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 volatile promotional, adoption, funding, release-timeline, and universal benchmark claims with a primary-source account of vLLM's scope, PagedAttention, V1 architecture, interfaces, execution features, deployment, evaluation, security, and limitations; align the platform, preemption, and shared-memory wording with the exact v0.26.0 documentation.
- editorial
Replace volatile popularity, funding, adoption, comparison, and release claims with a July 28, 2026 version-scoped account of Scikit-learn's history, estimator API, workflows, evaluation, computing, persistence, governance, and documented limitations.
- editorial
Rebuild Gradient Descent with assumption-bounded convergence results, renderer-safe formulas, exact distinctions among deterministic, stochastic, accelerated, and adaptive methods, and fully mapped primary and academic citations; explicitly distinguish Nesterov acceleration from basic gradient descent in the convergence table.
- editorial
Replaced outdated release, universal AI and speedup claims, and conflated hardware material with a primary-source account of CUDA's model, memory, toolchain, libraries, compatibility, licensing, and portability. Correct the PyTorch CUDA Semantics source-update date to the official July 17, 2026 record.
- editorial
Replace volatile pricing, roadmaps, comparison tables, and universal performance claims with a primary-source account of GPU architecture, programming, AI workloads, measurement, distributed scaling, and limitations. Correct the Owens survey record, state the Volta Tensor Core operation precisely, and identify the peer-reviewed Megatron-LM and ZeRO versions of record.
- editorial
Replace speculative project and operator catalogs with a standards-based Data Center article; correct the Tier availability mapping; distinguish measured load from capacity and projections; document PUE, WUE, AI rack-density, reliability, grid, water, security, and reporting boundaries using primary and academic sources. Correct a misattributed DieHard reference and limit its claim to the paper's 2016 correlated power/network failure-domain study.
- editorial
Rebuilt the article around functional dataset roles, defensible splitting, leakage control, quality, representation, foundation-model corpora, privacy, security, governance, and reproducibility; removed arbitrary ratios and volatile legal claims.
- editorial
Replace duplicated and weakly sourced material with a research-grounded account of prompt design, evaluation, reasoning methods, limitations, security, multimodal use, and context engineering; qualify inconclusive self-enhancement evidence.
- editorial
Replace unsupported and volatile benchmark claims with source-bounded guidance; remove contradictory BetterBench counts, qualify LLM-judge self-enhancement evidence, correct GAIA and Tau-bench publication records, and add construct-validity, uncertainty, contamination, resilience, and agent-evaluation coverage.
- editorial
Distinguish the perplexity metric from Perplexity AI; repair the metric's bounds, tokenization, context-window, pseudo-perplexity, speech, topic-model, scoring-rule, and AI-detection claims; replace unsupported company figures and volatile product assertions with dated primary-source or explicitly attributed evidence.
- editorial
Replace unreliable prominence, paper-history, attendance, oral-count, acceptance-rate, review-consistency, and future-impact claims with edition-specific primary evidence; distinguish main, Datasets and Benchmarks, and Position Paper tracks; add the AI Events category.
- editorial
Distinguish loss, risk, objective, regularization, and metrics; repair equations and optimization claims; replace unsupported examples with source-bound coverage.
- editorial
Replace categorical train-test-gap, sample-size, parameter-count, regularization, double-descent, benign-overfitting, and grokking claims with a source-scoped treatment of empirical risk, valid evaluation, leakage, shift, and modern interpolation.
- editorial
Refocus Tesla (robotics) on the verified Optimus program, distinguish demonstrations from autonomous operation and designed capacity from production, and remove unsupported FSD, pricing, deployment, and Gen 3 claims.
- editorial
Replace promotional and unsupported market claims with source-backed coverage of lidar physics, architectures, calibration, processing, applications, standards, safety, history, and documented limitations.
- editorial
Distinguish ChatGPT routing from API models; fix context, modality, benchmark, safety, and lifecycle claims.
- editorial
Rebuild the overview around model-specific generations, licenses, weight and API access, training disclosures, architecture changes, and bounded independent evaluation.
- editorial
Remove unsupported product, benchmark, architecture, and compute claims; rebuild the overview around sourced modality definitions, architectures, training, evaluation, failure modes, and reporting practice.
- editorial
Align the bounded 2024 battery-handling account with Tesla's primary shareholder update, which described the internal task as autonomous while publishing no reliability or intervention rates.
- editorial
Render 32 inline equations with the site's supported math delimiters while retaining the independently reviewed factual and citation corrections.
- editorial
Reconstruct Recurrent Neural Network from 30 primary, peer-reviewed, and official sources; correct the Jordan, Mamba, and echo-state bibliographic records; narrow historical and implementation claims; preserve identity and categories.
- editorial
Fix rendered equations, qualify a generative benchmark result, add primary support for multimodal connector examples, and replace an inaccessible citation URL.
- editorial
Rebuild the Tesla Optimus article around verified prototype milestones, demonstration controls, internal trials, production preparation, safety boundaries, and current commercial status.
- editorial
Replace the legacy Boston Dynamics article with a source-led reconstruction that fixes Atlas chronology, ownership and manufacturing status, early history, legacy-platform claims, current products, and responsible-use coverage.
- editorial
Fix one grammar error and describe the Conformer block order precisely; retain the reviewed article structure, sources, links, identity, scope, redirect, and categories.