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

    Series A investor is Lenovo Capital (not Legend Capital); ~70% throughput wording per StepFun; App Store listing title; Zhu Yibo citations; angel round date (independent verification V1)

  2. editorial

    Bionic hand date (27 Mar 2026 not April), hand cooling (~10 W not 100 W) and DoF figures, fabricated Lu Weibing 99% claim, unsupported citations (cost, MWC 2023, 'nervous and thrilled' quote, sale status), Optimus comparison cells, 6-DoF-per-arm, unsourced connectivity/API/stairs claims, misquoted Lei Jun line; update: 2022-2025 history, third-gen copyright, Xiaomi-Robotics-0, Apr 2026 new generation, IFA 2026 and 98% success rate

  3. editorial

    Tencent Tech disputed (not supported) the one-eighth cost claim; Step-3 context 65,536; AA-sourced rows are an inference; co-founded (independent verification V2)

  4. editorial

    CyberDog weight (3 kg was payload; about 14 kg), CyberDog sensor/ports specs, CyberOne specs (no RealSense, no 96 Nm/kg), factory size (718,000 m2 not 720,000) and automation figure, bionic hand date (March not April 2026) and misattributed 150,000-cycle durability, Lei Jun quotes and Two Sessions proposals (one robotics proposal, not two), MiMo-Embodied deployment claims removed, Robotics-0 benchmark details, Xynova round details, comparison table cells; removed RoboHub tweet quote. Update: Xiaomi-Robotics-1 and U0, 98% factory result, new-generation humanoid at Investor Day, WRC 2026 and IFA 2026, lab history and leadership, robotics investment portfolio

  5. editorial

    2023 seed round (US$200M from HongShan and ZhenFund, not "more than"); Zhou/Wu prior employers per ITIF; K3 subscription pause date (July 19, not July 20); K3 quantization (QAT with MXFP4/MXFP8) and 69 KDA + 24 gated MLA layer mix; minitriton/nano-kpu README attribution; Kimi CLI now archived; star-count ranking; Bessent post date and Caixin signatory date; Moonshot did respond to distillation claims (Huang Zhenxin denial); CAISI K2 Thinking findings; unverifiable Chapter 18C, CITIC and "no response" claims removed; update: K2.8 Preview, Kimi Code Desktop, ARR above US$1B, confidential Hong Kong IPO filing, Anthropic September 2026 report, AISI K3 cyber figures, funding-round investors

  6. editorial

    Founding date (company says early 2022; Cayman entity June 2021, not Dec 2021), founders per prospectus (Yan Junjie, Yun Yeyi), Zhou Yucong Chinese name, parent/subsidiary structure, funding rounds and totals (US$1.56B over 7 rounds), IPO proceeds (HK$4.82B incl. upsize, greenshoe separate), Huang March 2026 remarks (OpenClaw), M2 API price and license, M2.1/M2.5 dates and descriptions, abab6 claims, Video-01/Speech-01/Hailuo AI launch dates, Talkie stats, lawsuit defendants, removed unverifiable claims; update: 2026 share-price swings, STAR Market plan, M3 launch and billing apology, July 2026 placing and convertible bonds, Yan salary pledge, H1 2026 results, H3, MiniMax Code

  7. editorial

    Benchmark table misaligned (AIME 2025 82.9 not 73.0, GPQA-Diamond 73.0 not 67.1, LiveCodeBench 67.1 not 83.7, CNMO 2024 83.7 not 82.9), weights not hosted on GitHub, 'over ten languages' applied to text tokens only, AFD heterogeneous-throughput overstatement, benchmark positioning vs open VLMs/o3/Gemini; update: 25 July 2025 unveiling, MFA details, vision encoder, 800K capacity vs 65,536 context, chip alliance, known issues, Successors (Step 3.5 Flash, Step 3.7 Flash, Step 5 Preview)

  8. editorial

    Terminal-Bench 2.1 described as harder than 2.0 (it scored higher); Opus 4.8 self-report was 74.6% with Terminus-2, not 87.6% with an internal harness; Gemini CLI free personal tier ended 18 Jun 2026; Codex Pro 'up to $200' and Cursor '$20 usage included'/'4x speed' unsourced; Augment 51.8% SWE-bench Pro cited to wrong post; broken Opus 4.8 and Artificial Analysis links; update: Terminal-Bench 4.0 leaderboard, Opus 5.5, Fable 5.1, Opus 5, GPT-6 Astra/Sol, Grok 4.7/Grok Build, Cursor-SpaceX, Devin Desktop/SWE-2, Copilot/Cursor/Claude pricing, open-weight models

  9. editorial

    20+ fonts misattributed to launch post; 'Qwen Studio desktop application' unsupported; Qwen Chat t2i link was in the X post not the blog; unsupported 'Qwen cloud console' claim and lifecycle 'invite-only' attribution; 'internal' AI Arena; 2.0 distribution and 1,000-token citations; uncited 2.1 chart sentence; update: Aug 5 general availability and pricing, Sep 22 API doc (no preview notice, Standard model, regions), LMArena Sep 21 (11th of 79), Artificial Analysis rank, Qwen-Image-2.0 tech report, Qwen-Image-2.1 relation, still no weights or report

