About This Wiki
Entry
Transparency & Credibility
PurposeHelp you assess whether to trust this resource
Last UpdatedDecember 2024
Before using this wiki to inform your views on AI risk, you should know who made it, what biases it has, and how to evaluate its claims.
Who Made This?
Section titled “Who Made This?”This wiki is an experimental project exploring how to structure and communicate AI safety arguments. It was created primarily using Claude Code (Anthropic’s AI coding assistant) as an experiment in AI-assisted content creation.
What this means for you:
- The content reflects what an AI system trained on AI safety discourse would produce
- It may inherit biases present in that discourse
- It has been reviewed but not comprehensively fact-checked
- It should be treated as a starting point for exploration, not a definitive reference
Wiki Statistics
Section titled “Wiki Statistics”351
Content Pages
315
Entities Tracked
3133
External Resources
~970K
Words (approx)
| Metric | Value |
|---|---|
| Content pages | 351 |
| Entities tracked | 315 |
| External resources/citations | 3133 |
| Approximate word count | ~970K |
| Average quality score | 4.4 / 5.0 |
| Created | December 2024 |
| Last data build | 1/7/2026 |
Content Breakdown
Section titled “Content Breakdown”| Category | Count |
|---|---|
| Risks & Risk Factors | 61 |
| Responses (Agendas, Interventions, Policies) | 38 |
| Organizations & Labs | 19 |
| People | 28 |
| Analytical Models | 78 |
| Concepts & Capabilities | 38 |
Quality Distribution
Section titled “Quality Distribution”| Quality Level | Count |
|---|---|
| Low (1-2) — Needs improvement | 6 |
| Adequate (3) — Meets basic standards | 19 |
| High (4-5) — Well-developed | 282 |
For detailed quality metrics and entity gap analysis, see the Dashboard.
What Perspective Does This Wiki Represent?
Section titled “What Perspective Does This Wiki Represent?”This wiki is written from within the AI safety / x-risk community perspective. This is a real intellectual community with:
- Shared assumptions (AI could be transformatively powerful, alignment is hard, existential risk is possible)
- Shared vocabulary (TAI, P(doom), alignment, mesa-optimization)
- Shared heroes (Bostrom, Yudkowsky, Russell, etc.)
- Shared funding sources (Open Philanthropy, EA funders)
- Social dynamics that can create groupthink
The wiki tries to present counterarguments, but it does so from within this perspective. The structure of the arguments, the choice of what to cover, and the framing all reflect AI safety community assumptions.
Embedded Assumptions
Section titled “Embedded Assumptions”Assumptions this wiki takes for granted (which are contested elsewhere):
| Assumption | Contested By |
|---|---|
| ”AGI” is a coherent concept | Some ML researchers (LeCun, Marcus) |
| Existential risk from AI is possible | AI ethics community, some ML researchers |
| Theoretical risk analysis is valid before empirical evidence | Empiricists, mainstream ML |
| Long-term future matters morally | Person-affecting ethicists |
| The AI safety research agenda is roughly correct | AI ethics researchers, some ML researchers |
If you reject these assumptions, this wiki may still be useful for understanding what AI safety researchers believe—but you should not treat it as a neutral source.
How Was Content Created?
Section titled “How Was Content Created?”Sources
Section titled “Sources”- Alignment Forum and LessWrong posts
- Academic papers on AI safety
- Public statements from researchers and organizations
- AI safety organization websites
- Forecasting platforms (Metaculus, AI Impacts surveys)
Process
Section titled “Process”- Initial content generated with AI assistance
- Structured around “cruxes” (key disagreement points)
- Reviewed for accuracy against primary sources
- Counterarguments added to major claims
- Limitations acknowledged explicitly
What’s NOT Included
Section titled “What’s NOT Included”- Systematic literature review
- Original research or interviews
- Comprehensive fact-checking
- Peer review by subject matter experts
- Input from AI ethics / critical AI studies perspectives
Known Biases and Limitations
Section titled “Known Biases and Limitations”Coverage Biases
Section titled “Coverage Biases”| Well-Covered | Under-Covered |
|---|---|
| Technical alignment research | AI ethics / fairness |
| X-risk arguments | Mundane AI harms |
| Western/Anglophone perspectives | Non-Western perspectives |
| EA-adjacent organizations | Academic ML safety |
| Arguments for concern | Skeptical/dismissive views |
Structural Biases
Section titled “Structural Biases”- Selection of “cruxes”: The cruxes chosen reflect what the AI safety community considers important, not necessarily what’s actually decision-relevant
- Probability estimates: Numbers are drawn primarily from AI safety-adjacent sources
- Organization coverage: More detail on orgs the AI safety community considers important
- Framing of debates: The “Is AI x-risk real?” debate is framed as a debate, not as one side potentially being correct
Potential Conflicts of Interest
Section titled “Potential Conflicts of Interest”- Created using Anthropic’s Claude, which has interests in AI safety being taken seriously
- Content reflects perspectives likely to increase AI safety funding/attention
- No financial conflicts, but intellectual/tribal ones exist
How to Evaluate Claims in This Wiki
Section titled “How to Evaluate Claims in This Wiki”For Factual Claims
Section titled “For Factual Claims”- Check the sources provided
- Look for the original source, not just citations
- Check if claims are contested by credible critics
- Note the date—AI safety discourse moves fast
For Probability Estimates
Section titled “For Probability Estimates”- Treat as rough intuitions, not rigorous calculations
- Note whose estimates are cited (selection bias)
- Remember these are uncertain by definition
- Your informed estimate is as valid as anyone’s
For Arguments
Section titled “For Arguments”- Check if counterarguments are addressed
- Look for the strongest version of opposing views
- Ask: “What would falsify this?”
