Corrections Policy
How we handle errors and maintain accuracy in AI journalism. Transparency and accountability guide our process.
Our Commitment to Accuracy
We strive for accuracy in all reporting. When errors occur, we correct them promptly, with clear notation and timestamps.
Types of Corrections
Factual Errors
Incorrect information about dates, names, model specs, metrics, datasets, or other facts.
Response: Immediate correction with a clear note of the change.
Clarifications
Content that is technically accurate but potentially misleading or missing key context.
Response: Add clarifying language or additional context.
Updates
New information significantly changes the story or adds crucial context (e.g., revised paper, new benchmark).
Response: Update with timestamped note summarizing the change.
Our Correction Process
Error Identification
Errors are flagged via reader feedback, internal review, or routine fact-checks.
Verification
We confirm corrections against primary sources and authoritative references.
Correction Implementation
We amend the content and append a clear correction note with timestamp.
Reader Notification
For significant errors, we may notify readers via social channels or email updates.
How We Mark Corrections
Correction (August 12, 2025, 3:15 PM UTC):
This article originally stated that the model trained on 1T tokens; the correct figure is 900B tokens. We regret the error.
Example placement and tone for corrections
How to Report Errors
Contact Information
Report issues or propose corrections at: editorial
Please Include:
- The article URL
- Specific description of the error
- Correct information with supporting sources
- Your contact information (optional)
Response Times
Critical Errors
Within 2 Hours
Standard Errors
Within 24 Hours
Clarifications
Within 48 Hours