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.
- editorial
A study finding that contradicts the cited paper, the Rockwell acquisition date, university course platforms, the TurtleBot3 Friends list, and the TurtleBot 4 announcement sourcing
- editorial
A CNY 10 billion gap given as CNY 1 billion, SuperPoD figures used for the SuperCluster, a 5-to-10-years claim absent from its citation, chip-volume figures in neither cited source, and sourcing for the Ascend 960DT date
- editorial
Remove an unsourced AMD MI300 parity claim and an unsupported claim about the rate of new architecture support
- editorial
Remove three uncited claims (an engine-spread attribution absent from the cited post, an sm_121 prediction written as fact, and a sweeping claim that three unrelated lines of work build on the FA3 codebase)
- editorial
The Open LLM Leaderboard backend claim is past tense, an unsourced superlative removed, and a misattributed motivation
- editorial
Invented throughput figures and claims attributed to sources that do not contain them, a stale current release, a pipeline-parallelism feature that does not exist, MXFP4 scope, FlashMLA day-one timing, and model-table errors
- editorial
The FP8 ablation table had its two rows reversed and the conclusion drawn from them was backwards; register reallocation is dynamic; several figures re-sourced from the NeurIPS camera-ready rather than the PyTorch blog; FA4 and FlexAttention claims corrected; the FlashMLA inheritance claim reduced to what its README says
- editorial
Link Gemini 3.1 Pro to its own page rather than the Gemini 3 article
- editorial
The $18 billion Colossus figure is October 2025 reporting on Colossus 2, after Grok 4 was trained
- editorial
Remove a vendor-sourced PwC comparison and an uncomparable hallucination-rate row
- editorial
Remove an uncited sick-days claim from the enterprise-adoption block
- editorial
The model card is a safety report with no architecture content, a self-refuting Opus 4 pricing comparison, wrong retirement slugs, an unshipped Grok 4 Code product with invented benchmark figures, and an uncited Colossus spend
- editorial
Gemini 3.1 Pro has a 1M context window not 2M, plus fabricated attributions to Pichai and VentureBeat, the Windsurf licence described as an acquisition, wrong Priority pricing and several misattributed benchmark scores
- editorial
Three benchmark figures are in the cited model card, reversed temperature guidance, a safe-rate figure not in Google's model card, and a contradictory MAU sentence
- editorial
The lead stated the BMW deployment twice and in the company's voice; condense to one sentence attributed to the outlet, and attribute the positioning claim to the outlet rather than to Figure
- editorial
Remove an uncited count of available MCP servers
- editorial
Remove an unsourced IPO-window forecast, a net-worth sentence duplicated from a later section, and family details resting on a citation with no URL
- editorial
Remove four claims that no cited source supports (a market-size forecast, a monthly production estimate, a 50,000-unit run-rate target and an operating-system row)
- editorial
A fabricated attribution, agent list, usage figures resting on an unreadable source, competitor comparison rows, a reference that supports neither claim, and a mis-wired internal link
- editorial
Install count, final release version, a misattributed account and quote, provider count, Cline codebase-indexing claim, Cursor MCP row, rating figures and dead references
- editorial
A misquote, an unsourced customer list, wrong release dates and install and rating figures, a dead reference, and Roo Code shown as live after its May 2026 shutdown
- editorial
Archer departure reason, industry count, Vettery date and an unsourced IPO-window forecast
- editorial
The BMW engagement is a logistics sequencing demonstration per BMW and Figure, not a 40-unit body-shop and assembly contract; plus market-size, production, skills, quote and funding-round corrections
- editorial
Point the Sierra links at the existing sierra_ai page
- editorial
The Holistic Agent Leaderboard paper's first author is Sayash Kapoor