.plan - January 2025
What I did in January
I’m working on building better personal feedback loops. I have two primary goals in mind, and I think these info dumps will serve both well. The first is to keep a stream of conciousness record of content that I’ve reviewed or stirred on. Since these updates are intended to be quick lip service anyway, I think this will mitigate the sense that I have to put polish on everything that I write. The second is to remind myself of the progress that I’ve made over the course of weeks/months/years. It’s obviously difficult to see the forest from the trees, but I hope that I can provide that vantage point to my future self.
January 1 - ~January 25
I had a bunch of PTO days over the holidays, so I did some scouting for a blogging tech stack that would suit my needs. Putting my writing out into the world has been on my mind for a while and it felt like the right time to take the plunge. I still don’t really have a good answer to questions like “Why not Substack?”, but I’m happy with what I have for now.
This is how I felt every time I chose to tweak styling instead of write
I did however encounter a lot of bikeshedding. I think I was avoiding the inevitable discomfort of actually writing something that would see the light of day. I have a whole post hand-wringing about why this blog exists. In any case, it’s deployed to the world now for better or for worse.
Twitter started melting down about the DeepSeek R1 paper, so I’ve been reviewing it. I need to publish a separate, quick-and-dirty post about it for posterity’s sake. GRPO is cool, but it’s bizarre to me that it took the internet 5(!!!) days to react.1 2 3
January 26 - Feburary 2
More or less finished collecting my thoughts on R1, which is surprisingly simple to the point that I think I’m missing some subtleties. Should turn it into a post, but given how much attention it’s already received, I think it may just end up being collected resources.
Started reading Stanley Chan’s Tutorial on Diffusion Models. I’ve never really played with diffusion before and I need to implement some of the ideas myself, but it’s a decent crash course for now. I’ve found myself getting stuck on a loop rereading the first section two or three times now for whatever reason as well.