Most AI Compute Isn't Used for Cutting-Edge Research
A new report reveals that only a small fraction of AI computing power is used by labs working on the most advanced models. Most AI compute is still used for simpler, less innovative tasks. This highlights a significant gap in how AI resources are allocated.
Epoch AI published a report showing that frontier labs—those working on the most advanced AI models—use only a small portion of the world's AI computing power. Most AI compute is still dedicated to less ambitious tasks like training smaller models and running existing applications.
This gap matters because it means we're not fully leveraging our resources to push the boundaries of AI. Think of it like having a supercomputer but mostly using it to check emails. If more of that power went to cutting-edge research, we might see faster breakthroughs in AI capabilities.
If you're curious about how AI compute is used, check out the full report on Epoch AI's website. Look for the section on resource allocation to see where the biggest opportunities for improvement lie.