@sd141414: Recursive self-improvement needs a result that can be scored automatically. Search advertising has that. Most business processes do not. An invoice reminder can be scored in days. A big enterprise deal produces one data point a year. Check for the answer key before you build the loop. The Systems Thinking Quiz on prodblueprint.dev is a quick read on how you think about systems like this. Most teams choose the process before they check whether it can be scored.
I just find it hard to believe a model will be able to adjust its weights in a way that actually improves the intelligence. Sure you can make better tools and harnesses to send these models to perform tasks, but intelligence comes from training, and that’s a much different and harder problem
2026-09-21 23:22:00
0
To see more videos from user @sd141414, please go to the Tikwm
homepage.