A €500,000 prize for losslessly compressing a 1 GB snapshot of English Wikipedia. The premise: better compression = better world-model = better AI. Compression and prediction are formally equivalent.
Key facts
Optimal compression length ≈ Shannon entropy of source: L≥H(X).
Compressor that predicts P(xn+1∣x1..n) well achieves codelength −log2P via arithmetic coding.
Prize threshold: beat the previous record by ≥1%.
Hutter prize essentially states that AI and compression are the same problem