Google DeepMind AI Writes 75% of Code, But Isn't Self-Improving Yet

Google DeepMind's AI Code Contribution

Google DeepMind’s initial internal assessment of its AI-assisted research processes shows a significant reliance on artificial intelligence for code generation. The audit found that AI systems now write approximately 75% of the code used within the organization’s research efforts. Despite this high level of automation, the study indicates no measurable increase in the speed of AI progress itself.

No Acceleration in AI Development

The findings challenge the notion of rapid, self-accelerating AI development. While AI is proving adept at automating tasks like code writing, it hasn’t yet reached a point where it can independently drive breakthroughs at a faster rate. This suggests that human oversight and innovation remain critical components of AI advancement.

Implications for AI Research

This assessment provides a grounded perspective on the current state of AI capabilities. It highlights that current AI tools are powerful assistants, but not autonomous engines of discovery. The research suggests that continued investment in human expertise alongside AI tools is essential for sustained progress in the field. The report did not detail the specific AI models used for code generation, nor did it specify the scope of the internal audit beyond its focus on code-related tasks.

Source: Decrypt