michael-dean-k/

I just set up something with AI that really unlocked my 3 year logging practice. The insight is pretty simple. For context to be useful, you want atomize it. A 1,000,000 word dump into the context window isn’t that useful. I mean, each model will get better at the “needle in a haystack” problem, but if you atomize your context, you have far more precision. I wrote a python script that breaks each monthly log file into individual log.md files ([DATE]-[TIME]-[HASH].md). This means every thought turns into its own semantic coordinate. When I ask a prompt, it’s able to rapidly pull in the most 100 most relevant relevant logs, and factor that into its answer. I wonder if there’s value in auto-splitting essays into paragraph files for a similar degree of resolution.