Skip to content
This site is deprecated. See the new version.

Track Records

This section documents the epistemic track records of influential figures in AI. Understanding where experts have been right or wrong—and their patterns of over/underconfidence—helps calibrate how much weight to give their current views.

PersonSummaryKey Pattern
Yann LeCunStrong on long-term architectural intuitions; underestimates near-term LLM capabilitiesConsistent skeptic
Sam AltmanDirectionally correct on AI trajectory; overoptimistic on specific timelinesSafety rhetoric vs. deployment tension
Eliezer YudkowskyEarly timeline errors; vindicated on AI generalization; core doom predictions unfalsifiableUpdated from early overconfidence
Elon MuskPrescient safety warnings; consistently missed product timelines by 6+ yearsShifting goalposts

Each track record page documents:

  1. Resolved predictions - Claims that can now be evaluated with ✅/❌/⚠️ status
  2. Pending predictions - Testable claims with target dates
  3. Unfalsifiable claims - Positions that cannot be empirically tested
  4. Accuracy analysis - Patterns of where the person is right/wrong
  5. Position evolution - How their views have changed over time

All claims include source citations where available.