The initial idealistic vision held by figures like Demis Hassabis of a single, collaborative lab developing AGI safely has been replaced by the reality of a fierce, multi-polar competition. This race dynamic was solidified by events like Elon Musk founding OpenAI after a DeepMind safety summit and the recent emergence of powerful Chinese models.
The episode contrasts the strategies of different AI labs. While Google DeepMind has vast resources allowing it to pursue multiple research avenues, smaller players like OpenAI (with its early bet on transformers) and Anthropic (with its focused push on coding agents) have achieved breakthrough success by making concentrated, high-risk bets.
Despite his brilliance and Nobel Prize-winning work, Demis Hassabis is portrayed as having significant blind spots. His deep roots in neuroscience and reinforcement learning contributed to his underestimation of the transformer architecture's power, a delay he is still reluctant to fully acknowledge.
Demis Hassabis's view on AI safety has evolved from believing a single lab could manage the risk to concluding that it's a collective action problem requiring government intervention. The lack of trust between lab leaders makes a government enforcer necessary to create a level playing field for safety protocols.
The competition between labs extends beyond technology to the control of public narrative and the recruitment of top talent. Sam Altman's much larger social media presence gives him an outsized ability to shape the conversation and attract researchers, even when DeepMind is releasing competitive products.
Keep pulling the thread on Sebastian Malaby.