I'm wondering if 'slow AI' like this is a temporary bridge, or a whole new category we need to get used to. Is the future really about having these specialized 'deep thinkers' alongside our fast, everyday models? Or is this just a clunky V1 until the main models get this powerful on their own in seconds?
It's not unreasonable to think that with improvements on the software side - a Saturn-like model based on diffusion could be this powerful within a decade - with 1s responses.
I'd highly doubt in 10 years, people are waiting 30m for answers of this quality - either due to the software side, the hardware side, and/or scaling.
It's possible in 10 years, the cost you pay is still comparable, but I doubt the time will be 30m.
It's also possible that there's still top-tier models like this that use absurd amounts of resources (by today's standards) and take 30m - but they'd likely be at a much higher quality than today's.
The pressure in the other direction is tool use. The more a model wants to call out to a series of tools, the more the delay will be, just because it the serial process isn't part of the model.
we’re optimizing for quality over performance right now, at some point the pendulum will swing the other way, but there might be problems that require deep thinking just like we have a need for supercomputers to run jobs for days today.