Making LLMs more accurate by using all of their layers

Google Research introduces SLED (Self Logits Evolution Decoding), a decoding strategy that improves LLM factual accuracy by using all layers of the model rather than just the last layer. It requires no external data or fine-tuning and was featured at NeurIPS 2024.

From the source
We introduce SLED, a decoding strategy that enhances the accuracy of LLMs by aligning their output with the model’s intrinsic knowledge, without the need for external data or additional fine-tuning.
research.google

Visit source ↗research.google

Announced: 17 Sep 2025 17:00 UTC
Detected: 03 Aug 2026 21:12 UTC (+320d after)

Other stories

Google Research · 27 AugPlanetary prediction engine: Automating global models via Earth AIResearch paper

OpenAI · 27 AugBetter answers, broader thinking: What students gain from ChatGPT and critical-thinking trainingResearch paper

LG AI Research · 27 Aug[ACL 2026] ReSQL: A Self-Improving Text-to-SQL Framework through Retrieval-Augmented Error ReasoningResearch paper

xAI · 29 Aug

Grok Bot now works with X

Capability change

GitHub · 28 AugGitHub Copilot in Visual Studio — August update

GitHub · 28 AugGitHub Copilot weekly releases — August 24Capability change