I completed this blog post as my final project for BlueDot Impact's AI Alignment course. While it's nowhere near as polished as I'd like it to be given the short-term nature of the course, I'd love to hear what others think of this idea, and better yet, find someone who can test its hypotheses experimentally based on the research agenda it proposes.
How can we steer large language models (LLMs) to be more aligned from the very beginning? This post explores a novel data curation method that leverages existing LLMs to assess, filter, and revise pre-training datasets based on user-defined... (read 6826 more words →)
You probably already know this (low hanging fruit), but in case your or another reader doesn't, eliminating processed meat and limiting red meat consumption seem very likely to reduce your risk of some types of cancer, particularly of the colon. There are studies that argue otherwise, and the purported relative risks also vary, but my general takeaway has been that processed meat = definitely bad, and red meat = possibly bad if eaten daily or in consistently large portions.
Recommendations are a little varied depending on where you look, but what I've opted for is having processed meat at most once a month and red meats at most 3x/week (this was a big... (read more)