, we’ll be able to control its outputs more effectively, leading to better AI systems in the future.","url":"https://www.technologyreview.com/2024/11/14/1106925/the-download-understanding-ai-and-what-to-expect-from-the-uns-climate-conference/","relationship":"Source-stated"},{"text":"The newsletter reports that COP29’s central issue is a new goal for financing developing countries’ climate efforts, and says the US president-elect’s different climate approach complicates the talks.","excerpt":"Countries need to agree to set a new goal on how much money should go to developing countries to help them finance the fight against climate change. Complicating things? A US president-elect whose approach to climate is very different from that of the current administration","url":"https://www.technologyreview.com/2024/11/14/1106925/the-download-understanding-ai-and-what-to-expect-from-the-uns-climate-conference/","relationship":"Source-stated"}],"sourceMedia":{"status":"unchecked","sourceUrl":"https://www.technologyreview.com/2024/11/14/1106925/the-download-understanding-ai-and-what-to-expect-from-the-uns-climate-conference/","checkedAt":null,"items":[]},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"What can the Google DeepMind tool reveal about how AI models generate outputs, and how might researchers use that understanding to identify flaws or control outputs?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Scott J Mulligan in MIT Technology Review’s The Download on looking inside AI models","summary":"Published November 14, 2024, Scott J Mulligan’s section in MIT Technology Review’s The Download argues that researchers need a better way to inspect what happens inside AI models. He says that without understanding how AI works, a system could be used in a sensitive field such as medicine while critical flaws in its workings remain unknown. Mulligan reports that a Google DeepMind team studying mechanistic interpretability has released a tool to help researchers understand what happens when AI generates an output. He presents this work as part of a broader effort: understanding a model’s internal workings could allow its outputs to be controlled more effectively and lead to better AI systems. The piece describes that outcome as a possibility, not a demonstrated result of the tool. It leaves open what the tool can reveal in practice and whether that understanding will be enough to identify flaws or improve control. The newsletter also previews COP29, reporting that countries need to agree on a new financing goal for developing countries and that the US president-elect’s different approach to climate could shape the talks’ significance.","question":"What can the Google DeepMind tool reveal about how AI models generate outputs, and how might researchers use that understanding to identify flaws or control outputs?","claims":[{"text":"Scott J Mulligan warns that not understanding how AI works could leave critical flaws undiscovered when AI is used in sensitive fields such as medicine.","excerpt":"It could lead us to deploy an AI system in a highly sensitive field like medicine without understanding that it could have critical flaws embedded in its workings.","support":"source-stated"},{"text":"Mulligan reports that a Google DeepMind team studying mechanistic interpretability released a tool to help researchers understand what happens when AI generates an output.","excerpt":"A team at Google DeepMind that studies something called mechanistic interpretability has been working on new ways to let us peer under the hood. It recently released a tool to help researchers understand what is happening when AI is generating an output.","support":"source-stated"},{"text":"Mulligan argues that better understanding of what happens inside an AI model could help control its outputs more effectively and lead to better systems.","excerpt":"If we do, we’ll be able to control its outputs more effectively, leading to better AI systems in the future.","support":"source-stated"},{"text":"The newsletter reports that COP29’s central issue is a new goal for financing developing countries’ climate efforts, and says the US president-elect’s different climate approach complicates the talks.","excerpt":"Countries need to agree to set a new goal on how much money should go to developing