H-Net update shows 3-5x efficiency gains in raw-byte language modeling
omarsar0 · x · 2026-10-29
Albert Gu argues intelligence should come from simple general building blocks learning abstractions directly from raw data — the bitter lesson applied. The new H-Net achieves up to 3-5x efficiency gains in controlled scaling-law experiments and at larger scale matches open-source tokenizer baselines, a step toward LLMs that learn end-to-end from bytes without BPE tokenization.
Related event: Cartesia's New H-Net Learns from Raw Bytes, 3-5x Efficiency Gains(12 posts)→
More from Models
- Model trained via Tinker beats rival on finance benchmark at 9% of the cost — simonguozirui · 2026-10-29
- Uncensored GGUF Quant of LiquidAI's LFM3-1.7B Trends on Hugging Face — AtomicChat · 2026-10-29
- Claude Pro vs ChatGPT Plus at $20: robotics dev crowdsources which holds up — Wild_Shower6854 · 2026-10-29
- Odyssey-3 world models now simulate general environments in real time with higher physical fidelity — Scobleizer · 2026-10-29
- Qualia Beats Flora and GPT-6 Astra Across Three Robot-Data Benchmarks — lukas_m_ziegler · 2026-10-29
- Particle-1 inference goes live on OpenRouter, claiming 28-52% fewer output tokens — ycombinator · 2026-10-29