Tag
This paper proposes a principle of 'constraint-enhanced physical search' where temporal correlations in exploration are matched to constraint-induced spatial correlations in update dynamics, demonstrated via a tug-of-war bandit model. The authors show that efficient search emerges not from maximal randomness but from matching temporal correlation to the physical update scale that converts feedback into evidence.
This Twitter thread highlights Stanford CS221 lecture 6 on heuristics, explaining how A* search improves agent efficiency by using heuristics to guide decision-making. Key takeaways include building heuristics by relaxing problems, the danger of bad heuristics, and the optimality of A* with the right estimate.
Discusses how brand mentions across social platforms and AI-generated answers are becoming as important as traditional backlinks for search visibility, suggesting a shift in search ranking signals.