Summary
Forecast views
The timeline, two ways
Read every claim in sequence, then inspect how far resolved claims cleared or missed their stated test.
Timeline ledger302 claims
2025 · Q310 total✅ 8 · ❌ 22025 · Q412 total✅ 5 · ❌ 72026 · Q212 total✅ 9 · ❌ 32026 · Q414 total✅ 1 · ⏳ 13
Claim ledger
Every claim that can be checked
Vague predictions stay out. Direct measures, declared proxies, and bounded inferences stay in.
408 of 408 claims
Method
What gets counted
The tracker keeps testable measures of AI progress. Scenario politics, popularity, and other reactions stay out.
01Extract the claim
Preserve the prediction’s date, scope, threshold, and branch. Restate it plainly without making it easier to satisfy.
02Write the test first
Set the evidence rule before looking at the outcome. Use a proxy only when the direct quantity is private or undefined.
03Attach public evidence
Prefer primary records and stable benchmark reports. Every resolved status points to the observation behind it.
04Show the working
State the proxy, assumptions, and arithmetic. Use bounded inference when it can resolve a claim, and reserve “unclear” for a real impasse.
Primary sources
Read the forecasts in full
This project is independent of both author groups. Page references follow the publisher PDFs.
01
AI 2027
Daniel Kokotajlo, Scott Alexander, Thomas Larsen, Eli Lifland, and Romeo Dean
A concrete scenario tracing agent adoption, frontier compute, automated AI research, and two possible endings through 2027.
02
Situational Awareness: The Decade Ahead
Leopold Aschenbrenner
An essay series forecasting capability gains, an intelligence explosion, industrial buildout, and government mobilization.