countries to help them finance the fight against climate change. Complicating things? A US president-elect whose approach to climate is very different from that of the current administration","support":"source-stated"}],"limitations":["Text as retrieved on October 31, 2026; originally published November 14, 2024.","The newsletter does not describe what the Google DeepMind tool revealed, how it was evaluated, or whether it identified flaws or improved control.","The newsletter previews COP29 but does not report the conference’s outcome."],"mediaId":null,"correctionReason":null,"id":"01eb1b6c6fec865074612bfe90a6ffdfb9e9cc0d3fcda6dbe4f66408f77453ee","publicationId":"dfa990b2-ac03-87b8-066f-de2ec0f600ea","version":1,"sourceUrl":"https://www.technologyreview.com/2024/11/14/1106925/the-download-understanding-ai-and-what-to-expect-from-the-uns-climate-conference/","sourceTitle":"The Download: understanding AI, and what to expect from the UN’s climate conference","publisher":"MIT Technology Review","publishedAt":"2024-11-14T13:10:00Z","originalPublished":{"value":"2024-11-14T13:10:00Z","precision":"instant","basis":"backfill:api_published"},"firstObservedAt":"2026-10-31T04:20:34.399Z","createdAt":"2026-11-16T13:23:23.530Z","historical":true,"entityIds":[],"topicIds":["work-and-business--jobs-and-occupations","job-displacement-and-creation","investment-booms-and-busts"],"media":[],"sourceMedia":{"status":"unchecked","sourceUrl":"https://www.technologyreview.com/2024/11/14/1106925/the-download-understanding-ai-and-what-to-expect-from-the-uns-climate-conference/","checkedAt":null,"items":[]},"editorialDecision":{"outcome":"accept","mediaId":null,"inputHash":"fd214debdca7e40e838a0fa797f1a61f5e8761538c62620f48de6852d8bc5e2a","rubricVersion":"unrecorded","provider":"anthropic","model":"claude-opus-5-5","reasonCodes":["bridge_admitted_by_model","second_opinion_categorical"],"confidence":1,"decidedAt":"2026-10-31T04:20:31.045Z"},"provenance":{"textHash":"16408efa8557926386c62c300fcf596abfc02d272df6dc55f04e9c865d01c24f","observationId":"25cbf46cd5bdd3b3daaf9ee0c927676e387c430797cfe4d385ab564e74095fda","retrievedAt":"2026-10-31T04:22:07.853Z","kind":"publisher-page","preparer":{"model":"openai/gpt-6-luna","runId":"c014de1b-a600-38a7-4ec2-b917414c27a3"},"reviewer":{"model":"anthropic/claude-opus-5.5","runId":"68dfba89-b7e0-260a-1cdd-c374d1ac8c19"}},"audit":{"status":"passed","reviewedAt":"2026-11-16T13:18:37.862Z","model":"anthropic/claude-opus-5.5"}},"canonicalUrl":"https://escapemeridian.com/readings/scott-j-mulligan-in-mit-technology-review-s-the-download-on-01eb1b6c","formats":{"html":"/readings/scott-j-mulligan-in-mit-technology-review-s-the-download-on-01eb1b6c","markdown":"/readings/scott-j-mulligan-in-mit-technology-review-s-the-download-on-01eb1b6c.md","json":"/readings/scott-j-mulligan-in-mit-technology-review-s-the-download-on-01eb1b6c.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":[{"id":"term-interpretability","name":"Interpretability","explanation":"Research into how a model reaches its outputs, often by examining its internal workings.","referenceUrl":"https://developers.google.com/machine-learning/glossary#interpretability"}],"entities":[{"id":"google-deepmind","name":"Google DeepMind","kind":"organization","context":"Google DeepMind develops AI systems for general tasks and scientific research, including AlphaFold.","referenceUrls":["https://deepmind.google/blog/","https://deepmind.google/science/alphafold/"],"path":"/entities/google-deepmind","json":"/entities/google-deepmind.json","markdown":"/entities/google-deepmind.md"}]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"01eb1b6c6fec865074612bfe90a6ffdfb9e9cc0d3fcda6dbe4f66408f77453ee","revision":"dfa990b2-ac03-87b8-066f-de2ec0f600ea","revisionKind":"publication-id","title":"Scott J Mulligan in MIT Technology Review’s The Download on looking inside AI models","url":"https://escapemeridian.com/readings/scott-j-mulligan-in-mit-technology-review-s-the-download-on-01eb1b6c"},"source":{"type":"source","id":"https://www.technologyreview.com/2024/11/14/1106925/the-download-understanding-ai-and-what-to-expect-from-the-uns-climate-conference/","revision":"16408efa8557926386c62c300fcf596abfc02d272df6dc55f04e9c865d01c24f","revisionKind":"sha256-source-text","title":"The Download: understanding AI, and what