- A study by the independent AI safety monitoring group Model Evaluation and Threat Research found that Anthropic’s AI progress is accelerating twice as fast as it was a year ago.
- The rapid advancement is driven primarily by recursive self-improvement, where AI systems help construct the next generation of models, rather than an increase in training computing power.
- Anthropic chief executive Dario Amodei warned that unchecked recursive self-improvement could outrun the ability to control these systems, a concern echoed by OpenAI chief executive Sam Altman.
- Researchers at METR warned that further acceleration in AI development could lead to the sudden emergence of extreme and risky capabilities.
- Although human oversight remains essential for critical aspects of model training, researchers noted signs of life suggesting AI systems may soon make autonomous contributions to development.
IN FULL
