AI Safety and the Age of Dislightenment

The article argues that proposals for strict AI model licensing and surveillance will likely be ineffective or counterproductive, concentrating power in unsustainable ways and potentially undoing Enlightenment-era gains. It contends that regulating full access to foundation models creates a severe power imbalance, since those with full model access gain advantages over users limited to restricted services like ChatGPT. The authors favor openness, such as open-source model development, and regulating applications and disclosure instead. The article notes that the whitepaper "Frontier AI Regulation" (FAR) has authors connected to OpenAI and Google, and says that models can be copied easily and trained collectively by decentralized groups.