Mochi

1. wake up, grab phone and some researcher posts insane scaling law analysis with charts showing how compute needs grow with model size?! bookmark into "ai math stuff" (that folder hasn't been opened since mid 2025 lmao) and quickly picked up the "scaling laws for neural language models" to read it again and get my head right 2. then I start on improvising the autocompletion models, yeah, doing this to beat the github copilot on the custom repos, but turns out the custom tokens have their own chaos to use them in the templates (more later) 3. then pleasing the claude code to fix my sft dataset script for scraping github prs?! hitting those api requests per hour limits and getting throttled every time i try to scrape at scale?! like dude, just let me build my nemotron-style sft training data in peace without waiting for rate limit windows to reset? 4. switched back to X to find a thread about underrated hill stations / beaches near bangalore with actual good weather?! bookmark to "places to visit" (who am i kidding, i'll just stay at some hostel again) 5.stumble on that obscure pytorch optimization technique using custom CUDA kernels that might make my transformer training 20% faster?! bookmark to "ml hacks" (the folder where good intentions go to die and dude, when I will get time to open these) 6. now it's 11 pm (night), long day, tired! but wanted to read something, well uhhh, "How To Build A Car from Adrian Newey" is nice, i mean guy won more championships combined as architect than teams alone, 7. ok magnesium, omega both done, what's next! well close browser for time being, open spotify, code until 3am on personal project with my love (claude opus) done now! good night