to expect from the UN’s climate conference","url":"https://www.technologyreview.com/2024/11/14/1106925/the-download-understanding-ai-and-what-to-expect-from-the-uns-climate-conference/"},"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":"01eb1b6c6fec865074612bfe90a6ffdfb9e9cc0d3fcda6dbe4f66408f77453ee","title":"Scott J Mulligan in MIT Technology Review’s The Download on looking inside AI models","argument":"Published November 14, 2024, Scott J Mulligan’s section in MIT Technology Review’s The Download argues that researchers need a better way to inspect what happens inside AI models. He says that without understanding how AI works, a system could be used in a sensitive field such as medicine while critical flaws in its workings remain unknown. Mulligan reports that a Google DeepMind team studying mechanistic interpretability has released a tool to help researchers understand what happens when AI generates an output. He presents this work as part of a broader effort: understanding a model’s internal workings could allow its outputs to be controlled more effectively and lead to better AI systems. The piece describes that outcome as a possibility, not a demonstrated result of the tool. It leaves open what the tool can reveal in practice and whether that understanding will be enough to identify flaws or improve control. The newsletter also previews COP29, reporting that countries need to agree on a new financing goal for developing countries and that the US president-elect’s different approach to climate could shape the talks’ significance.","topicIds":["work-and-business--jobs-and-occupations","job-displacement-and-creation","investment-booms-and-busts"],"domain":"other","claimType":"attributed reading","limits":"Text as retrieved on October 31, 2026; originally published November 14, 2024. The newsletter does not describe what the Google DeepMind tool revealed, how it was evaluated, or whether it identified flaws or improved control. The newsletter previews COP29 but does not report the conference’s outcome.","speaker":{"name":"MIT Technology Review","roleAtSource":"Publisher"},"source":{"title":"The Download: understanding AI, and what to expect from the UN’s climate conference","publisher":"MIT Technology Review","format":"Article","url":"https://www.technologyreview.com/2024/11/14/1106925/the-download-understanding-ai-and-what-to-expect-from-the-uns-climate-conference/","publishedAt":"2024-11-14T13:10:00Z","dateBasis":"Date from the publisher’s own records, through its API."},"evidence":[{"text":"Scott J Mulligan warns that not understanding how AI works could leave critical flaws undiscovered when AI is used in sensitive fields such as medicine.","excerpt":"It could lead us to deploy an AI system in a highly sensitive field like medicine without understanding that it could have critical flaws embedded in its workings.","url":"https://www.technologyreview.com/2024/11/14/1106925/the-download-understanding-ai-and-what-to-expect-from-the-uns-climate-conference/","relationship":"Source-stated"},{"text":"Mulligan reports that a Google DeepMind team studying mechanistic interpretability released a tool to help researchers understand what happens when AI generates an output.","excerpt":"A team at Google DeepMind that studies something called mechanistic interpretability has been working on new ways to let us peer under the hood. It recently released a tool to help researchers understand what is happening when AI is generating an output.","url":"https://www.technologyreview.com/2024/11/14/1106925/the-download-understanding-ai-and-what-to-expect-from-the-uns-climate-conference/","relationship":"Source-stated"},{"text":"Mulligan argues that better understanding of what happens inside an AI model could help control its outputs more effectively and lead to better systems.","excerpt":"If we do, we’ll be able to control its outputs more effectively, leading to better AI systems in the future.","url":"https://www.technologyreview.com/2024/11/14/1106925/the-download-understanding-ai-and-what-to-expect-from-the-uns-climate-conference/","relationship":"Source-stated"},{"text":"The newsletter reports that COP29’s central issue is a new goal for financing developing countries’ climate efforts, and says the US president-elect’s different climate approach complicates the