Sources
Sources, freshness and caveats
AIFear uses public data where possible and marks unavailable data clearly. The goal is to show directional anxiety signals without pretending the measurement is more precise than it is.
AI Fear Index signals
The headline index combines market, news-attention, prediction-market, incident and public-attention signals. Unavailable signals are excluded until reliable data is available.
| Source | Used for | Freshness | Caveat |
|---|---|---|---|
| AI stock basket | Market / bubble signal and stock pages | Revalidated about every 15 minutes when live market data is configured | The basket is curated rather than exhaustive, so it measures a representative AI-market pulse rather than every AI-adjacent listing. |
| Hacker News Algolia API | AI-news attention signal | Revalidated about hourly | HN attention is a technology-community proxy. It can overrepresent developer and startup concerns. |
| Manifold Markets | Prediction-market signal | Revalidated about hourly | Prediction markets are small and self-selected. They are useful directional signals, not official forecasts. |
| AI Incident Database | AI incidents signal (report velocity over the last 30 days) | Revalidated about every 6 hours via their public feed | Report counts measure documentation activity, not harm itself, and the public feed's ~100-item depth caps extreme readings. |
| Wikimedia Pageviews API | Public-attention signal (views of AI-risk articles vs a 5-week baseline) | Revalidated about every 6 hours; pageview data lags ~2 days | Wikipedia reading is an attention proxy: spikes can reflect curiosity or news cycles, not only anxiety. |
Job-risk inputs
Job pages blend published automation research with task-level occupation data. Scores estimate exposure, not individual outcomes.
| Source | Used for | Freshness | Caveat |
|---|---|---|---|
| O*NET Online | Occupation titles, task profiles and SOC mappings | Updated when the local occupation dataset is rebuilt | Task descriptions are broad occupational profiles. Individual employers and roles can differ materially. |
| BLS Occupational Employment and Wage Statistics | Median pay and employment counts where available | Current dataset uses May 2024 OEWS figures | BLS figures are U.S.-only medians and should not be treated as a salary offer or local wage estimate. |
| Frey and Osborne, The Future of Employment | Automation probability anchor for curated roles | Research anchor | The paper predates modern LLMs, so AIFear blends it with a separate LLM-exposure estimate. |
| Eloundou et al., GPTs are GPTs | LLM task-exposure framing | Research anchor | Exposure is not replacement. A task can be exposed to AI and still require human review or accountability. |
Financial-data posture
Stock data is informational and exists to explain market sentiment around AI. It is not investment advice.
| Source | Used for | Freshness | Caveat |
|---|---|---|---|
| Finnhub | Delayed quote data when FINNHUB_API_KEY is configured | Revalidated about every 15 minutes | If the API key is not configured, the site clearly switches to deterministic demo data. |
Trust rules
- Show the live data mode instead of hiding fallbacks.
- Separate measured signals from planned signals.
- Explain formulas, weights and limitations in public pages.
- Treat job-risk scores as exposure estimates, not predictions about specific workers.
- Treat market pages as sentiment context, not investment recommendations.