Most LLMs are first trained on a broad dataset that informs the model’s overall knowledge and capabilities, and then are customized for a particular domain. That wouldn’t work for Rufus, since our aim was to train it on shopping data from the very beginning

Citation needed for “wouldn’t work” lol. That’s what foundational models were supposed to be for — foundations. I don’t think I want any answers from a service like this that aren’t directly citing some RAG results… training an AI from scratch on product reviews of all things seems like a good way to get it to speak like an alien. I guess they also included a lot of other more traditional datasets?