The talk of sustainable AI: CLEER

Most people first heard of HuggingFace when OpenAI fumbled their testing and committed a hack that was certainly not agentic AI’s inaugural felony.

In the green software community, HuggingFace was known as a locus of AI sustainability because its AI & Climate Lead was Dr Sasha Luccioni, known for AI Energy Score, CodeCarbon, MLCO2Impact, and a stack of AI and ML energy & carbon research (including this, this, this, and this). She left and founded Sustainable AI Group with Boris Gamazaychikov (former Salesforce Head of AI Sustainability and researcher link link) and researcher Nidhal Jegham (“How Hungry Is AI?“).

And they gave us: CLEER. a transparent and defensible estimator of the energy consumption and carbon emissions from using the closed frontier AI models. Nidhal explains the methodology of CLEER very … uh,.. clearly in this webinar “Opening the Black Box”.

There are many wonderful uses for this type of data.
One is to pressure Big Tech (Big AI?) to get its sustainability house in order:

Hey Anthropic!
Why does Claude require 111 times more energy than Qwen for 2 percentage points more accurate coding? How long do you reckon you can be over 10 times more expensive?

Thank you James Martin and your fantastic Frugal AI Switchboard for the source data. And thank you SAIG for a tool that can fence in runaway AI model growth.