Use of Tokens Is Skyrocketing

Nathan Lambert offers thought-provoking insights about how fast we are scaling our use of tokens with genAI applications. He uses numbers offered by Google, OpenAI, and more. Good perspective. Maybe I am a nerd.

Raises some questions (not unlike the ones we considered with mobile voice, SMS, and data):

- Does seat-based pricing make sense for SaaS when the variable costs of usage are non-trivial?

- Will we reward genAI tools for being efficient? or choose not to absorb the costs? (i.e., pay for outcomes, not work).

- Why use inference if we don't have to do so?

For anyone who needs a reminder on mobile ... consumer use of mobile services "took off" when there were flat fees for service (ok, they would choke your stream if you really over-used). When carriers were stubborn and charged per SMS (a LOT), IP-based message apps took hold.

PS As we scale our use of inference, we also scale our use of energy - lots of implications here.