IRS Acquiring Nvidia Supercomputer, ChatGPT Energy Use Overestimated, and Complex Brains Evolved Independently

The IRS is acquiring an NVIDIA Superpod AI supercomputer to enhance its machine learning capabilities for fraud detection and taxpayer behavior analysis. This system, consisting of 31 servers with advanced Blackwell processors, will be installed at the IRS Computing Center in Martinsburg, West Virginia. This acquisition is part of a broader AI transformation at the IRS, which has 68 AI-related projects in development. The Superpod is expected to significantly improve the IRS's ability to identify complex patterns indicative of fraudulent activities.

A study by EPIC AI has revealed that the energy consumption of ChatGPT's latest model, GPT-40, is approximately 0.3 watt-hours per query, much lower than previous estimates. The reduction in energy use is attributed to advancements in AI technology and improved calculation methods. Notably, typical household activities consume significantly more energy than ChatGPT's queries. The AI industry is increasingly investing in energy-efficient hardware and renewable energy sources to reduce environmental impacts.

Recent research led by Dr. Fernando Garcia Moreno has shown that birds and mammals have independently evolved complex brains through distinct pathways, illustrating convergent evolution. The study highlights that while the pallium, responsible for cognitive functions, develops differently across species, both groups achieve similar functional outcomes. The research indicates that inhibitory neurons have remained similar across species for millions of years, while excitatory neurons have evolved along dramatically different pathways, challenging traditional views of brain evolution and demonstrating that advanced cognition can arise from different neural architectures.