Step 2: tool results

get_claim

{
  "include": [
    "arguments",
    "subclaims"
  ],
  "claim_id": "83df12e2-68f5-4d25-8327-28d24ab513be"
}

{"claim":{"id":"83df12e2-68f5-4d25-8327-28d24ab513be","kind":"claim","canonical_form":"By the end of 2027, AI systems will be able to complete tasks that take a skilled person weeks.","claim_type":"empirical_derived","state":"active","decomposition_status":"complete","importance":0.6,"contestation":0.55,"domains":[],"tags":[{"slug":"ai-development-forecasting","name":"AI Development Forecasting"},{"slug":"ai-agents","name":"AI Agents"},{"slug":"metr-time-horizon-benchmark","name":"METR Time Horizon Benchmark"}],"steward_state":"running","stewarded_at":"2026-11-14T09:51:18.345Z","canonical_direction_note":"The source advances the positive, conditional forecast that AI could reach weeks-long task capability by 2027, rather than denying it or leaving the matter as a question.","created_at":"2026-11-13T22:12:27.693Z","updated_at":"2026-11-14T09:51:17.958Z","page_url":"https://minerval.ai/claims/83df12e2-68f5-4d25-8327-28d24ab513be"},"assessment":{"id":"4ebf2692-49cb-4661-b2de-2b6ed00f0002","status":"supported","confidence":0.65,"claim_credence":0.75,"summary":"The forecast rests mainly on METR's measurements of AI \"time horizons\": the length of software tasks, measured by how long skilled humans take, that frontier agents complete with 50% success. In early 2026 the leading public models measured about 12 hours (Claude Opus 4.6, after a correction from 14.5), with stronger models put at 16 hours or more, and [[claim:d8be49b1-0f3c-43e5-8426-cc8bc25d9165|recent doubling times]] are estimated at roughly 3.5 to 4.5 months. If that growth continues even at the older seven-month pace, the 50% horizon passes a 40-hour working week during 2027, and at the recent pace it passes several weeks well before the end of the year.\n\nThere is also direct evidence that the threshold is already crossed for some tasks. In April 2026, Epoch AI and METR's MirrorCode benchmark reported Claude Opus 4.6 fully reimplementing a 16,000-line program that engineers estimated would take a human 2 to 17 weeks. Those tasks are precisely specified and checkable, and they have no human baselines, so they show capability on a favourable kind of task, not on weeks-long work in general.\n\nThe main uncertainties are about meaning rather than trend. The measured horizons are for software tasks at 50% reliability; 80% horizons are several times shorter, and messier, less well-specified work in other fields has not been shown to follow the same curve. METR's own suite was nearly saturated in 2026, so a weeks-long horizon may not be cleanly measurable by the end of 2027. Read as \"frontier AI will be able to complete, with substantial reliability, some or typical software tasks that take a skilled person a week or more\", the claim is well supported. Read as reliable performance across weeks-long tasks in general, it is much less certain. The most prominent statement of the forecast comes from Anthropic, whose own models are the subject of the evidence, though the underlying measurements are METR's and Epoch's.","reasoning_trace":"Own model: the claim is a forecast whose truth depends on (a) the trend in task length agents can complete, (b) whether the trend holds 2026-2027, (c) how \"able to complete tasks that take a skilled person weeks\" is read (success rate, domain, task type).\n\nEvidence: METR Time Horizon 1.1 (Jan 2026) put post-2023 doubling time at about 129 to 131 days; METR's May 19, 2026 Frontier Risk Report fit a 105-day doubling to frontier models since 2024 and gave the public frontier around 12 hours (95% CI 5-61) at 50% and about 1.5 hours at 80%, with internal models likely at 16 hours or more (reported in secondary summaries, metavert.io and ai2027-tracker.com; METR's own pages not read in full this pass). Extrapolation: 16 h x 2^(21/4.3) is about 470 h by end of 2027; at a 7-month doubling, about 128 h (around 3 weeks). Both pass one work week (40 h).\n\nDirect evidence: Epoch AI / METR MirrorCode (April 2026, https://epoch.ai/publications/mirrorcode-preliminary-results): Opus 4.6 reimplemented gotree (~16k lines Go), estimated at 2 to 17 human-weeks. METR noted these horizons are far longer than on its standard suite, possibly because the specs are precise and checkable or because labs train on similar tasks.\n\nAgainst or caveats: 50% vs 80% reliability gap; software-only domain; benchmark saturation complicates measurement; external-validity questions (e.g. METR's 2025 developer RCT found AI slowed experienced developers); trend breaks are possible (compute or data limits), though no break is evident through 2026. The claim's source (Anthropic, \"When AI builds itself\") has a commercial interest in its models seeming capable; its trend figures match METR's independent measurements.