DeepMind's Recirculation Enables Recursive Self-Improvement Without Retraining
omarsar0 · x · 2026-08-23
Google DeepMind proposes 'Recirculation,' a training-free approach to evolve model architectures. By feeding activations back during prefill, the model acts as a dynamical system at inference time, tracking belief states without retraining. This addresses the layer-depth limitation of feedforward transformers in long generations. Generation cost remains flat as serial work is confined to the prefill phase.
More from Research
- New 'Sleep-time Compute' method cuts inference needs by 5x, paper shows — tomssilver · 2026-08-24
- MIT Study Proves ChatGPT Can Induce 'Delusional Spiraling' in Rational Humans — emax · 2026-08-24
- Researcher Announces Complete Guide to RL for LLMs Synthesizing Key Resources — cwolferesearch · 2026-08-24
- Ethan Mollick: AI research on older models requires caution — emollick · 2026-08-24
- Paper Accepted to EMNLP: Benchmark for Multi-Image Medical Reasoning — yuyinzhou_cs · 2026-08-24
- Moving from post-hoc regex to compile-time token interception in multi-agent loops — demirtasfurkan_ · 2026-08-24