Continuous diffusion LM RePlaid accepted at NeurIPS 2026, matches discrete diffusion scaling
ArashVahdat · x · 2026-09-25
The paper "Continuous Diffusion Scales Competitively with Discrete Diffusion for Language" has been accepted at NeurIPS 2026. The team introduces RePlaid, a continuous diffusion language model (DLM) featuring a discrete likelihood bound and scaling laws competitive with state-of-the-art discrete DLMs. Co-authors include ArashVahdat, Yongxin Chen and others; the paper is publicly available.
More from Research
- DARLING: diversity-aware RL beats standard RL on both quality and diversity — DanielKhashabi · 2026-09-25
- TTIC launches LEIF Lab to study Transformer expressivity via formal language theory — lambdaviking · 2026-09-25
- Researcher lands 4 NeurIPS papers including one Oral, spanning synthesis to protein diffusion — abeirami · 2026-09-25
- Calibration-Free Quantization Method TQ Open-Sourced, Hits 92.4% Top-1 on Qwen 27B 4-bit — textclf · 2026-09-25
- Yoav Goldberg: LLM reasoning traces are 'too good' — unclear how they emerge from RL — yoavgo · 2026-09-25
- Neuralese Recurrence Explained: Why OpenAI's Astra Loop Is Not Yet Unreadable AI Thought — Astral Codex Ten · 2026-09-25