Abstract
Neutrino oscillations prove that neutrinos have mass and effectively puts a lower bound to the neutrino mass scale, but the absolute mass scale still remains unknown. The most sensitive direct, model-independent measurement of the effective electron neutrino mass $m_\beta$ comes from observing the tritium beta-decay endpoint energy spectrum. Using this method, the KATRIN experiment currently sets the most stringent direct upper mass limit at 450 meV, and is designed to reach ~200 meV sensitivity. Project 8 experiment is aiming to go further & probe the entire inverted-ordering region using the novel technique of cyclotron radiation emission spectroscopy (CRES) - which infers each tritium beta-electron’s kinetic energy from its cyclotron frequency. Following the successful demonstration of CRES with waveguides, the upcoming phase of Project 8 will demonstrate the first realization of the CRES technique in cylindrical cavities using Cavity CRES Apparatus (CCA) with the goal to further improve energy resolution by an order of magnitude. A cubic-meter scale volume apparatus called Low-Frequency Apparatus (LFA) is planned to boost statistics and address remaining technical risks on the path to reaching the final mass sensitivity goal of 40 meV. Large Language Models (LLMs) like GPT-4 have demonstrated remarkable capabilities through pre-training on vast text corpora. However, their raw outputs often fall short of the nuanced expectations of human users. In this talk, I will explore the critical role of post-training techniques in aligning LLM behavior with human preferences. We begin with an overview of LLM pre-training through next-token prediction, then investigate why this objective alone is insufficient. Drawing on insights from reinforcement learning, we delve into post-training strategies including supervised fine-tuning and preference optimization methods like RLHF and Direct Preference Optimization (DPO). I will also present our recent findings—highlighting the surprising power of on-policy, suboptimal data in preference learning—and discuss the implications for future LLM alignment research. This talk is intended for audiences interested in LLMs, alignment and post-training.