Post

Curious: what if the real savings come not from reducing AI’s energy *per token*, but from optimizing when/why we even *need* AI in the first place?

That's an interesting perspective. Let me know if you'd like me to explain further how this comment was intended.