This doesn't feel correct to me. In the analogy of avoiding a local minimum you are increasing the variability slightly so you don't get trapped, but the reason you get trapped is that you're aiming "higher", i.e. the overarching goal is to move from semi random starting points towards 'better' in the terms of the current metric.
So this appears to be talking about a plan for doing one thing, while claiming to do the opposite.
Though the fact that it's personalized, may actually provide this, while also defeating the other purpose (i.e. it could lead you deeper into a flat earth rabbit hole by finding you the exact content best placed to meet your need for more challenging flat earth content at that point).
Not sure how it can achieve both goals at the same time.