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

    Fix Chinese railroad workforce proportions and the Louis Agassiz campus-feature description; align institutional claims with authoritative sources and set the research cutoff to July 28, 2026.

  2. editorial

    Fix dataset counts and item-weighted scoring, remove unsupported future leaderboard claims, and clarify protocols, data quality, contamination evidence, and distinct variants.

  3. editorial

    Replace unsupported universal hardware, cost, and performance claims with independently verified coverage of fine-tuning methods, evidence, evaluation, reproducibility, safety, privacy, and failure modes.

  4. editorial

    Document the stable 2026-07-28 protocol, replace obsolete release-candidate and session behavior, and independently reverify architecture, security, governance, and adoption claims.

  5. editorial

    Replace promotional and unsupported market, autonomy, product, and financial claims with a source-bounded company history, product-family overview, control-mode distinctions, the latest audited full-year business data, and documented safety and policy context through 28 July 2026.

  6. editorial

    Rebuild Figure AI from checked primary, customer, regulatory, legal, independent, and academic sources; distinguish company claims from corroborated results; correct dates and source scope; and normalize citations in first-use order.

  7. editorial

    Replace mixed corporate, litigation, market, and model claims with a checkpoint-specific primary-source account of the Stable Diffusion pipeline, release genealogy, training disclosures, licenses, safeguards, adaptation methods, and reproducibility limits.

  8. editorial

    Replace the legacy glossary with a sourced, scoped navigation glossary; correct NIST source identity and tighten definitions for dropout, hallucination, logistic regression, random forests, recommender systems, and tensors.

  9. editorial

    Distinguish full ImageNet from ILSVRC/ImageNet-1K, correct the 2012 data condition and 2013 winner, qualify human comparisons, and reconcile benchmark, curation, privacy, rights, and access claims with primary sources.

  10. editorial

    Replace unsupported named footnotes, volatile citation and adoption claims, and mixed operator-kernel-serving taxonomies with a primary-source account of attention, its mathematics, variants, efficiency tradeoffs, applications, and interpretability limits.

  11. editorial

    Apply title case; distinguish the moving descriptive frontier, model and system evaluation, risk-management usage, and jurisdiction-specific legal definitions; remove unsupported product rosters, obsolete US policy, and false universal compute thresholds.

  12. editorial

    Replace unsupported API-size mappings, pricing, application, cost, adoption, and emergence claims with a primary-sourced account of GPT-3's research family, architecture, training mixture, evaluation protocol, historical results, contamination, prompt sensitivity, limitations, deployment history, lifecycle, and boundaries from later OpenAI models.

  13. editorial

    Replace an oversized, volatile catalogue of models, pricing, benchmark claims, company estimates, and future projects with a source-bounded company article covering Mistral AI's identity, founders, financing, mixed licensing, model strategy, distribution, enterprise products, compute infrastructure, acquisition, regulation, safety transparency, environmental reporting, and unresolved evidence limits.

  14. editorial

    Distinguish configuration mobility, task-space rank, actuation, statistical residual and distributional degrees of freedom, effective model degrees of freedom, and two-degree-of-freedom control terminology while removing volatile robot-specification catalogues.

  15. editorial

    Replace an overextended catalogue with a source-bounded account of RLHF terminology, history, human feedback, reward modeling, PPO and non-PPO policy optimization, adjacent methods, evaluation, and limitations.

  16. editorial

    Replace unsupported current-adoption, citation-count, cost, ranking, tokenization, and model-family claims with a primary-sourced account of BERT's historical context, architecture, input representation, objectives, original training and results, fine-tuning behavior, interpretation evidence, limitations, lineage, and released implementation.

  17. editorial

    Replace a vendor-heavy catalogue with a source-bounded account of context-window meaning, token accounting, attention and cache costs, extension methods, effective context, evaluation, and application policy.

  18. editorial

    Replace volatile adoption, market, ranking, and product-catalogue material with a primary-sourced technical account of TensorFlow's execution model, differentiation, input pipelines, distribution, serialization, compatibility, security, governance, licensing, and project boundaries.

  19. editorial

    Replace a volatile vendor catalog and universal tuning claims with a source-bounded account of RAG architecture, design, evaluation, failure modes, long-context tradeoffs, and security.

  20. editorial

    Replace a vendor-heavy and volatile survey with a source-bounded account of embodied AI scope, agent-environment coupling, learning and control, simulation, generalist policies, evaluation, safety, and evidence limits.

  21. editorial

    Replace a volatile, overbroad catalog with a source-bounded account of AI safety scope, risk classes, controls, lifecycle management, governance, and evidence limits.

  22. editorial

    Replace a volatile product and market catalogue with a focused, primary-sourced account of robotics systems, mechanics, sensing, planning, control, learning, human interaction, evaluation, safety, deployment, and history.

  23. editorial

    Replace unsupported invariance, feature-hierarchy, architecture-comparison, benchmark, medical, and training claims with a sourced operator-level account that distinguishes conditional equivariance, effective receptive fields, measurement, and deployment limits.

  24. editorial

    Replace an overbroad, internally mixed survey with a source-bounded account of diffusion formulations, sampling, conditioning, discrete extensions, applications, evaluation, and limitations.

  25. editorial

    Replace unsupported LaTeX delimiters with the wiki renderer's supported math syntax; factual content and citations are unchanged.