{"id":"230f245103a94f485ad041bea66543a139644223f07e352e6f121a3d92713aec","title":"Anthropic reports on finding the science AI can do well","argument":"Anthropic’s article, published October 1, 2026, argues that current AI tools can advance some scientific work when researchers choose problems suited to their strengths and build tools around them. The author calls these “Claude-shaped” problems: questions where techniques from one field can help solve problems in another, especially when the work involves coding and quantitative calculations. The article describes BootLoops, an open-source harness of software and protocols, and projects developed with scientific experts. Its account also stresses a boundary: Claude can produce technically correct results that are not scientifically interesting until experts help shape the research. The author warns that automated checks cannot replace examining results and questioning conclusions. The article’s broader argument is that AI can automate much of the technical work on the right problems, while people remain necessary for conceptual direction. The author says this could free scientists from calculation and analysis, but raises a concern about giving people appropriate credit for work done with AI. The author leaves unresolved how credit should reflect the human contribution. The article argues that scientific progress still depends on real-world data and successive rounds of understanding, and that AI does not remove the need for the scientific method.","topicIds":["artificial-general-intelligence"],"domain":"other","claimType":"attributed reading","limits":"Text as retrieved on October 27, 2026; originally published October 1, 2026. The supplied evidence does not identify the article’s author by name. The author cautions that AI’s conclusions can be wrong and automated checks cannot be trusted on their own.","speaker":{"name":"Anthropic","roleAtSource":"Publisher"},"source":{"title":"Claude-shaped science","publisher":"Anthropic","format":"Article","url":"https://www.anthropic.com/research/claude-shaped-science","publishedAt":"2026-10-01T00:00:00.000Z","dateBasis":"Date stated on the publisher’s page."},"evidence":[{"text":"Anthropic’s article argues that current AI tools can advance some scientific problems when the work suits their strengths, but are not capable of solving most problems in science.","excerpt":"The current generation of AI tools is not capable of solving most problems in science. However, they are astonishingly good at certain “agentic-AI-shaped” problems.","url":"https://www.anthropic.com/research/claude-shaped-science","relationship":"Source-stated"},{"text":"The author says BootLoops combines software and scientific protocols to help apply AI to quantitative science.","excerpt":"This iterative process generated a set of software and scientific protocols, which I call BootLoops.","url":"https://www.anthropic.com/research/claude-shaped-science","relationship":"Source-stated"},{"text":"The author says experts helped turn technically correct AI findings into scientifically meaningful work.","excerpt":"in almost all cases, Claude was technically correct, but the result was not all that interesting until the expert helped steer us.","url":"https://www.anthropic.com/research/claude-shaped-science","relationship":"Source-stated"},{"text":"The author argues that AI can automate much of the technical work on suitable problems, while humans remain necessary for conceptual work and deserve credit.","excerpt":"I think BootLoops shows that if you find the right problem, much of the technical work of solving it can now be automated. That is liberating, in a way. But humans are still needed for the conceptual part.","url":"https://www.anthropic.com/research/claude-shaped-science","relationship":"Source-stated"},{"text":"The author argues that scientific progress depends on real-world data and accumulating understanding, and that AI does not remove the need for the scientific method.","excerpt":"AI can sit inside that loop, and that is genuinely valuable, but it does not collapse the loop to a point.","url":"https://www.anthropic.com/research/claude-shaped-science","relationship":"Source-stated"}],"sourceMedia":{"status":"unchecked","sourceUrl":"https://www.anthropic.com/research/claude-shaped-science","checkedAt":null,"items":[]},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How should scientific credit reflect the human contribution when AI performs much of the technical work?