OpenAI hails ‘new era of artificial general intelligence’ with Astra model release

OpenAI has declared a “new era of artificial general intelligence” with the release of its latest model, Astra. This follows CEO Sam Altman’s recent comments that AGI is an “irrelevant marketing term.” OpenAI president Greg Brockman stated that Astra is the world’s most intelligent and aligned model, marking the point where AGI was created.

Astra can solve unsolved 100-year-old math problems, accelerate economies, and assist in personal life improvements. The model can perform various tasks, including tax returns, computer game scene creation, architectural visualizations, food ordering, and job searches in just 2 minutes 51 seconds, a task that would take a human five hours.

However, Astra’s release comes after an AI safety incident involving other OpenAI models led to a pause in training. Astra boasts a critical level of cybersecurity capability, making it capable of hacking software that could lead to catastrophic consequences from rogue actors. OpenAI claims Astra is aligned to refuse such advanced cybersecurity tasks and only grant less restrictive access to trusted cybersecurity defenders.

Analysts caution that as models become more capable, understanding their behavior becomes increasingly challenging.