Keep pulling the thread on Geoffrey Hinton.
Advances in AI capabilities are primarily driven by new training techniques, such as step-by-step reasoning, and inference-time enhancements, rather than by simply training larger models, according to the International AI Safety Report 2025.
Improvements in AI capabilities increase the risks associated with biological weapons and cyber attacks, according to the International AI Safety Report 2025.
General-purpose AI systems can solve more complex problems in scientific research and software development due to recent capability improvements, according to the International AI Safety Report 2025.
The performance of general-purpose AI systems has improved on benchmarks for coding, mathematics, and expert-level science questions, according to the International AI Safety Report 2025.
General-purpose AI systems have persistent reliability challenges, demonstrating high performance on some tasks while failing completely on others, as noted in the International AI Safety Report 2025.
Advances in AI capabilities create new challenges for monitoring and controlling AI systems, as stated in the International AI Safety Report 2025.