Jeffrey Quesnelle, co-founder of Nous Research, argues that massive capital concentration in large corporations is stifling innovation within the AI industry. He highlights how decentralization technologies offer solutions to both the funding and operational challenges inherent in AI development.
Quesnelle points to inefficiencies in current AI data centers, noting that around 50% of GPUs remain underutilized, leading to significant costs. He explains that crypto rails can provide permissionless access to computing resources, thereby enhancing decentralization. This distributed model, he suggests, can democratize access to AI development and training.
Crucially, Quesnelle emphasizes the role of smart contracts in decentralized AI training. These contracts are essential for assigning tasks and ensuring consensus and accountability among participants, maintaining system integrity in a permissionless environment. He also warns of regulatory capture, citing instances where legislation could threaten open-source AI by imposing criminal liability on developers.
For AI to remain competitive, Quesnelle states that efficiency is paramount, with a strategic goal of achieving thousandfold improvements in intelligence per unit of energy. He believes there is significant untapped potential for breakthroughs in AI efficiency, drawing parallels with natural systems, which could redefine the future of the technology.