Sign up to save tools and stay up to date with the latest in AI
bg
bg
1

Developing AI models or giant GPU clusters? Uncle Sam would like a word

Nov 05, 2023 - theregister.com
The US government, under an executive order signed by President Joe Biden, is requiring operators of large AI compute clusters and models to disclose their activities. The order aims to ensure the safe and productive development of AI technologies, with a focus on potential dual-use foundation models that can be used for both civilian and military applications. However, the thresholds set for reporting are so high that only the largest ML companies and infrastructure providers are likely to be affected.

The White House has set interim thresholds that trigger reporting obligations, including any model trained using more than 1026 integer or floating point operations total, or more than 1023 floating point operations for biological sequence data. Another threshold is set for compute clusters located in a single datacenter and networked at more than 100Gb/s. Facilities exceeding 1020 FLOPS of AI training capacity will be subject to reporting rules. More precise reporting requirements are expected to be defined in the next 90 days.

Key takeaways:

  • The White House has issued an executive order requiring operators of large AI compute clusters and models to report their activities, specifically those exceeding certain thresholds.
  • The reporting requirements are currently set at interim thresholds, including any model trained using more than 1026 integer or floating point operations total, or more than 1023 floating point operations for biological sequence data.
  • Another threshold is set for compute clusters located in a single datacenter and networked at more than 100Gb/s, with facilities exceeding 1020 FLOPS of AI training capacity subject to reporting rules.
  • Several major tech companies, including OpenAI, Google DeepMind, Amazon, Microsoft, Anthropic, Mistral, and Meta, have signed a non-binding agreement to test their powerful ML models for national security and other risks before releasing them to the wider world.
View Full Article

Comments (0)

Be the first to comment!