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 3 reports and reviewed 1, taking a median of 4.8 days. 2 are 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

    Removed duplicate references 1-9 and rebuilt the bibliography as supported numeric footnotes; corrected benchmark sizes, tree-versus-neural findings, categorical encoding guidance, XGBoost and foundation-model claims; updated the article through July 2026; and preserved the verified classification scope and internal links.

  2. editorial

    Repaired a missing citation and corrected product chronology, feature descriptions, ownership, terminology, and source support

  3. editorial

    Repaired the unbound bibliography with renderer-supported footnotes; corrected fund chronology, 2025 target-versus-close language, assets, leadership, OpenAI and portfolio claims; removed speculative return estimates and tangential personal material; and preserved the verified firm history, strategy, risks, and useful internal links.

  4. editorial

    Replaced renderer-invalid numbered markers with supported footnotes; corrected the original evaluation table, release history, licensing, and benchmark claims; removed unsupported data-cleaning, lineage, shutdown, and current-use claims; and preserved the verified model scope and internal links.

  5. editorial

    Repaired citations and corrected release history, licensing, algorithms, configuration, and integration claims

  6. editorial

    Replaced unsupported numeric reference definitions with renderer-supported footnotes; corrected the production backbone from 15B to 3B ESM-2, the CASP14 set from 71 to 51 targets, frozen-stem and v1 attention-map behavior, benchmark and speed scopes, pLDDT interpretation, multichain support, and the Atlas update; removed unsupported compute, adoption, and downstream-use claims while preserving all 25 live internal-link targets.

  7. editorial

    Replaced renderer-invalid references with supported footnotes; corrected the institute-network launch timeline and membership, participant and signatory claims, severe-risk scope, and successor dates; removed unsupported attendee, causality, and reception claims; and preserved the verified event scope and internal links.

  8. editorial

    Repaired unsupported citations; removed unverifiable community and mixed-vendor claims; corrected benchmarks, image limits, vulnerability attribution, pricing, and lifecycle facts

  9. editorial

    Repaired citations and corrected stale funding, hardware, and deployment claims

  10. editorial

    Repaired renderer-invalid citations and seven claim/source-fit defects; corrected staged rollout, disclosed architecture and training-data scope, tokenizer examples, output limit, and Custom GPT retirement sourcing; restored first-party GPT-4o evaluations, the Studio Ghibli image trend, independent Sky voice analysis, and the documented gpt-image-1-mini lifecycle; and retained defensible removals of unsupported or non-comparable claims.

  11. editorial

    Repaired malformed citations and clarified reported financing and site-specific safety claims

  12. editorial

    Unified the mixed named and numeric citations into one numbered reference system, restored all citation targets, corrected the Whisper training-data breakdown and unsupported product figures, and replaced a speculative legal forecast with the July 28, 2026 litigation status

  13. editorial

    Resolved all undefined citations with ten primary or first-party references, updated version-drifted ReasoningBank and MaTTS results to the ICLR 2026 paper, repaired malformed result tables, and removed unsupported promotional reception claims.

  14. editorial

    Preserved the supported biography, trial, Future Fund, Anthropic-sale, and charitable-recovery history; updated the April and June 2026 post-trial rulings and July 2026 clemency record; removed the unsupported Alameda-alone $500 million and initial 8 percent Anthropic claim; distinguished grant commitments from delivered funds; corrected the two estate transactions to 44,539,240 shares for approximately $1.34 billion; and replaced the stale $350 billion Anthropic term-sheet valuation with the official May 2026 $965 billion post-money valuation.

  15. editorial

    Qualified Oracle's cloud backlog and partnership terms, distinguished announced GPU capacity from deployed hardware, verified Michigan construction status, refreshed corporate-role and policy details, and removed unstable real-time wealth figures.

  16. editorial

    Fixed H100 SXM5 and PCIe dense versus sparsity-assisted Tensor Core throughput, corrected the DGX H100 FP8 aggregate, removed two ambiguously labeled cross-generation throughput rows, and refreshed the two supporting NVIDIA citation targets.

  17. editorial29 articles

    Repaired broken image URLs

    23 more articles
  18. editorial249 articles

    Removed redundant H1 title heading (rendered by the page template)

    243 more articles
  19. editorial

    OpenAI $100B LOI superseded by $30B equity stake ($852B round), added $20B Groq deal and Groq 3 LPU, Q1 FY2027 results ($81.6B), removed departed exec Manuvir Das, fixed 8 minor figures.

  20. editorial

    Sonnet 5 (June 30, 2026) and Gemini table fixes; EU AI Act high-risk deadlines deferred to Dec 2027/Aug 2028; GPT-5 error-rate baseline; ChatGPT, Waymo, consciousness-paper updates.

  21. editorial

    GPT-4 left ChatGPT on April 30, 2025 (replaced by GPT-4o), not at GPT-5's launch; removed fabricated system-card evaluators (ARC is the only one named); fixed SOTA columns and dates.

  22. editorial

    Williams REINFORCE timeline redated to 1992; AlphaGo Zero 4.9M games was the 3-day run (40-day run: 29M); R1-Zero (not R1) skipped SFT; ACM Turing quote restored verbatim.

  23. editorial

    Kavukcuoglu's chief AI architect title dated to June 11, 2025 (was mid-2024) in prose, table, and reference; Lila Ibrahim's role column updated to Chief AI Readiness Officer.

  24. editorial

    Pentagon appeal was filed in the D.C. Circuit, not the Federal Circuit; ASL-3 applied to Opus 4 only; updated for the June 2026 Claude 5 releases and the US export-control suspension.

  25. editorial

    EU AI Act high-risk deadlines updated per the June 2026 Digital Omnibus: Annex III deferred to Dec 2, 2027, Annex I to Aug 2, 2028; Samuel quote and 'deep learning' coinage reattributed.