BCI deep dive
Reading through the state of the art in brain-computer interface software, one piece at a time.
The hardware for reading the brain has gotten good enough to leave the lab. People with implants are now using them at home every day to move cursors, type, and play games. What decides how well that works is mostly software. The decoders that turn neural activity into intent are machine learning models, and a lot of the hard problems in BCI now look like familiar ML problems: drift, scarce labels, and generalization across people.
This series is my notes on the work moving that frontier. Each part takes one paper or update and summarizes what was done and why it matters.