updates on learnings - the feed back from hn about ground truths was helpful passed this thread into the ai and it started to base truth on exa search. - it produced alot of slop so it added a review cycle to remove alot of the slop. - it worked out it was doing performative work then removed the data as code and moved this to toml. it then started to focus heavy on processes and doing things like the ALDH2 metabolic process.

honest take - decent at getting a quick understanding of what the ALDH2 process is like but struggles with harder modelling, the experiment gave better insights into ai then medical. (slop is countable with exa mcp (ground truthing, research, anti hallucinations).