Are there any videos/streams of people writing non-trivial features for a non-trivial project (e.g. a more compact serialization format for Redis, a keybinding manager for Emacs, etc.) with significant output from ChatGPT/Claude? Nothing I could found on Youtube fits my non-triviality criteria so far. I want to know if I’m missing out by not using these AI tools properly.
IMO the problem lies on "magical thinking" that we can just say a spell and have the computer presenting the best solution ever without even knowing what we specifically want at the first space. Kind of like asking the AI for the answer to the meaning of life, the universe, and everything - "forty-two" (from <>) Personally, when I use them a lot of time I realised I am basically typing the whole detailed requirement specification out. First, it will be much more efficient for me to just do it myself. Second, in such cases existing technologies like formal methods seem more suitable as we can verify our specification's soundness too. And most modern approach to formal methods can do code generation too. Latest research (https://arxiv.org/abs/2404.04125v1) on image-text and text-to-image models shows that these pre-trained few-shots learners exhibit a log-linear relationship with "concept frequencies" and their performance on tasks. Hopefully this does not translate to LLMs but my bet is that they do.