Exploring ideas in software engineering, AI systems, and the craft of building great products.
The OpenAI API could do all of that beautifully. But at scale, even a modest number of daily users would rack up a bill I couldn't justify as a third-year CS student building a side project between assignments. So I asked myself: what if I just ran the model myself?
There's a conversation most of us have never had. Not because we don't want to have it, but because we don't know how to start it. Maybe it's something you did that you're not proud of. Maybe it's a feeling you've been carrying alone for months. Maybe it's just a thought that would sound strange coming from your mouth, attached to your name, in front of people who know you.
If you've ever asked a virtual assistant a question, gotten a suspiciously accurate product recommendation, or watched a spam filter quietly protect your inbox — you've already benefited from machine learning. But what exactly is it, and why is everyone talking about it?
For years, the AI industry chased one goal: bigger models. Then a 70-billion-parameter underdog named Chinchilla beat models four times its size — not by being larger, but by being better trained. The 2022 DeepMind paper that introduced it rewrote the rules of scaling, revealing that the field had been systematically leaving performance on the table. Here's what the Chinchilla Law says, why it shook the research world, and what it means for the future of AI development.