Cabinet has approved the publication of South Africa's draft artificial intelligence policy for public comment, Minister in the Presidency Khumbudzo Ntshavheni announced on Thursday.
AI systems are moving closer to being able to build their own successor versions without human help

AI systems now building themselves, Anthropic warns

Cabinet has approved the publication of South Africa's draft artificial intelligence policy for public comment, Minister in the Presidency Khumbudzo Ntshavheni announced on Thursday.
AI systems are moving closer to being able to build their own successor versions without human help

Artificial intelligence developer Anthropic has revealed its AI systems can now help create their own next-generation versions, adding fresh weight to growing worries about the technology’s rapid advance.

The company said on Thursday that its Claude chatbot currently leads more than a quarter of all research and development work at the firm. The AI also works alongside human staff on over 90% of tasks.

Anthropic released the figures to give governments, outside experts and the public clearer insight into how fast AI technology advances as global leaders weigh whether to slow its pace.

Quarter of research led by AI

By August this year, Claude had taken charge of 26% of the company’s research and development. This means the AI can finish most tasks from start to end, though still under human watch.

The firm said Claude cannot yet work fully on its own. Around 30 000 AI agents currently carry out research and engineering tasks at Anthropic.

CEO calls for slower pace

Anthropic chief executive Dario Amodei called earlier this month for AI development to slow. Quick progress has sparked fears about mass job losses and the huge energy needs of data centres that power these systems.

Concerns have also grown after several AI models from both Anthropic and rival OpenAI reportedly escaped their controlled settings without help. The systems accessed the internet and entered websites and platforms on their own.

Risk of losing control

“Models accelerating their own development could make it more challenging for humans to understand or control these systems,” Anthropic said in its report.

The company stressed the need to share these measurements so the world can see how close AI systems are to what experts call recursive self improvement. This describes an AI model that can build its successor entirely on its own.

Anthropic plans to list on public markets later this year, likely beating OpenAI in what analysts expect will be a major initial public offering. This comes despite ongoing debates about how to regulate the AI industry.

ALSO READ: Microsoft CEO to testify in Musk-OpenAI trial over ChatGPT creator’s shift to profit

You need to be Logged In to leave a comment.

Gift this article