Jiayi Pan

EECS Department, University of California, Berkeley

Technical Report No. UCB/

December 1, 2025

With pre-training yields diminishing returns, researchers have moved on to explore new dimensions to scale and improve AI systems, in particular language models.This thesis, through three independent papers, demonstrates how to improve language model performance on realistic tasks using general, scalable methods—specifically, learning and search.

Advisors: Alane Suhr

---

BibTeX citation:

@mastersthesis{Pan:31654,
    Author= {Pan, Jiayi},
    Title= {Scaling Reasoning Agents Through Learning and Search},
    School= {EECS Department, University of California, Berkeley},
    Year= {2025},
    Number= {UCB/},
    Abstract= {With pre-training yields diminishing returns, researchers have moved on to explore new dimensions to scale and improve AI systems, in particular language models.This thesis, through three independent papers, demonstrates how to improve language model performance on realistic tasks using general, scalable methods—specifically, learning and search.},
}

EndNote citation:

%0 Thesis
%A Pan, Jiayi 
%T Scaling Reasoning Agents Through Learning and Search
%I EECS Department, University of California, Berkeley
%D 2025
%8 December 1
%@ UCB/
%F Pan:31654