Submission + - Leak-Proof AI Benchmark Posts Biggest Gain in 20 Years, on One CPU Core (spasim.org)
Baldrson writes: The Hutter Prize for lossless compression of human knowledge has paid out nearly €50,000: its largest advance in 20 years: three entries together cut the record for compressing a 1GB Wikipedia snapshot by nearly 10%: a compression ratio of more than a factor of 10. Every entry ran on a single CPU core, in about 50 hours and under 10GB of RAM, so the gain came from algorithms.
Unlike most AI benchmarks, the prize requires no validation data, so a leak can't inflate a score. The full file is public, and the size of the decompressor counts against the entry, so anything memorized is paid for in bits. Vladimir Ivanov's fx2-cmix-T now holds the record, after earlier 2026 gains by Ibrahim Marcouch and Kaido Orav (cmix-lex) and David Freelan (cmix-obias). Sponsor Marcus Hutter, who advised DeepMind co-founder Shane Legg's 2008 thesis "Machine Super Intelligence," calls it "the largest progress in its 20-year history." With trillions riding on claims of AI progress and data-center power demand, is a leak-proof, energy-limited benchmark the yardstick the industry should be watching?
(Disclosure: the submitter is on the Hutter Prize committee.)
Unlike most AI benchmarks, the prize requires no validation data, so a leak can't inflate a score. The full file is public, and the size of the decompressor counts against the entry, so anything memorized is paid for in bits. Vladimir Ivanov's fx2-cmix-T now holds the record, after earlier 2026 gains by Ibrahim Marcouch and Kaido Orav (cmix-lex) and David Freelan (cmix-obias). Sponsor Marcus Hutter, who advised DeepMind co-founder Shane Legg's 2008 thesis "Machine Super Intelligence," calls it "the largest progress in its 20-year history." With trillions riding on claims of AI progress and data-center power demand, is a leak-proof, energy-limited benchmark the yardstick the industry should be watching?
(Disclosure: the submitter is on the Hutter Prize committee.)
