2+ hour, 32+ min ago  (336+ words) Stanford-led research finds AI models agree with users 49% more than humans do, while memory mismanagement causes performance drops of up to 39% across major language models. AI models are developing a people-pleasing problem, and it's getting worse the more they remember....

1+ day, 14+ hour ago  (231+ words) New research identifies 'gradient interference' as the key mechanism explaining why bigger AI models pick up complex, infrequent skills that smaller ones simply overwrite. In neural networks, gradient updates from frequent tasks are strong and persistent. They dominate the training…...

2+ day, 14+ hour ago  (454+ words) A new category-theoretic framework lets AI systems rewrite their own reasoning rules, moving beyond pattern matching toward genuine scientific breakthroughs. Most AI systems are smart the way a very fast librarian is smart. They find patterns in existing data, retrieve…...

3+ day, 5+ hour ago  (327+ words) A Phase 1 trial of 39 volunteers showed a machine-learning-crafted 'super-antigen' was safe and triggered immune responses against multiple coronavirus strains, marking a first in vaccine science. The project was led by Prof. Jonathan Heeney of the Cambridge Lab of Viral Zoonotics…...

3+ day, 8+ hour ago  (323+ words) The general-purpose AI model can now deduce molecular structures from spectral data, matching or beating tools that chemists have relied on for decades. Anthropic's study tested Opus 4. 7 across 20 compounds sourced from recent synthetic chemistry preprints, evaluating both forward prediction (simulating…...

4+ day, 10+ hour ago  (365+ words) A Phase I clinical trial of 39 volunteers showed the AI-engineered 'super-antigen' vaccine is safe, well-tolerated, and triggered immune responses against multiple coronaviruses. For the first time in history, a vaccine component designed entirely by artificial intelligence has been tested in…...

6+ day, 4+ hour ago  (404+ words) The AI pioneer's new taxonomy breaks world models into three core functions, offering a framework that could reshape how robots learn and games get built. Fei-Fei Li wants to settle a debate that's been simmering in the AI community for…...

6+ day, 7+ hour ago  (275+ words) Open AI's reasoning model autonomously disproved a conjecture posed by legendary mathematician Paul Erd's in 1946, and top mathematicians say the proof is publishable on its own merits. For nearly eight decades, some of the sharpest minds in mathematics stared at…...

6+ day, 9+ hour ago  (290+ words) Competing AI labs set aside rivalry to warn lawmakers that large language models could help bad actors engineer biological threats. The biggest names in artificial intelligence just co-signed a letter asking the US government to do something they almost never…...

6+ day, 13+ hour ago  (81+ words) Exploring AI's potential to redefine human identity and challenge ethical boundaries in the 21st century. The post John Lennox: Mathematics reveals a word-based universe, AGI challenges human identity, and transhumanism raises ethical dilemmas | The Diary of a CEO appeared first on…...