AI Wiki

Explore AI Wiki

Corrections

When we get something substantively wrong, we fix it and log the fix here; small copyedits live in each article's revision history instead. Found an error? Use the "Report issue" button on the article (no account needed). So far we have received 2 reports and reviewed 1, taking a median of 4.8 days. 1 is waiting, the oldest since September 26, 2026. That is our record rather than a promise. See the verifiability policy for how checking works.

321 articles were corrected in the last 30 days. Every correction note and resolved report is listed below, newest first. A fix applied to several articles with the same note is shown once, listing each article; "See the change" opens that version in the article's revision history.

  1. editorial

    Fixed topping-out date, grid utility, GPT-5.5 training site, building and site figures; removed unsupported claims; added 2026 Microsoft campus and delivery updates

  2. editorial

    Fixed Gallup support share, PJM auction year, local dates, quotes and citations; added 2026 state, local, federal and international moratorium developments

  3. editorial

    B300 is a revised chip, not B200 silicon; sparse FP8/FP4 figures relabeled with dense values. Added B300 specs, systems, deployments, MLPerf, sales, prices, roadmap.

  4. editorial

    Fixed EU cap status, Taiwan deal terms, China antitrust date and BIS certification list; added Aug-Oct 2026 tariff, filing, shipment, court and enforcement updates

  5. editorial

    Merged Crusoe Energy into this article; fixed funding-round, site and deal dates against filings and releases, and added 2026 funding, sites and dated prices

  6. editorial

    Merged NVIDIA Blackwell B200 into this article; fixed memory, FP4, core-count, price and B300 figures against NVIDIA documents and added MLPerf, deployments and dated rental prices

  7. editorial3 articles

    OpenAI stated the value of its Oracle partnership (more than $300 billion over five years) on 2025-09-23; Oracle has not stated it

  8. editorial

    Clarified humanoid design-patent evidence and corrected the internal deployment claim

  9. editorial

    Codex 5-hour limit restored for Plus Aug 25 (Pro still exempt); Max second weekly cap is Fable; Claude Code boost ended Sep 13 with permanent +25%; Cowork boost ran to Aug 19; banked/paid/saved resets; reset catalogue through Oct 2026

  10. editorial

    Clarified active-parameter accounting and forthcoming weight release; refreshed API pricing snapshot

  11. editorial

    Stated Vinge's forecast, Bostrom's 1998 and 2009 claims, the oracle-AI checks, the International AI Safety Report's deployment factors and Willett et al.'s error rates as the sources give them; aligned the definition with Bostrom's "best human brains in practically every field". Also condensed the organizational section now covered in Artificial superintelligence, added the 2024 AI Impacts survey, and added cross-references

  12. editorial

    Dated the FLI Statement on Superintelligence to its October 22, 2025 launch (not March 27, 2026); added the May 2024 dissolution of OpenAI's Superalignment team; stated the weak-to-strong result ("consistently"), Turner et al.'s "most reward functions" and the conclusion of Bostrom's 2026 timing paper; quoted Bostrom's definition and Good's proviso exactly; removed editorial instructions that had been left in the text

  13. editorial

    Quoted Bostrom's and Good's definitions exactly; stated the weak-to-strong result accurately ("consistently"), METR's seven-month doubling, Turner et al.'s "most reward functions", and the stated conclusion of Bostrom's 2026 timing paper; added the May 2024 dissolution of OpenAI's Superalignment team; dated and quoted the CAIS (May 30, 2023) and FLI (October 22, 2025) statements; removed commentary misattributed to Bostrom (2009)

  14. editorial

    Updated CORLEO concept status and development targets.

  15. editorial

    Clarified artificial superintelligence, capability evidence, conditional risk arguments, and forecasts

  16. editorial

    Clarified token limits and added multi-agent controls

  17. editorial

    The link covered only part of the Stack Exchange Paired dataset name and pointed at The Stack, a different dataset

  18. editorial4 articles

    The generic phrase "the stack" was auto-linked to The Stack, the BigCode source-code dataset

  19. editorial

    OpenRouter's trailing window ends at the most recent complete daily bucket, and the weekly token totals are restated from a 2 October 2026 read

  20. editorial

    OpenRouter's trailing windows end at the most recent complete daily bucket rather than aggregating through the day; the Freebuff entity went from Manicode, Inc. to Freebuff, Inc. directly

  21. editorial

    The Hacker News item id, the Limited Codebase Evaluation section dates, and the OpenRouter window description (its windows end at the most recent complete daily bucket rather than aggregating through the day)

  22. editorial

    The /design skill is not in the 21 May 2026 changelog entries or the cited post

  23. editorial

    Remove four claims whose cited sources do not contain them, including a token-reduction figure resting on a dead citation and company financial figures attributed to a product-launch story

  24. editorial

    Remove three uncited teaching-adoption claims that name the wrong platforms (ETH Zurich uses Husky in Gazebo, MIT 6.4200 uses the RACECAR, and JPL's Open Source Rover has its own stack)

  25. editorial

    Competitor pricing wrong by up to 8x and the conclusion drawn from it, a fabricated Cursor figure, adaptive thinking attributed to the wrong model, a dead citation carrying six claims, a long-context pricing tier contradicted by Anthropic, and subgeneration and benchmark-leadership claims