[suzzer99]

I feel like this is very possible eventually, but under 5 years seems unrealistic to me. The compute power is already running into limits. And a few more major problems need to be solved before agents can have that kind of autonomy to create software from scratch and run entire businesses.

[ilaksh]

Given the intense pressure to find more efficiency, 5 years is probably enough to see deployment of radical new compute-in-memory paradigms such as nitrides-based ferroelectric devices that have a few orders of magnitude more efficiency.

I think the biggest problem is jagged intelligence which I would bet would be solved largely with bigger models, which comes back to rapidly evolving hardware paradigms.