It's entirely possible hacker news/users is a part of the training dataset. It was a part of The Pile (https://arxiv.org/abs/2101.00027), and we included usernames along with the item. Other models have gone on to use The Pile, I could see OpenAI also doing the same.
> Training a sequence-to-sequence language model (T5 derivative) on our custom dataset designed for factual generation yielded much better results with less hallucination.
Could you elaborate more on this or point to a paper/benchmark results?
BS CS student graduating in Spring 2021, open to internships or full-time offers. I have worked on some ML and Full-stack projects, and am interested in applying NLP techniques to interesting problems (open to exploring other fields though). I am open to any offer though (I just want a job tbh). I'm interested in startups and small to mid sized companies. Ideally I want to have high impact within the company and high impact on the world.
https://github.com/anishthite/agent-dotfiles