- Consider whether the argument structure privileges certain conclusions
For Recommendations
Section titled “For Recommendations”- Consider who benefits from the recommendation
- Check if alternatives are fairly presented
- Ask if the recommendation follows from the arguments
- Consider your own situation and values
How Has This Wiki Updated Over Time?
Section titled “How Has This Wiki Updated Over Time?”A healthy knowledge resource should update based on evidence. Here’s our track record:
Changes Made Based on Criticism
Section titled “Changes Made Based on Criticism”- Added explicit “Limitations & Perspective” section
- Added criticism sections to major argument pages
- Changed fake-precise probabilities to ranges
- Added sources for alternative viewpoints
- Acknowledged non-neutral framing
What Would Make Us Update Further
Section titled “What Would Make Us Update Further”- Evidence of deceptive behavior in AI systems → would increase concern emphasis
- Continued absence of concerning behaviors → would moderate concern emphasis
- Major alignment breakthroughs → would increase optimism
- Alignment techniques failing at scale → would increase pessimism
- Feedback from AI ethics researchers → would improve coverage of that perspective
What We’re NOT Tracking (Limitation)
Section titled “What We’re NOT Tracking (Limitation)”We don’t have systematic tracking of predictions made in this wiki vs. outcomes. This is a significant limitation for assessing calibration.
How to Contribute or Critique
Section titled “How to Contribute or Critique”If You Disagree With Coverage
Section titled “If You Disagree With Coverage”- Open an issue on GitHub explaining the gap
- Suggest specific sources that should be included
- Write a critique we can link to
If You Find Errors
Section titled “If You Find Errors”- Factual errors: Open an issue with correction and source
- Broken links: Open an issue or PR
- Outdated information: Suggest update with new source
If You Have Expertise
Section titled “If You Have Expertise”- AI safety researchers: Help ensure technical accuracy
- AI ethics researchers: Help improve coverage of that perspective
- ML practitioners: Help ground theoretical claims in practice
- Forecasters: Help improve probability estimates
The Meta-Question: Can You Trust AI-Generated Content About AI Risk?
Section titled “The Meta-Question: Can You Trust AI-Generated Content About AI Risk?”This wiki was largely created with AI assistance. This creates an interesting epistemological situation:
Reasons for caution:
- AI systems might have biases toward certain conclusions about AI
- The AI safety community’s perspectives may be overrepresented in training data
- AI assistance may create false confidence through fluent, confident-sounding prose
Reasons this might be okay:
- AI assistance is transparent (you know it was used)
- The content can be checked against primary sources
- Human review and editing occurred
- Explicit acknowledgment of limitations
Our recommendation: Treat this wiki as a well-organized summary of AI safety community views, not as an authoritative source. Use it to navigate the landscape and find primary sources—then form your own conclusions.
Summary: How Credible Is This Wiki?
Section titled “Summary: How Credible Is This Wiki?”| Dimension | Assessment |
|---|---|
| Factual accuracy | Moderate—based on reputable sources but not comprehensively fact-checked |
| Balanced coverage | Low—systematically biased toward AI safety perspective |
| Source quality | Moderate—mix of academic papers, forum posts, and public statements |
| Transparency | High—explicit about limitations and perspective |
| Updatability | High—can incorporate feedback and new evidence |
Bottom line: Use this wiki to understand what the AI safety community believes and why. Don’t use it as your only source on AI risk. Seek out critics and alternative perspectives.