"Query" is ambiguous, because ChatGPT runs on "tokens", not "queries".
In real life you will want an order of magnitude more tokens if you're generating something useful and not just testing/playing.
Compute doesn't get cheaper over time, it gets more expensive.
This isn't 1993 anymore.
> but high enough that I don't ever see regular humans pay for it for day-to-day tasks.
And “640K ought to be enough for anyone”.
Of course this will get cheaper over time as compute gets cheaper and the cloud providers design custom chips optimized for it.