talks.","excerpt":"Countries need to agree to set a new goal on how much money should go to developing countries to help them finance the fight against climate change. Complicating things? A US president-elect whose approach to climate is very different from that of the current administration","url":"https://www.technologyreview.com/2024/11/14/1106925/the-download-understanding-ai-and-what-to-expect-from-the-uns-climate-conference/","relationship":"Source-stated"}],"sourceMedia":{"status":"unchecked","sourceUrl":"https://www.technologyreview.com/2024/11/14/1106925/the-download-understanding-ai-and-what-to-expect-from-the-uns-climate-conference/","checkedAt":null,"items":[]},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"What can the Google DeepMind tool reveal about how AI models generate outputs, and how might researchers use that understanding to identify flaws or control outputs?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Scott J Mulligan in MIT Technology Review’s The Download on looking inside AI models","summary":"Published November 14, 2024, Scott J Mulligan’s section in MIT Technology Review’s The Download argues that researchers need a better way to inspect what happens inside AI models. He says that without understanding how AI works, a system could be used in a sensitive field such as medicine while critical flaws in its workings remain unknown. Mulligan reports that a Google DeepMind team studying mechanistic interpretability has released a tool to help researchers understand what happens when AI generates an output. He presents this work as part of a broader effort: understanding a model’s internal workings could allow its outputs to be controlled more effectively and lead to better AI systems. The piece describes that outcome as a possibility, not a demonstrated result of the tool. It leaves open what the tool can reveal in practice and whether that understanding will be enough to identify flaws or improve control. The newsletter also previews COP29, reporting that countries need to agree on a new financing goal for developing countries and that the US president-elect’s different approach to climate could shape the talks’ significance.","question":"What can the Google DeepMind tool reveal about how AI models generate outputs, and how might researchers use that understanding to identify flaws or control outputs?","claims":[{"text":"Scott J Mulligan warns that not understanding how AI works could leave critical flaws undiscovered when AI is used in sensitive fields such as medicine.","excerpt":"It could lead us to deploy an AI system in a highly sensitive field like medicine without understanding that it could have critical flaws embedded in its workings.","support":"source-stated"},{"text":"Mulligan reports that a Google DeepMind team studying mechanistic interpretability released a tool to help researchers understand what happens when AI generates an output.","excerpt":"A team at Google DeepMind that studies something called mechanistic interpretability has been working on new ways to let us peer under the hood. It recently released a tool to help researchers understand what is happening when AI is generating an output.","support":"source-stated"},{"text":"Mulligan argues that better understanding of what happens inside an AI model could help control its outputs more effectively and lead to better systems.","excerpt":"If we do, we’ll be able to control its outputs more effectively, leading to better AI systems in the future.","support":"source-stated"},{"text":"The newsletter reports that COP29’s central issue is a new goal for financing developing countries’ climate efforts, and says the US president-elect’s different climate approach complicates the talks.","excerpt":"Countries need to agree to set a new goal on how much money should go to developing countries to help them finance the fight against climate change. Complicating things? A US president-elect whose approach to climate is very different from that of the current administration","support":"source-stated"}],"limitations":["Text as retrieved on October 31, 2026; originally published November 14, 2024.","The newsletter does not describe what the Google DeepMind tool revealed, how it was evaluated, or whether it identified flaws or improved control.","The newsletter