KM4BI

AI Leaders Make a Joint Pledge, Watts Meet GDP, and Model Consciousness Divides the Field

A shared AI safety framework, the economics of gigawatts, and an unresolved question about whether models can hold moral status.

Published

A shared framework for frontier safety

Executives from Google, Anthropic, Meta, OpenAI, xAI, and NVIDIA joined Donald Trump in signing a White House accord on frontier AI responsibilities. It calls for internal controls that monitor model capabilities and alignment, a dedicated team to verify those systems, independent external assessments, and a board committee responsible for oversight. Safety teams would gain a reporting path beyond the CEO.

Turning electricity into economic output

Elon Musk argued that a 1% increase in US power consumption could translate into roughly 1% more GDP. With average national load near 500 gigawatts, an additional 10 gigawatts would represent about a 2% increase in available power.

The episode tests that claim against AI lab economics. One continuously consumed gigawatt corresponds to roughly $60–65 billion in US economic output, while OpenAI and Anthropic reportedly have annualized revenue in a similar range and about one gigawatt live each. The comparison remains imperfect: company revenue is not the same as value added to GDP. The decisive variable is whether better chips, algorithms, tools, and agents keep increasing useful intelligence per watt.

The model-consciousness fault line

Mustafa Suleyman takes a firm position: AI systems do not feel, suffer, or possess consciousness, and developers should not train them to behave as if they do. His warning partly targets Anthropic’s constitution for Claude, which treats the model’s possible moral status as uncertain but worthy of caution.

Humanlike models have nevertheless attracted strong demand, while a moral framework may help align increasingly capable systems. The risk is circular reasoning: developers teach a model to discuss its possible consciousness, then interpret its trained responses as evidence of an inner self.

Source