Kipply Chen and Allison Duettmann discuss AI alignment, focusing on data-related tasks at the pre-training level. Chen believes that alignment is in its early stages, but expects this to progress faster than AI capabilities. Challenges include aligning predictive and base models, which require concentrated efforts and coordination between labs. She also highlights that security, and policy work are crucial in AI safety. Furthermore, she discusses workshop participant Matjas’ proposal on using cryptography techniques for watermarking language models. Finally, she anticipates an increase in lawsuits related to training data misuse and calls for an open-source alignment community to address alignment challenges.