previews COP29 but does not report the conference’s outcome."],"mediaId":null,"correctionReason":null,"id":"01eb1b6c6fec865074612bfe90a6ffdfb9e9cc0d3fcda6dbe4f66408f77453ee","publicationId":"dfa990b2-ac03-87b8-066f-de2ec0f600ea","version":1,"sourceUrl":"https://www.technologyreview.com/2024/11/14/1106925/the-download-understanding-ai-and-what-to-expect-from-the-uns-climate-conference/","sourceTitle":"The Download: understanding AI, and what to expect from the UN’s climate conference","publisher":"MIT Technology Review","publishedAt":"2024-11-14T13:10:00Z","originalPublished":{"value":"2024-11-14T13:10:00Z","precision":"instant","basis":"backfill:api_published"},"firstObservedAt":"2026-10-31T04:20:34.399Z","createdAt":"2026-11-16T13:23:23.530Z","historical":true,"entityIds":[],"topicIds":["work-and-business--jobs-and-occupations","job-displacement-and-creation","investment-booms-and-busts"],"media":[],"sourceMedia":{"status":"unchecked","sourceUrl":"https://www.technologyreview.com/2024/11/14/1106925/the-download-understanding-ai-and-what-to-expect-from-the-uns-climate-conference/","checkedAt":null,"items":[]},"editorialDecision":{"outcome":"accept","mediaId":null,"inputHash":"fd214debdca7e40e838a0fa797f1a61f5e8761538c62620f48de6852d8bc5e2a","rubricVersion":"unrecorded","provider":"anthropic","model":"claude-opus-5-5","reasonCodes":["bridge_admitted_by_model","second_opinion_categorical"],"confidence":1,"decidedAt":"2026-10-31T04:20:31.045Z"},"provenance":{"textHash":"16408efa8557926386c62c300fcf596abfc02d272df6dc55f04e9c865d01c24f","observationId":"25cbf46cd5bdd3b3daaf9ee0c927676e387c430797cfe4d385ab564e74095fda","retrievedAt":"2026-10-31T04:22:07.853Z","kind":"publisher-page","preparer":{"model":"openai/gpt-6-luna","runId":"c014de1b-a600-38a7-4ec2-b917414c27a3"},"reviewer":{"model":"anthropic/claude-opus-5.5","runId":"68dfba89-b7e0-260a-1cdd-c374d1ac8c19"}},"audit":{"status":"passed","reviewedAt":"2026-11-16T13:18:37.862Z","model":"anthropic/claude-opus-5.5"}}}],"audits":[{"publicationId":"dfa990b2-ac03-87b8-066f-de2ec0f600ea","verdict":"passed","findings":[{"dimension":"attribution","status":"pass","explanation":"This is one newsletter issue that comments on several items in its writers' own words, not a bare listing. The reading attributes the AI section to Scott J Mulligan and the COP29 preview to the newsletter, matching the bylines. The source's claims stay attributed rather than presented as fact.","excerpt":"—Scott J Mulligan"},{"dimension":"summary","status":"pass","explanation":"The reading leads with Mulligan's argument and the DeepMind tool release, and explains the stakes, including the medicine example. Mechanistic interpretability is defined in the page definitions. The source's confident conditional ('If we do, we'll be able to') is rendered as a possibility; that is acceptable, since it remains conditional. The question follows from the evidence. The COP29 mention is brief and accurate.","excerpt":"It’s all part of a push to get a better understanding of exactly what is happening inside an AI model."},{"dimension":"date","status":"pass","explanation":"The reading's November 14, 2024 matches the UTC day of originalPublished (2024-11-14T13:10Z). The dateModified value precedes the publication time and does not indicate a later edit. Later Deep Dive teasers, such as the OpenAI math findings, are not used by the reading. The limitations say the COP29 outcome is not reported, and nothing is framed with hindsight. The evidence kind is publisher-page, so no archive disclosure is needed.","excerpt":null},{"dimension":"support","status":"pass","explanation":"Each claim is directly supported by the quoted source text. The limitations correctly note that the newsletter gives no tool results or evaluation. The COP29 framing, that the president-elect's approach complicates the talks and could set the tone, is grounded in the text. The reading invents no benefits or costs.","excerpt":"This is a big moment that could set the tone for what the next few years of the international climate world looks like."},{"dimension":"media","status":"pass","explanation":"No verified media candidates were supplied, and the reading selects none, which is appropriate.","excerpt":null}],"createdAt":"2026-11-16T13:18:37.862Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}