Google DeepMind's Recirculation cuts perplexity by 23% without retraining
TheTuringPost · x · 2026-08-21
Google DeepMind introduces "Recirculation," an inference-time architectural enhancement designed to prevent Transformers from losing context.
Core Mechanism: It passes a portion of the deeper layers' contextualized representations back to earlier layers for processing subsequent inputs. The model weights remain frozen; only the inference flow changes.
Performance:
- Reduced contextualization errors by 60%
- Reduced perplexity by 23%
- Improved GSM8K accuracy by 21%
Difference from CoT: Unlike Chain-of-Thought, which adds computation via generated reasoning tokens, Recirculation reuses the model's internal understanding for state tracking rather than explicit reasoning.
Related event: DeepMind's Recirculation boosts LLM reasoning without retraining(10 posts)→
More from Research
- UCSD Researchers Propose Dataset Watermarking Detectable via Black-Box API Access — yuxiangw_cs · 2026-10-16
- EmbodiedEvo tops RoboDojo as agents inherit reusable skills and lessons — Aiden_Tech_Ai · 2026-10-16
- CHI reviewer laments: papers' Discussion sections now cite nothing — IanArawjo · 2026-10-16
- Casper: Understanding 90% of neurons ≠ understanding the network — StephenLCasper · 2026-10-16
- NVIDIA Launches cuPotential in CUDA-X to Bring AI Agents to Semiconductor Materials Simulation — tomaszbednarz · 2026-10-16
- Agentic Search Systems Are Rewarded Only for Final Answers, Raising Faithfulness Concerns — burkov · 2026-10-16