Your AI bill depends on which model you pick and how much of it wakes up per token.
Mixture of Experts is Tokenminning inside the model. Tokenminning decides whether AI gets called at all.
@sriramgopalan.bsky.social
@sriramgopalan.bsky.socialYour AI bill depends on which model you pick and how much of it wakes up per token.
Mixture of Experts is Tokenminning inside the model. Tokenminning decides whether AI gets called at all.
@sriramgopalan.bsky.socialYour AI bill has a hidden lever: quantization.
Model weights carry more precision than needed. Round them down (FP16→INT8→INT4) and one benchmark saw a 70B model's GPU cost drop from $24K to $4K. Same model, 83% cheaper.
sgopalanbtg.substack.com/p/explainer-...
#TokenMinning #AI #FinOps
@sriramgopalan.bsky.socialYour AI's "on second thought" isn't reconsideration. It's autocomplete predicting a correction pattern learned from humans thinking aloud. No draft, no delete key — you pay for the wrong turn AND fix.
sgopalanbtg.substack.com/p/the-actual...
#AI #FinOps #TokenMinning