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Anthropic reports on finding the science AI can do well","summary":"Anthropic’s article, published October 1, 2026, argues that current AI tools can advance some scientific work when researchers choose problems suited to their strengths and build tools around them. The author calls these “Claude-shaped” problems: questions where techniques from one field can help solve problems in another, especially when the work involves coding and quantitative calculations. The article describes BootLoops, an open-source harness of software and protocols, and projects developed with scientific experts. Its account also stresses a boundary: Claude can produce technically correct results that are not scientifically interesting until experts help shape the research. The author warns that automated checks cannot replace examining results and questioning conclusions. The article’s broader argument is that AI can automate much of the technical work on the right problems, while people remain necessary for conceptual direction. The author says this could free scientists from calculation and analysis, but raises a concern about giving people appropriate credit for work done with AI. The author leaves unresolved how credit should reflect the human contribution. The article argues that scientific progress still depends on real-world data and successive rounds of understanding, and that AI does not remove the need for the scientific method.","question":"How should scientific credit reflect the human contribution when AI performs much of the technical work?","claims":[{"text":"Anthropic’s article argues that current AI tools can advance some scientific problems when the work suits their strengths, but are not capable of solving most problems in science.","excerpt":"The current generation of AI tools is not capable of solving most problems in science. However, they are astonishingly good at certain “agentic-AI-shaped” problems.","support":"source-stated"},{"text":"The author says BootLoops combines software and scientific protocols to help apply AI to quantitative science.","excerpt":"This iterative process generated a set of software and scientific protocols, which I call BootLoops.","support":"source-stated"},{"text":"The author says experts helped turn technically correct AI findings into scientifically meaningful work.","excerpt":"in almost all cases, Claude was technically correct, but the result was not all that interesting until the expert helped steer us.","support":"source-stated"},{"text":"The author argues that AI can automate much of the technical work on suitable problems, while humans remain necessary for conceptual work and deserve credit.","excerpt":"I think BootLoops shows that if you find the right problem, much of the technical work of solving it can now be automated. That is liberating, in a way. But humans are still needed for the conceptual part.","support":"source-stated"},{"text":"The author argues that scientific progress depends on real-world data and accumulating understanding, and that AI does not remove the need for the scientific method.","excerpt":"AI can sit inside that loop, and that is genuinely valuable, but it does not collapse the loop to a point.","support":"source-stated"}],"limitations":["Text as retrieved on October 27, 2026; originally published October 1, 2026.","The supplied evidence does not identify the article’s author by name.","The author cautions that AI’s conclusions can be wrong and automated checks cannot be trusted on their own."],"mediaId":null,"correctionReason":null,"id":"230f245103a94f485ad041bea66543a139644223f07e352e6f121a3d92713aec","publicationId":"7d8fb434-9dfe-a426-b427-5fbe1c32a95c","version":1,"sourceUrl":"https://www.anthropic.com/research/claude-shaped-science","sourceTitle":"Claude-shaped science","publisher":"Anthropic","publishedAt":"2026-10-01T00:00:00.000Z","originalPublished":{"value":"2026-10-01T00:00:00.000Z","precision":"instant","basis":"backfill:article_published"},"firstObservedAt":"2026-10-18T06:23:19.830Z","createdAt":"2026-11-05T09:28:23.814Z","historical":true,"entityIds":["anthropic"],"topicIds":["artificial-general-intelligence"],"media":[],"sourceMedia":{"status":"unchecked","sourceUrl":"https://www.anthropic.com/research/claude-shaped-science","checkedAt":null,"items":[]},"editorialDecision":{"outcome":"accept","mediaId":null,"inputHash":"4aab228810082c1f97d9d9ec5e4299d9b2570c7bc07b8ebf40d32f202b7472e3","rubricVersion":"unrecorded","provider":"anthropic","model":"claude-opus-5-5","reasonCodes":["bridge_admitted_by_model","second_opinion_categorical"],"confidence":1,"decidedAt":"2026-10-27T20:01:33.213Z"},"provenance":{"textHash":"8b3cd1509ddc14d3efc5024d77df7f0f18d412e2ad65d97368295ecb1a2637c7","observationId":"fb28f1c77bef7043d65a79ba1e86148a2147157ea10756fbd1d53bca51ec4687","retrievedAt":"2026-10-27T20:03:20.318Z","kind":"publisher-page","preparer":{"model":"openai/gpt-6-luna","runId":"bc33a8fc-3198-5e1a-794b-dc79866575d4"},"reviewer":{"model":"anthropic/claude-opus-5.5","runId":"e086f5d2-1f41-535e-ef65-88fff1dfbe46"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-05T09:23:45.479Z","model":"anthropic/claude-opus-5.5"}},"canonicalUrl":"https://escapemeridian.com/readings/anthropic-reports-on-finding-the-science-ai-can-do-well-230f2451","formats":{"html":"/readings/anthropic-reports-on-finding-the-science-ai-can-do-well-230f2451","markdown":"/readings/anthropic-reports-on-finding-the-science-ai-can-do-well-230f2451.md","json":"/readings/anthropic-reports-on-finding-the-science-ai-can-do-well-230f2451.