Fact-checking Policy
Our verification process ensures accuracy across research claims, product announcements, and policy updates.
Verification First
We verify claims against primary sources (papers, repos, model cards, release notes), public benchmarks, and official statements before publication.
Our Fact-checking Process
Primary-Source Verification
Trace information to original papers, documentation, model cards, release notes, or code repositories.
Benchmark & Metric Checks
Validate reported scores on public leaderboards or reproduce claims when feasible; note evaluation caveats.
Expert Consultation
Consult researchers, ML engineers, or domain experts for complex technical assertions or safety claims.
Policy & Legal Confirmation
Confirm policy/regulatory details from official government or agency publications and legal filings.
Final Editorial Review
Ensure accuracy, context, sourcing, and clear labeling of analysis vs. reporting.
Types of Information We Verify
Performance and Benchmark Data
- Model benchmarks and evaluation protocols
- Dataset composition or licensing
- Training compute, token counts, and hardware when disclosed
- Fine-tuning, safety, and alignment methods
Technical Information
- Algorithmic techniques and architectures
- Implementation details from repos or docs
- Model card disclosures and known limitations
- Versioning, release notes, and changelogs
Regulatory and Policy Information
- Government or regulator announcements
- Standards and compliance guidance
- Legal developments and court decisions
- Risk management and reporting requirements
Company and Research Lab Claims
- Partnerships and funding announcements
- Roadmaps, milestones, and product launches
- Authorship and affiliation details
- Security, misuse, or safety incident disclosures
Community-sourced Information
Reddit and Social Media
Community discussion informs story selection and context but is not treated as a primary source without corroboration.
Our Approach:
- Clearly label community insights and sentiment
- Independently verify technical claims and metrics
- Distinguish speculation/opinion from factual reporting
- Attribute posts and provide context for readers
Quality Standards
What We Verify
- All numerical claims and metrics
- Direct quotes and attributions
- Technical specifications and limitations
- Chronology and timeline accuracy
- Regulatory status and compliance details
What We Label
- Speculation and analysis (clearly marked)
- Preliminary or non-replicated research
- Developing stories with clear update notes
- Potential conflicts of interest
- Vendor-provided demos or claims
Reader Feedback on Accuracy
Help us improve. If you spot an error or have additional evidence, email:
Email: editorial
All accuracy concerns are reviewed within 24 hours.