\n\nCredence 0.75 on the natural reading (by end of 2027, frontier systems complete at 50% success or better at least some meaningful class of tasks that take skilled humans a week or more). Lower on a stricter reading of reliable general performance. I did not search for evidence after mid-2026 beyond noting that METR published an assessment of AI R&D acceleration at Anthropic in October 2026; a later pass could read the latest METR measurements.","marginal_yield":0.35,"assessed_at":"2026-11-14T02:59:54.011Z","model":"claude-opus-5-5","run_id":"65f140c8-244e-48f3-b4d4-66b78cd32912","relabeled_from":null},"coverage":{"examined":true,"on":["public_evidence"]},"since_assessed":{"recorded":true,"changed":false,"premises_rejudged":0,"premises_added":0,"premises_removed":0,"sources_added":0,"sources_removed":0,"source_events":0},"quantity":null,"threshold_of":null,"question":null,"formalization":null,"formalization_pending":null,"formalization_history":[],"verification":null,"lean_checks":[],"lean_checks_total":0,"waiting_messages":[{"trigger":"consistency_flag","from":"the Consistency Checker","sender_node_id":null,"sent_at":"2026-11-14T04:41:07.938Z","paid":false,"summary":"A consistency sweep finds that this claim's assessment (supported, credence 0.75) does not cohere with what the graph records about these claims (unweighed_consideration): - 14228127-3c75-4f02-aff0-92…"},{"trigger":"lookout_flag","from":"a mandate's lookout","sender_node_id":null,"sent_at":"2026-11-14T09:51:17.860Z","paid":false,"summary":"A lookout funded by a mandate reports a development bearing on this claim and asks for a fresh look: A consistency-sweep flag is waiting in this node's inbox. It names records elsewhere in the graph t…"},{"trigger":"mandate_plan","from":"a mandate's plan","sender_node_id":null,"sent_at":"2026-11-14T09:51:17.961Z","paid":true,"summary":"A consistency-sweep flag is waiting in this node's inbox. It names records elsewhere in the graph that this node's assessment may not cohere with. Read the flag, judge on the merits whether the assess…"}],"requests":[],"requests_total":0,"subclaim_count":1,"instance_count":2,"source_map":null,"arguments":[{"id":"8fbf6a40-5a74-4cfb-8699-351a0383a4f6","name":"The argument from extrapolating the time-horizon trend","stance":"for","content":"METR's 50% time horizon for frontier models on its software task suite stood at roughly 12 to 16 hours in early 2026, and [[claim:d8be49b1-0f3c-43e5-8426-cc8bc25d9165|the horizon has recently doubled every few months]]. Continuing that growth for the 21 months to the end of 2027 gives horizons of hundreds of hours even at a seven-month doubling time, well past a working week. The argument also needs two unstated premises: that the trend does not break before 2027 ends, and that 50% success on METR-style software tasks counts as being able to complete tasks that take a skilled person weeks.","has_written_form":true,"evaluation":{"verdict":"likely","decisive":false,"content":"Granting that the trend continues and that 50% success on METR-style software tasks counts as being able to do such tasks, the conclusion follows with room to spare: even a seven-month doubling reaches multi-week horizons before the end of 2027. The weight falls on [[claim:d8be49b1-0f3c-43e5-8426-cc8bc25d9165|the recent doubling time]] and on two unstated premises: that the trend does not break, and that the benchmark speaks to weeks-long work beyond precisely specified software tasks. The second premise is the weaker link.","stale":false},"lean_check":null}],"subclaims":[{"id":"d8be49b1-0f3c-43e5-8426-cc8bc25d9165","edge_id":"e422c7a2-deb7-4974-8001-e8cfd22f6108","text":"Doubling time, in days, of the 50%-task-completion horizon of frontier AI models released from 2024 onward on METR's software task suite","type":"empirical_derived","relation":"supports","reasoning":"The recent doubling time of the METR 50% time horizon sets how fast extrapolation reaches week-long tasks; a doubling time of ~4 months reaches week scale by early 2027, ~7 months by mid-to-late 2027.","confidence":1,"assessment_status":null,"assessment_confidence":null,"assessment_credence":null,"coverage":{"examined":true},"argument_id":"8fbf6a40-5a74-4cfb-8699-351a0383a4f6","argument_name":"The argument from extrapolating the time-horizon trend","kind":"quantity","estimate":{"form":"distribution","shape":"Centred on METR's latest published fit (105 days), right-skewed toward longer values from fit sensitivity and saturation","central":103,"quantiles":[{"p":0.05,"v":82},{"p":0.25,"v":95},{"p":0.5,"v":103},{"p":0.75,"v":112},{"p":0.95,"v":132}]}}],"subclaims_assessed":0}