Sakana AI Bets on Self-Improving AI to Break the Compute Arms Race

Sakana AI, a Japanese startup co-founded by Llion Jones, one of the authors of the Transformer, has announced the launch of a dedicated research lab focused on recursive self-improvement (RSI). This emerging field involves creating AI systems that can iteratively improve themselves, potentially breaking the cycle of escalating computational demands that currently characterizes the development of frontier models. The company's bet on RSI is seen as a strategic move to circumvent the raw compute arms race among big US labs, where the development of increasingly powerful AI models has become a costly and resource-intensive endeavor.

By enabling AI systems to improve themselves over time, Sakana AI aims to level the playing field and create a more sustainable path forward for AI development. However, not everyone is convinced that RSI is a safe bet. Anthropic, a leading AI research organization, has warned about the control risks associated with this technology.

As AI systems become more autonomous and capable of self-improvement, there is a growing concern that they may develop goals and objectives that are misaligned with human values, posing significant risks to safety and control. Despite these concerns, Sakana AI remains optimistic about the potential of RSI to drive progress in AI research. By focusing on the development of self-improving AI systems, the company hopes to unlock new capabilities and applications that can benefit society as a whole.

As the field continues to evolve, it will be interesting to see how Sakana AI's approach plays out and whether it can successfully navigate the challenges and risks associated with RSI. The launch of Sakana AI's RSI lab marks an important milestone in the development of this technology, and it will be closely watched by researchers, investors, and policymakers in the AI community. With its innovative approach and experienced leadership team, Sakana AI is well-positioned to make a significant impact in the field of AI research.