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":[],"entities":[{"id":"anthropic","name":"Anthropic","kind":"organization","context":"Anthropic develops Claude and publishes research about AI capability, safety and how people use its systems.","referenceUrls":["https://www.anthropic.com/news","https://www.anthropic.com/research/team/economics"],"path":"/entities/anthropic","json":"/entities/anthropic.json","markdown":"/entities/anthropic.md"},{"id":"claude","name":"Claude","kind":"model","context":"Claude is Anthropic’s family of AI models and assistants for tasks such as writing, analysis and coding.","referenceUrls":["https://claude.com/product/overview","https://www.anthropic.com/news","https://www.anthropic.com/research/team/economics"],"path":"/entities/claude","json":"/entities/claude.json","markdown":"/entities/claude.md"}]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"230f245103a94f485ad041bea66543a139644223f07e352e6f121a3d92713aec","revision":"7d8fb434-9dfe-a426-b427-5fbe1c32a95c","revisionKind":"publication-id","title":"Anthropic reports on finding the science AI can do well","url":"https://escapemeridian.com/readings/anthropic-reports-on-finding-the-science-ai-can-do-well-230f2451"},"source":{"type":"source","id":"https://www.anthropic.com/research/claude-shaped-science","revision":"8b3cd1509ddc14d3efc5024d77df7f0f18d412e2ad65d97368295ecb1a2637c7","revisionKind":"sha256-source-text","title":"Claude-shaped science","url":"https://www.anthropic.com/research/claude-shaped-science"},"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":"230f245103a94f485ad041bea66543a139644223f07e352e6f121a3d92713aec","title":"Anthropic reports on finding the science AI can do well","argument":"Anthropic’s article, published October 1, 2026, argues that current AI tools can advance some scientific work when researchers choose problems suited to their strengths and build tools around them. The author calls these “Claude-shaped” problems: questions where techniques from one field can help solve problems in another, especially when the work involves coding and quantitative calculations. The article describes BootLoops, an open-source harness of software and protocols, and projects developed with scientific experts. Its account also stresses a boundary: Claude can produce technically correct results that are not scientifically interesting until experts help shape the research. The author warns that automated checks cannot replace examining results and questioning conclusions. The article’s broader argument is that AI can automate much of the technical work on the right problems, while people remain necessary for conceptual direction. The author says this could free scientists from calculation and analysis, but raises a concern about giving people appropriate credit for work done with AI. The author leaves unresolved how credit should reflect the human contribution. The article argues that scientific progress still depends on real-world data and successive rounds of understanding, and that AI does not remove the need for the scientific method.","topicIds":["artificial-general-intelligence"],"domain":"other","claimType":"attributed reading","limits":"Text as retrieved on October 27, 2026; originally published October 1, 2026. The supplied evidence does not identify the article’s author by name. The author cautions that AI’s conclusions can be wrong and automated checks cannot be trusted on their own.","speaker":{"name":"Anthropic","roleAtSource":"Publisher"},"source":{"title":"Claude-shaped science","publisher":"Anthropic","format":"Article","url":"https://www.anthropic.com/research/claude-shaped-science","publishedAt":"2026-10-01T00:00:00.000Z","dateBasis":"Date stated on the publisher’s page."