Google DeepMind’s AI Progress Shows No Acceleration
- Google DeepMind’s capability index has increased steadily by about 0.1 points per month since April 2026.
- More than 75% of production code at the lab is AI-written, yet ~88% of code changes still require human intervention.
- Gemini 4 Argon completed only half of its internal tasks independently, well below the threshold for full automation.
- The report indicates no significant acceleration in machine learning research and development driven by AI.
Despite concerns about recursive self-improvement, Google DeepMind’s latest findings suggest that AI is not accelerating its own development significantly faster than before. The steady growth in capability index and reliance on human oversight indicate that current risks from AI-driven advancements remain low.
This data highlights the ongoing need for human involvement in AI development processes, as evidenced by Gemini 4 Argon’s performance and the high percentage of human-steered code changes. (Source)
