A new study by Google Research and Technion shows that frontier models like GPT-5 and Gemini-3 encode 95-98% of tested facts, but fail to directly recall 26-34% of them. Inference-time thinking recovers 40-65% of those facts, suggesting recall, not knowledge, is the main bottlen…
Frontier AI models can recover up to 65% of facts they fail to recall by thinking longer
A new study by Google Research and Technion shows that frontier models like GPT-5 and Gemini-3 encode 95-98% of tested facts, but fail to directly recall 26-34% of them. Inference-time thinking recovers 40-65% of those facts, suggesting recall, not knowledge, is the main bottleneck.
This item was produced with AI assistance under the editorial responsibility of Haydamax OÜ.
Monitoring item. The full text is not distributed. Extract and source below.
Same event, other desks
Story file →
uk
Звіти OpenAI про атаку AI-агентів на Hugging Face змушують переглянути підходи до кібербезпеки
01.09 20:01
en
IKEA invests $1.4 billion in price cuts across Europe to win back cost-conscious shoppers
01.09 20:02
uk
Anthropic випустила Claude Fable 5.1 і Mythos 5.1 зі зниженням вартості кешованого контексту на 75%
01.09 20:03
de
The-Witcher-Remake wartet auf Technik von The Witcher 4
01.09 21:01