OpenAI: Astra Could Have Advanced Some Projects by Six Months

A developer at OpenAI credits the internal tool Astra with accelerating projects by six months. Meanwhile, a recent study conducted by Severin Field reports that the majority of researchers surveyed fear the automation of research and anticipate that the most powerful models will remain internal. Anthropic, for its part, claims that Claude writes the majority of its production code.

Researchers Worry About Automation and Model Retention

IAPS researcher Severin Field conducted a study among 25 researchers from OpenAI, Anthropic, Google DeepMind, and Meta. Twenty of them consider the automation of AI research to be one of the main risks associated with AI. Several milestones mentioned by these researchers have already been reached. Furthermore, half of the respondents believe that the most powerful models will remain internal and will never be made available to the public.

Anthropic: Claude Generates the Majority of Its Code

Anthropic states that Claude now writes over 80% of its own production code. This statement aligns with the trend of using models to automate code creation.

Thibault Sottiaux Claims Astra Provided a Major Advantage to OpenAI

Thibault Sottiaux, a developer at OpenAI, believes that Astra likely represented the company's main competitive advantage while it was not publicly accessible. According to him, the internal use of Astra has increased productivity to the point of advancing certain projects by six months. However, not everyone shares this enthusiasm, particularly regarding self-improving AI models.