Completely agree with this. While there are some novel applications possible with reasoning models, the main value has been the ability to trace specific chains of thought and redefine/reprompt accordingly. Makes the system (slightly) less of a black box
I really enjoyed this, and especially appreciated the detailed assumptions and corresponding projections in the research tab.
Your treatment on energy requirements is interesting, but unless I’m missing something I haven’t seen anything explicit on this - is your assumption that energy requirements won’t be a bottleneck that slows down the US competitive advantage?
I read your chart (figure 23) to suggest that compute will require 3.5% of US energy and assuming no increase in US energy production ( 7x increase from Jan 2025 baseline).
Are you concerned at all about energy production capacity limiting the US ability to deploy these models at scale and fully ramp up the SEZs? I think that would either require a significant decline in broader power consumption for other purposes, a significant increase in efficiency, or broad deployment of new energy sources quickly (which has not been achievable quickly in the US).