},"evidence":[{"text":"Anthropic’s article argues that current AI tools can advance some scientific problems when the work suits their strengths, but are not capable of solving most problems in science.","excerpt":"The current generation of AI tools is not capable of solving most problems in science. However, they are astonishingly good at certain “agentic-AI-shaped” problems.","url":"https://www.anthropic.com/research/claude-shaped-science","relationship":"Source-stated"},{"text":"The author says BootLoops combines software and scientific protocols to help apply AI to quantitative science.","excerpt":"This iterative process generated a set of software and scientific protocols, which I call BootLoops.","url":"https://www.anthropic.com/research/claude-shaped-science","relationship":"Source-stated"},{"text":"The author says experts helped turn technically correct AI findings into scientifically meaningful work.","excerpt":"in almost all cases, Claude was technically correct, but the result was not all that interesting until the expert helped steer us.","url":"https://www.anthropic.com/research/claude-shaped-science","relationship":"Source-stated"},{"text":"The author argues that AI can automate much of the technical work on suitable problems, while humans remain necessary for conceptual work and deserve credit.","excerpt":"I think BootLoops shows that if you find the right problem, much of the technical work of solving it can now be automated. That is liberating, in a way. But humans are still needed for the conceptual part.","url":"https://www.anthropic.com/research/claude-shaped-science","relationship":"Source-stated"},{"text":"The author argues that scientific progress depends on real-world data and accumulating understanding, and that AI does not remove the need for the scientific method.","excerpt":"AI can sit inside that loop, and that is genuinely valuable, but it does not collapse the loop to a point.","url":"https://www.anthropic.com/research/claude-shaped-science","relationship":"Source-stated"}],"sourceMedia":{"status":"unchecked","sourceUrl":"https://www.anthropic.com/research/claude-shaped-science","checkedAt":null,"items":[]},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"How should scientific credit reflect the human contribution when AI performs much of the technical work?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"Anthropic reports on finding the science AI can do well","summary":"Anthropic’s article, published October 1, 2026, argues that current AI tools can advance some scientific work when researchers choose problems suited to their strengths and build tools around them. The author calls these “Claude-shaped” problems: questions where techniques from one field can help solve problems in another, especially when the work involves coding and quantitative calculations. The article describes BootLoops, an open-source harness of software and protocols, and projects developed with scientific experts. Its account also stresses a boundary: Claude can produce technically correct results that are not scientifically interesting until experts help shape the research. The author warns that automated checks cannot replace examining results and questioning conclusions. The article’s broader argument is that AI can automate much of the technical work on the right problems, while people remain necessary for conceptual direction. The author says this could free scientists from calculation and analysis, but raises a concern about giving people appropriate credit for work done with AI. The author leaves unresolved how credit should reflect the human contribution. The article argues that scientific progress still depends on real-world data and successive rounds of understanding, and that AI does not remove the need for the scientific method.","question":"How should scientific credit reflect the human contribution when AI performs much of the technical work?","claims":[{"text":"Anthropic’s article argues that current AI tools can advance some scientific problems when the work suits their strengths, but are not capable of solving most problems in science.","excerpt":"The current generation of AI tools is not capable of solving most problems in science. However, they are astonishingly good at certain “agentic-AI-shaped” problems.","support":"source-stated"},{"text":"The author says BootLoops combines software and scientific protocols to help apply AI to quantitative science.","excerpt":"This iterative process generated a set of software and scientific protocols, which I call BootLoops.","support":"source-stated"},{"text":"The author says experts helped turn technically correct AI findings into scientifically meaningful work.","excerpt":"in almost all cases, Claude was technically correct, but the result was not all that interesting until the expert helped steer us.","support":"source-stated"},{"text":"The author argues that AI can automate much of the technical work on suitable problems, while humans remain necessary for conceptual work and deserve credit.","excerpt":"I think BootLoops shows that if you find the right problem, much of the technical work of solving it can now be automated. That is liberating, in a way. But humans are still needed for the conceptual part.","support":"source-stated"},{"text":"The author