H-Nets v2 delivers 3-5x efficiency gains, betting on end-to-end learning from raw bytes
niloofar_mire · x · 2026-10-29
Albert Gu and Cartesia unveiled new H-Nets (hierarchical networks) work: learning abstractions end-to-end from raw data instead of handcrafted heuristics like tokenizers, embracing the bitter lesson. The new version shows up to 3-5x efficiency gains in controlled scaling-law experiments and matches an open-source baseline at larger scale. The team argues byte-level end-to-end learning is a central problem for handling all of the world's data, and published a series of blog posts with more to come.
Related event: Cartesia's New H-Net: Byte-Level Modeling Gains 3-5x Efficiency(8 posts)→
More from Research
- LoopVL: 1B-parameter VLM reuses 32 layers across 128 calls, beats 4B baselines on 5/8 benchmarks — jiqizhixin · 2026-10-29
- Geoffrey Irving puts >90% on unbounded-depth CSETH, unsure about finite depth — geoffreyirving · 2026-10-29
- davidad: lead poisoning may explain the 1971–2012 Great Stagnation, checked with GPT-4 — davidad · 2026-10-29
- New AI security benchmarks from Meta collaborators urge community to double down on robustness — maksym_andr · 2026-10-29
- Arena Releases Alignment Index, Arguing Alignment Is a Function of Model Capability — arena · 2026-10-29
- Researchers Push Back on AI Audit of Saez-Zucman Papers Over Missing Prompts — felpix_ · 2026-10-29