AI: Misjudged Uses and Giant Data Centers

⚡

Key Takeaways

1AI Observatory identifies distinct uses based on models (programming with Anthropic, social/role-playing with Gemini, homework with ChatGPT)

2Nvidia commits up to $105 billion for an OpenAI data center in Ohio, expected to be operational in 2028

3Tesla plans Cybercab robotaxi rides in Austin this month

💡Why it matters — These elements combine differentiated uses of AI, a major infrastructure investment, and tests of autonomous services on public roads, illustrating the scope and diversity of the current deployment.

A new project, AI Observatory, documents AI usage practices that corporate reports do not capture and highlights differing preferences based on models. At the same time, an investment of up to $105 billion has been announced for an OpenAI data center in Ohio, electricity demand is rising partly due to AI, and several product pilots—from robotaxis to camera-equipped headphones—are emerging. Design choices for surveillance systems and lawsuits targeting platforms complete this picture.

Power Grid Strain and Mega-Project in Ohio

The electric sector is facing a "trilemma" between reliability, affordability, and sustainability, as electricity demand surges, partly driven by AI. During a snowstorm in Nebraska, nearly 10% of the 150,000 customers of Lincoln Electric System lost power, with CEO Emeka Anyanwu monitoring outage maps during the response efforts. This type of storm is only a secondary concern for him, given a much larger issue that is expected to soon affect all electric companies.

In this context, Nvidia has announced an investment of up to $105 billion in an OpenAI data center located in Ohio. The total project cost could reach $500 billion, with operations expected to commence by 2028. OpenAI is set to lease an eight-gigawatt site for 20 years.

Measuring Actual Usage: An Independent Observatory Steps In

Anthropic and OpenAI regularly publish usage reports, but AI researchers believe they only release data they want to make public, with no independent source to corroborate it. A research program called AI Observatory aims to address this lack of external measurement and studies sensitive behaviors that do not appear in these reports, which are generally focused on professional use.

According to AI Observatory, preferences vary by model: Anthropic is more sought after for programming, Gemini for social interactions or role-playing, and ChatGPT for homework assistance. Researchers observe notable discrepancies between different models.

Data Collection and Access: From Flock Network to Chinese Datasets

Flock Safety, which operates around 120,000 automatic license plate readers in the United States, has announced changes aimed at preventing illegal or illegitimate uses, including harassment. Beyond the arguments made in favor of these devices—solving serious crimes or the alleged harmlessness of the rest—the crux of the matter lies in design choices: what information is collected, who can search it, how long it is retained, and to what extent it is shared. These decisions shape the trade-off between security and civil liberties. James O’Donnell suggests examining a more restricted Flock system.

Meanwhile, Beijing is distributing datasets described as reflecting "dominant Chinese values." Chinese authorities want their data to influence what global AI knows, highlighting the role of datasets and their governance in defining systems.

Pilots and Hardware: Robotaxi in Austin, Fast Humanoid, Camera Headphones

Tesla plans to offer rides in Cybercab robotaxis in Austin this month. The first phase of the launch may involve employee-only rides on public roads, before electric vehicles join the robotaxi service a few days later.

On the robotics front, Unitree has unveiled a humanoid claimed to be faster than any human: the "Superman" robot can cover 12.66 meters in one second, with the company's IPO scheduled for Wednesday. At the same time, American robotics startups are facing challenges due to new restrictions imposed by China. Finally, a resurfaced video suggests that Apple is adding cameras to AirPods, showing headphones identifying a book through visual capabilities.

Litigation and Usage Rules: A Massive Lawsuit Targets Meta

A lawsuit concerning children's privacy protection begins today, backed by more than half of U.S. states, and could have repercussions for Meta. The plaintiffs argue that the company intentionally designed social networks that create strong addiction. Among the requested changes are the elimination of "like" accounts and infinite scrolling. Four states are seeking $1.4 trillion in damages.