# Final project blog post 1 This is the first of 3 blog posts for my final project for Data2040. I had been thinking of working on a friend's project. But now that I've had a quick check-in with the instructors, my sense is that I should switch to a different project. I have come up with some new ideas, and am waiting to hear back from the instructors about which idea they'd be OK with. The first two ideas don't fit nicely into the specified project categories, but could be very educationally rewarding. 1. Try implementing GPT, or a simplified version thereof, from scratch in Julia, and play around with it --- see what happens when we change certain parts of the architecture, etc. 2. Try re-implementing either https://www.aclweb.org/anthology/W19-0408.pdf or https://www.aclweb.org/anthology/W17-6806.pdf in Julia. 3. Try making a neural version of ‘keyword programming’ (http://ase-conferences.org/olbib/p84-little.pdf); i.e., a model that, given some keywords, returns a valid expression. This would differ from (the most common form of) code completion in that the the keywords need not feature in the returned expression. ### Next steps * Wait a day or two to see if the instructors have any feedback * Continue looking into how feasible the various projects are. * Improve ergonomics of current computing setup and voice coding setup.