Study: Mask-based Diffusion Language Models Develop a Subliminal Clock
s_scardapane · x · 2026-07-23
A new study reveals that mask-based diffusion language models (DLMs) develop a subliminal mechanism to encode denoising progress in a low-dimensional latent structure, despite receiving no explicit timestep. Researchers demonstrated that this internal clock can be predictably steered, allowing observation of how the model corrects perturbations across layers.
Related event: Study: Diffusion Language Models Have Subliminal Clocks(2 posts)→
More from Research
- LightMTP: lightweight latent multi-token prediction boosts LLM planning in pretraining — gdm3000 · 2026-10-27
- EMNLP study: AI literary translation reads fine but varies more, even within one book — tuvllms · 2026-10-27
- Lab monkeys at $30-80K each: a 4-year breeding cycle is the real preclinical drug bottleneck — zebird0 · 2026-10-27
- OpenAI Foundation funds UNC tumor antigen project, hiring senior experimental scientist — iskander · 2026-10-27
- Steering attention heads toward honesty: up moral patienthood, down psychopathy — edelwax · 2026-10-27
- Are transcoders overkill? New paper argues plain neurons suffice for circuit analysis — burny_tech · 2026-10-27