April 13th 2025. Today was uneventful. I cooked again, although it's quite effort consuming to do so. Yesterday was the first I cooked in around 2 weeks. It's unfortunate there aren't good options to eat fast, cheap, and healthy.
Yesterday, I was ready to sleep around 12:30am, but I was up until about 4am yesterday, so I missed the chance to calibrate my sleep schedule. I've been getting better at fixing my sleep schedule though. Now I can normally fix it in just a day when I try.
I'm still trying to get the meta-rl environment to work. The current approach I'm using is to turn the task environment into a dictionary of functions, then passing that in as input to our meta-rl environment, so it should be able to work on anything that has the gym style api. Ideally, this should make training an inference process. We would be able to train a model by inputting the weights and some information about the episode to another model, which outputs updated weights. Obviously the problem is that this will require a gargantuan model and lots of compute as well as alchemy to make it actually learn to make effective updates with anything near the sample efficiency of the average child.
The dictionary thing somehow reminds me that theres also some dead project out there that converts all the keywords of python into chinese, but it's a bit out of date and the project is dead. I feel like massive amounts of money can be made with this type of stuff in China maybe similar to what oracle did with java. Although it is kind of useless in the grand scheme of things.