Google/DeepMind

Latest AI news, models and releases from Google/DeepMind. ['A2UI v0.9', 'AI Co-Scientist', 'AI Evaluator', 'AI Mode', 'AI Overviews', 'AlphaEvolve', 'AlphaFold', 'AlphaGenome', 'AlphaQubit', 'Antigravity', 'Auto frame', 'BERT', 'BioBERT', 'Chinchilla', 'ClinicalBERT', 'Computational Discovery', 'Confidential GKE Nodes', 'Co-Scientist', 'Empirical Research Assistance', 'Era', 'ERA', 'Executive LLM', 'Farmscapes 2020', 'Flood Hub', 'frontier AI', 'Frozen v2', 'FunctionGemma', 'Gemini', 'Gemini 1.5 Flash', 'Gemini-1.5-pro', 'Gemini 1.5 Pro', 'Gemini 2.0 Flash', 'Gemini 2.5', 'Gemini 2.5 Flash', 'Gemini-2.5 Flash', 'Gemini-2.5-Flash', 'gemini-2.5-flash-image', 'Gemini 2.5 Pro', 'Gemini-2.5 Pro', 'Gemini-2.5-Pro']

Official site ↗

Research 🇺🇸

Google: Reasoning Helps Language Models Recall Facts Through Buffer and Priming

Google Research finds that enabling reasoning in LLMs helps recall simple facts via two mechanisms: extra tokens act as a computational buffer, and generating related facts primes correct recall. The study tests Gemini-2.5 and Qwen3-32B on QA datasets.

Google/DeepMind

Alibaba/Qwen

Google Research27.07 · 10:04

Research 🇷🇺

Token Cramming: Ideal Text Compression Does Not Preserve Meaning – AIRI Study

Researchers at AIRI demonstrated that the token cramming method, which compresses up to 1568 tokens into a single embedding, does not guarantee meaning preservation: under greedy decoding, accuracy drops from 99.96% to 40%. A new method, progressive cramming, was proposed to achieve 100% reconstruction, but it was found that even a perfectly compressed embedding impairs the model's reasoning ability, and the cause lies in the early layers of the transformer.