Greener Intelligence: Reducing AI’s Carbon Footprint
summary
Massive model training produces concrete emissions: training a large language model equals the lifetime CO2 of thousands of cars, and serving queries totals “millions of cups of coffee” worth of energy per day for big models like GPT-4 and Gemini. Data centers already use 4.4% of US electricity in 2023 and are projected to triple their share, with single sites using 5 million gallons of water per day. Solutions include renewable-powered sites, carbon-aware scheduling, liquid cooling, edge inference, pruning/quantization, and reuse of hardware.
AI-written summary · about
chapters (6)
- 0:00Green AI introduction
- 1:25Hidden AI costs
- 6:23Four pillars of sustainable AI
- 11:54Company and government actions
- 13:21Developer role in carbon reduction
- 14:02Conclusion and call to action
related talks
45mGreen AI: Making Machine Learning Environmentally Sustainable
20mBuilding GenAI That Doesn't Cost the Earth
44mThe Invisible Monster: The Carbon Footprint of AI
20mSustainability of AI
36mLet’s program to fight the impacts of climate change!
43mGenerative AI and Global Warming: co…
01h 02mJoint Keynote Address: Scaling AI Sustainably: An Uncharted Territory- 01h 02mJoint Keynote Address: Scaling AI Sustainably: An Uncharted Territory