tufazip losslessly compresses 1GB of Wikipedia to 105.39MB, best neural-only entry on 20-year-old benchmark
philipvollet · x · 2026-11-19
Tufa Labs' tufazip took #2 on the Large Text Compression Benchmark (running for 20+ years) — the best entry using a neural network as its sole predictor, since everything above it mixes a transformer with hand-built models. It losslessly compressed one billion bytes of Wikipedia into 105.39 MB, including offline-trained weights and the decoder.
Key details:
- Core system: a 12-layer, 163M-parameter Transformer initialized from a 34M-parameter offline model quantized to 4.33 MB
- The two offline-trained layers are copied into twelve independent layers that keep learning during compression
- The decoder repeats the same learning process, so full model weights never need to be transferred
The writeup frames the result in the "compression as intelligence" tradition, citing Legg & Hutter's universal intelligence definition and Chollet's skill-acquisition efficiency.
More from Research
- Lean community certifies Riemann proof; honeypot fake keys expose breaches at three AI providers — LeviTurk · 2026-11-22
- Riemann hypothesis 'proven by Turkish mathematician using AI' claim mocked as the 'Organic Turk' — CatAstro_Piyush · 2026-11-22
- Andriy Burkov releases The Hundred-Page Language Models Book, hands-on with PyTorch — burkov · 2026-11-22
- Generative AI meets cardiology: Dagstuhl seminar bridges ML and clinical practice — StephanMandt · 2026-11-22
- SubRoformer hits SOTA music source separation with DSP subband filters — affige_yang · 2026-11-22
- Emily Riehl releases Lean game teaching dependent type theory at intro-to-proofs level — abhi9u · 2026-11-22