spending $1000 on inference over the weekend to vibecode your toy app is like spending $1000 on gas fees to mint an NFT in 2021. a few years from now this will seem insane and we’ll get whatever the AI equivalent is of an L2 blockchain…harness rigorously optimized for costs
The article compares expensive AI inference to high NFT gas fees in 2021, predicting future cost optimizations akin to L2 blockchains.
spending $1000 on inference over the weekend to vibecode your toy app is like spending $1000 on gas fees to mint an NFT in 2021. a few years from now this will seem insane and we’ll get whatever the AI equivalent is of an L2 blockchain…harness rigorously optimized for costs
Narrative violation People say they want local models until they realize how much they’ll spend on hardware and electricity to run worse models at lower tokens per second If the goal is cheaper inference, the answer usually isn’t local. It’s custom. Small models trained on your exact workload, lik
Narrative violation People say they want local models until they realize how much they’ll spend on hardware and electricity to run worse models at lower tokens per second If the goal is cheaper inference, the answer usually isn’t local. It’s custom. Small models trained on your exact workload, lik
Narrative violation People say they want local models until they realize how much they’ll spend on hardware and electricity to run worse models at lower tokens per second If the goal is cheaper inference, the answer usually isn’t local. It’s custom. Small models trained on your exact workload, lik
spending $1000 on inference over the weekend to vibecode your toy app is like spending $1000 on gas fees to mint an NFT in 2021. a few years from now this will seem insane and we’ll get whatever the AI equivalent is of an L2 blockchain…harness rigorously optimized for costs