A new benchmark called Neo-Classic, introduced in arXiv paper 2609.19154v1, shows that state-of-the-art large language models lose 20 to 50 percent accuracy when moving from historical Classical Chine…
A Google Research and Technion study of over 4 million responses across 13 large language models found that frontier models like Gemini-3-Pro and GPT-5 store facts correctly 95-98% of the time but fai…
Researchers introduced OraRL, a reinforcement learning method for video multimodal large language models that uses annotations as oracle rollouts to improve sample efficiency and scalability. OraRL re…
Google researchers found that frontier LLMs such as Gemini-3-Pro and GPT-5 encode 95–98% of tested facts but fail to directly recall 26–34% of them, with recall failures accounting for more than 70% o…
Google Research found that frontier AI models, including Gemini-3-Pro and GPT-5, encode 95-98% of brand facts in their neural weights but fail to recall 26-34% of them, indicating the bottleneck is re…