  10. editorial

    Removed unsupported claim that Zhang was deputy director of a Tsinghua big-data center (that post was Liu Debing's); added prospectus degrees (BS 2002, MS 2005, PhD 2025); fixed 2,400% stock gain (SCMP: 1,700%+); removed unsupported RMB 40bn valuation; retail oversubscription now 1,159.46x per allotment results; Entity List statement no longer claims 'did not rely on US tech'; fixed misattributed/unverifiable AGI quotes; policy timeline (Jan 2018 policy, 18 months); tutoring broker source; GLM-5.2 row (no 40B active claim, vendor table context); ref 1 is Qiming's release; update: 2025 annual and H1 2026 interim results, GLM-5.3/5.3-Flash, name change to Z.AI Co., A-share progress, 2026 placings and CB, AA26-251A advisory, TIME100 AI 2026

  11. editorial

    Series A investors (angel HongShan/Qiming/IDG, Series A 5Y/Shunwei/Legend), Xiaomi wrongly listed in Series B, Step-3 context (65,536 not 800K) and throughput comparison (4,039 vs 2,324, ~74%), Step 3.5 Flash AIME/HMMT settings and type (language, not multimodal), Step3-VL-10B date (Jan 2026), Galaxy M9 sales window (3 months), Zhu Yibo Chinese name and career order, Zhang Xiangyu not co-founder, Step-1 timeline, headquarters address, unsourced unicorn/valuation/170% claims removed; update: pre-IPO round, joint-stock conversion, reported HK IPO filing, Step 3.7 Flash, Step Plan, StepFun AI app, Step 5 Preview

  12. editorial

    GPT-6 Astra HealthBench values updated to the system card figures OpenAI revised on Sept 22, 2026 (misconfiguration); added GPT-6 Sol and Luna results

  13. editorial

    GPT-6 Sol and Luna released Sept 22 (no-variants sentences dated); system card HealthBench values for Astra revised Sept 22; updated alignment evals noted

  14. editorial

    Competitor cells in the frontier comparison table verified against vendor primaries (Kimi K2.6 is 32B active / 1T total and GLM-5.1 40B / 744B, both fully open-weight, not 'Partial'; GPT-5.5 SWE-bench Verified 88.7% had no OpenAI source and is now n/a; Opus 4.6 GPQA ~91% -> 91.3%; filled Opus 4.7, GPT-5.5, Kimi and GLM prices/scores from vendor sources); GPT-5.4/5.5 context 1.05M; benchmark caption no longer says DeepSeek re-ran all competitor figures; captions dated and sourced, refs 78-92 added

  15. editorial

    Add Muse's September 18, 2026 expansion to Canada (iOS and web, Android coming soon) from the official @Muse posts; fix Mac zero-day timeline (Ars Technica Sep 21, Meta hotfix Sep 22) and cite Ars directly; fix Amazon-block timing, browser-pause trigger, Apptopia 12-day comparison wording, Vibes description, glasses citation, US date for the connectors post, and ref 8 title.

  16. editorial

    Peak hours applied every day when peak/off-peak billing began on 16 Aug 2026; weekends became off-peak from 23 Aug and Chinese public holidays are now off-peak in full

  17. editorial

    Gemini 3.1 Pro list price in the frontier comparison table was ~$3.50/~$10.50; Google's rate is $2.00/$12.00 (prompts up to 200K). Also: Claude Code mapping sourced and clarified, Pro-0813 description, V4.1-Flash MIT citation, flat-rate dating, thinking-mode rule history

  18. editorial

    3FS was introduced on High-Flyer's blog before the SC24 paper (the lead said SC24 was its first public description); network topology, unit wording, Open Source Week and DSec dating fixed

  19. editorial

    The 3FS production result (cache hit rate 40% to 80%, 56% lower TTFT) is a Novita AI testimonial quoted in the LMSYS HiCache blog, not an SGLang-team measurement; linked Fire-Flyer File System

  20. editorial

    Brought DeepSeek V4 up to DeepSeek's change log as of Sep 23, 2026 (V4-Pro-0813 GA on Aug 13, peak/off-peak pricing from Aug 16, weekend off-peak from Aug 23, V4-Flash and V4-Flash-Vision-Exp retired Sep 10, withdrawn V4-Pro retirement); current pricing table with Aug 1 flat rates kept as history; removed stale 29x-34x comparison; replaced BigGo ref with SCMP; ref 59 section range fixed

  21. editorial

    The $1.4 billion (10 billion yuan) valuation was the post-February 2026 figure, not April's; Moz2 is a commercial service robot, not a companion robot; dead RoboArena link removed

  22. editorial

    Per Zhipu's HKEX prospectus, Tang Jie's chief-scientist title is a media usage (prospectus names Zhang Bo), the founding team is six people, Wang Shaolan is deputy general manager, and control rests with a concert-party group; listed English name changed to Z.AI Co., Ltd. in July 2026

  23. editorial

    Chinese legal name is 北京智谱华章科技股份有限公司 (not 知识图谱科技); H1 2026 R&D growth 33.6% per HKEX filing; total funding now includes ~$4B July and ~$5B September 2026 raises. Also: June record price, Entity List wording, unsupported first-Tiger claim removed, Zhang Peng PhD, primary citation for name change

  24. editorial

    H1 2026 R&D expenses rose 33.6% per Zhipu's HKEX interim results (Reuters said 36.6%); the listed company is not a parent of Z.ai; the June 22 HK$2,980 intraday record (not May's HK$1,993) took market value past HK$1 trillion. Also: post-IPO placements, Entity List citation and wording, tutoring filing, employees, Zhang Peng degrees, English name change citation

  25. editorial

    Hong Kong Investment Corporation invested in the May 2026 pre-IPO round (Sina), not the January 2026 Series B+; Zhu Yibo is not sourced as a Microsoft alumnus; Step 3.5 Flash context stated as 256K per model card