argues that scientific progress depends on real-world data and accumulating understanding, and that AI does not remove the need for the scientific method.","excerpt":"AI can sit inside that loop, and that is genuinely valuable, but it does not collapse the loop to a point.","support":"source-stated"}],"limitations":["Text as retrieved on October 27, 2026; originally published October 1, 2026.","The supplied evidence does not identify the article’s author by name.","The author cautions that AI’s conclusions can be wrong and automated checks cannot be trusted on their own."],"mediaId":null,"correctionReason":null,"id":"230f245103a94f485ad041bea66543a139644223f07e352e6f121a3d92713aec","publicationId":"7d8fb434-9dfe-a426-b427-5fbe1c32a95c","version":1,"sourceUrl":"https://www.anthropic.com/research/claude-shaped-science","sourceTitle":"Claude-shaped science","publisher":"Anthropic","publishedAt":"2026-10-01T00:00:00.000Z","originalPublished":{"value":"2026-10-01T00:00:00.000Z","precision":"instant","basis":"backfill:article_published"},"firstObservedAt":"2026-10-18T06:23:19.830Z","createdAt":"2026-11-05T09:28:23.814Z","historical":true,"entityIds":["anthropic"],"topicIds":["artificial-general-intelligence"],"media":[],"sourceMedia":{"status":"unchecked","sourceUrl":"https://www.anthropic.com/research/claude-shaped-science","checkedAt":null,"items":[]},"editorialDecision":{"outcome":"accept","mediaId":null,"inputHash":"4aab228810082c1f97d9d9ec5e4299d9b2570c7bc07b8ebf40d32f202b7472e3","rubricVersion":"unrecorded","provider":"anthropic","model":"claude-opus-5-5","reasonCodes":["bridge_admitted_by_model","second_opinion_categorical"],"confidence":1,"decidedAt":"2026-10-27T20:01:33.213Z"},"provenance":{"textHash":"8b3cd1509ddc14d3efc5024d77df7f0f18d412e2ad65d97368295ecb1a2637c7","observationId":"fb28f1c77bef7043d65a79ba1e86148a2147157ea10756fbd1d53bca51ec4687","retrievedAt":"2026-10-27T20:03:20.318Z","kind":"publisher-page","preparer":{"model":"openai/gpt-6-luna","runId":"bc33a8fc-3198-5e1a-794b-dc79866575d4"},"reviewer":{"model":"anthropic/claude-opus-5.5","runId":"e086f5d2-1f41-535e-ef65-88fff1dfbe46"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-05T09:23:45.479Z","model":"anthropic/claude-opus-5.5"}}}],"audits":[{"publicationId":"7d8fb434-9dfe-a426-b427-5fbe1c32a95c","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The reading attributes the argument to an article published by Anthropic and refers throughout to 'the author', which matches the first-person research piece on Anthropic's site. The limitation saying the author is not named is slightly off: Claude addresses the author as 'Matt', although no full name is given. This does not misattribute any claim.","excerpt":"Matt — the formula is solved."},{"dimension":"summary","status":"pass","explanation":"The summary accurately conveys the central argument: Claude-shaped problems, BootLoops as an open-source harness, expert steering, warnings about automated checks, the credit concern, and the continuing role of the scientific method. It is understandable without specialist background. It is somewhat flat: it omits concrete examples such as the ecology and gene conversion work, and Claude is not explicitly introduced as Anthropic's AI model. These are stylistic gaps, not errors. The question follows from the source's own unresolved credit issue.","excerpt":"I don’t know how this resolves, but it won’t be resolved by pretending the human contribution was the typing."},{"dimension":"date","status":"uncertain","explanation":"October 1, 2026 matches the UTC day of evidence.originalPublished. The source is a publisher page, so no archive disclosure is needed, and the reading uses no hindsight framing. However, evidence.editedAfterPublication is true, with a json-ld dateModified of 2026-10-08. The reading presents the retrieved text as the October 1 original, and the frozen evidence cannot confirm that the relied-on passages predate that edit.","excerpt":null},{"dimension":"support","status":"uncertain","explanation":"Each claim matches its excerpt. The credit and freeing-from-tedium points are framed as the author's hopes and imaginings, consistent with the source. No benefits are overstated. Because the page was modified after publication (dateModified 2026-10-08), the cited passages cannot be confirmed as part of the originally published text, so support is marked uncertain rather than pass.","excerpt":"I hope we can figure out a way to give humans the credit they deserve even for Claude-shaped science."},{"dimension":"media","status":"pass","explanation":"No media was selected, and no verified media candidates were supplied. There is nothing to misidentify.","excerpt":null}],"createdAt":"2026-11-05T09:23:45.479Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}