A look at the wave of Google DeepMind researchers who have exited recently to launch their own AI startups focused on alternatives to LLMs

If you build AI systems, alternative approaches to large language models are now in development.

Several researchers from Google DeepMind have recently departed to establish their own AI startups. Robert O'Callahan, formerly a technical expert at DeepMind in New Zealand, is one such individual. He cites the rapid pace of AI development and expresses concerns about the risks associated with superintelligent AI, including cognitive surrender, AI-induced psychosis, loneliness, power concentration, economic disruption, cybersecurity threats, and a lack of accountability. O'Callahan believes the potential harms of AI may outweigh its benefits. He links to external resources discussing the existential risks of advanced AI, referencing the effective altruism movement, which has faced criticism and been characterized by some as a fringe group. These departures suggest a growing divergence in views regarding the development and safety of advanced AI within leading research institutions.

The exodus of DeepMind researchers signals a growing unease with the current trajectory of large language model development and a desire for more controlled or fundamentally different AI architectures. These departing scientists are not just leaving a job; they are actively seeking to build alternatives that mitigate perceived risks, potentially leading to a more diverse AI landscape beyond the dominant LLM paradigm. This could foster innovation in areas previously overshadowed by the LLM race, offering new tools and approaches for AI development.

The concerns raised by these researchers, particularly regarding existential risks and ethical implications, highlight a critical debate within the AI community about responsible development. Their focus on 'unambiguously pro-human' work and the deliberate pursuit of alternatives to current LLM trends suggest a potential shift in research priorities and investment. This movement could influence future AI safety standards and encourage a broader industry conversation about the long-term societal impact of artificial intelligence.