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.
- editorial
Replace a speculative model catalogue and inaccurate claims about active compute, capacity, memory, specialization, and speed with a focused, sourced account of MoE architecture, routing, training, inference, and evaluation.
- editorial
Replace an overbroad survey, architecture-plus-weights definition, unsupported compression claims, weak model-system boundaries, and overstated reproducibility guidance with a sourced object-level account.
- editorial
Separated DeepSeek the company from its models and hosted products; corrected the V3 training-cost claim; replaced fabricated roadmap, unaudited finance, promotional benchmark, and blanket open-source assertions; and bounded privacy, regulatory, distillation, and hardware claims.
- editorial
Updated Microsoft's OpenAI partnership through April 2026; replaced stale leadership, exclusivity, financial, spending, market, and product claims; and clarified corporate scope, first-party AI, infrastructure, safety, regulation, and environmental impacts.
- editorial
Replace stale release, adoption, and ecosystem catalogues; inaccurate static-versus-dynamic framing; active TorchScript treatment; redirected links; and weak security and reproducibility coverage with a durable systems account.
- editorial
Replace volatile adoption and benchmark claims, unreleased-model speculation, repeated generation catalogues, broken links, and inaccurate openness, architecture, data, and context statements with a durable family history and deployment guide.
- editorial
Separated Google LLC from Alphabet and Other Bets; replaced incorrect 2025 financial figures and volatile product, benchmark, user, and market claims; and clarified corporate scope, AI organizations, governance, privacy, competition, and environmental impacts.
- editorial
Standardize the visible title and replace unsupported market forecasts, product catalogues, performance specifications, prices, and generalized capability claims with a source-audited account of humanoid-robot definitions, history, mechanics, control, learning, evaluation, deployments, safety, standards, and limitations.
- editorial
Replace market forecasts, product and framework catalogues, volatile rankings, malformed references, and unsupported autonomy claims with a primary-source systems account of agent scope, architecture, planning, tools, memory, multi-agent coordination, evaluation, security, oversight, and reproducible deployment.
- editorial
Standardize the visible title and replace unsupported release logs, vendor rankings, cost estimates, forecasts, regulatory equivalences, and capability claims with a primary-source account of foundation-model scope, history, training, adaptation, evaluation, risks, openness, and current governance.
- editorial
Separated the consumer assistant from Meta Platforms, FAIR, and Meta Superintelligence Labs; replaced unsupported growth, performance, licensing, and release claims with dated primary and independent sources; and clarified model lineage, privacy, safety, and regional availability.
- editorial
Standardize the visible title and replace unsupported product, market, parameter, and adoption claims with a primary-source account of NLP scope, history, tasks, methods, data, evaluation, applications, and limitations.
- editorial
Replace unsupported market forecasts, release claims, human-level generalizations, and unsourced architecture comparisons with a primary-source account of vision formulation, historical methods, learning paradigms, tasks, datasets, metrics, 3D and video, applications, failure modes, and reproducible evaluation.
- editorial
Replaced unsupported historical and performance claims, separated causal, masked, and denoising objectives, corrected tokenization and perplexity explanations, and rewired the article to verifiable academic sources.
- editorial
Normalize the visible AI in media title to the required Title Case without changing article prose, categories, or fact-check state.
- editorial
Normalize the visible Groq hardware title to the required Title Case without changing article prose, categories, or fact-check state.
- editorial
Normalize the visible AI chip title to the required Title Case without changing article prose, categories, or fact-check state.
- editorial
Normalize the visible AI accelerator title to the required Title Case without changing article prose, categories, or fact-check state.
- editorial
Replace unreliable history, vendor rankings, adoption forecasts, originality, legal, and environmental claims with a primary-source account of generative model families, training, modalities, evaluation, risks, copyright, provenance, and current governance.
- editorial
Rebuilt the company history, financing, Hub, software, hosted services, licensing, moderation, research, and security coverage with 40 explicit sources; removed unsupported private-company figures, duplicate sections, and pseudo-infobox markup; corrected repository-access and open-source distinctions; replaced broken and repeated internal links; and added the qualified July 2026 incident disclosures.
- editorial
Separated the USTC/Metastone benchmark from the similarly named FutureSearch project, replaced unsupported claims with primary-source findings, documented the current evaluator generation, and rewired the article to authoritative sources.
- editorial
Distinguish FutureSearch's Deep Research Bench from the separate DeepResearch Bench, correct RetroSearch and trace-statistic descriptions, date initial results, and replace unsupported promotional claims with primary-source evidence.
- editorial
Separate methodology from the paired Time Series survey, distinguish stochastic moving averages, qualify stationarity and cross-validation claims, and add sourced diagnostics, uncertainty, intervention, and change-point guidance.
- editorial
Retitle this data-concept page Time Series; add reciprocal scope with Time Series Analysis; preserve and extend verified data-structure, irregular-sampling, task, model-history, competition, and benchmark coverage; and reserve specification, diagnostics, validation, inference, intervention, and structural change for the companion article.
- editorial
Redirect the duplicate NVIDIA HGX page to the consolidated canonical article.