🔮 AI & energy; global slowdown; Tornado Cash and crab evolution++ #386

The near future

🧨 Apophis intelligence
At current growth rates, machine learning workloads might consume the entirety of global energy production, argues AMD’s CTO. The growth in demand is not merely in the cloud but in increasing uses by edge devices like cars, phones, and sensors. So I don’t think this will actually happen, but it’s a useful challenge to prompt action. Current approaches to increase the power of ML models involve more complexity. But such complexity comes at a disproportionate energy cost: a four-fold increase in parameters drove an 18,000-fold increase in energy usage. AI’s carbon footprint varies dramatically region-to-region and between different types of model architectures.

While valuable triggers for analysis, to some extent both of these challenges (growing energy use and the carbon footprint) only capture part of the picture. If ML workloads improve outcomes, then perhaps the energy investment is worthwhile. Take one simple use case: sensors from field to col…