As Luccioni shows me a digital map that compiles real-time data on electricity consumption and carbon intensity, she explains that it’s her “favorite website.”
His enthusiasm is reflected in his role as AI and climate manager at Cuddly facewhere Luccioni collects actionable data on the environmental impact of AI.
She has her work cut out for her, especially since companies like Nvidia and Google, which designs AI chips, is becoming more secretive about its proprietary offerings.
“The difficulty is that there is this race for secrecy,” Luccioni said. “Since ChatGPT arrived in November 2022, companies have really cracked down on the amount of information they share on their models.”
Luccioni co-developed and regularly contributes to CodeCarbonea program that helps developers estimate emissions and energy consumption from running AI models. Luccioni says tens of thousands of people have cited the program. She is most proud of how this allows technologists to address their environmental concerns and benchmark against real-world data.
Meta recently used the tool to estimate emissions linked to the operation of one of its latest llama models.
“Where the energy comes from really has the biggest impact on emissions,” Luccioni said. “The problem is that most supercomputers, whether it’s Google Cloud, Azure or AWS, are located in places that are not powered by renewable energy, mainly natural gas and coal, and that makes a huge difference.”
One of his latest projects is expected to make it less opaque to understand how much energy and computing power popular AI models and tools use. This work, carried out with the Organization for Economic Co-operation and Development, should ultimately help establish a standard for evaluating energy efficiency for AI models. She compares it to a Energy Star Rating.
“EPA has taken this trust-based approach, but in the case of AI, it would be a little more difficult to take that approach because there are so many variables,” she said. “You need to define specific data sets and specific hardware to compare these models.”
As The value of AI is soaring, perhaps into the billions, Luccioni’s work will help document the consequences of this technology on power grids and, ultimately, the environment.
“You train a model once, but you deploy it for a while. Even quantifying the amount of energy per query or per day is really powerful,” Luccioni said. “In the long run, it really pays off.”
See the full Business Insider AI Power List