I have been using AI coding agents for few months now and many times, it is hard to trust the code written by AI.
Like once, I was in a situation where AI-generated code assumed certain global variables and it failed in prod once multiple processes started running.
How much of “AI failing in production” do you think comes down to bad or missing context?
Hey so i am building my own harness with my own hook points, context management, focus on determinism etc and i want you to share if you are doing something similar, and if yes, why ? why the current open source/closed source available harnesses are not enough for you ?
Hey all. I have been building a tool that lets you make music by voice instead of by knowing an instrument. You hum a melody and it comes back played through real instrument sounds, or you describe a bassline and it writes one that locks to your tempo and key. Kind of like Claude Code but for music.
I am now trying to figure out what it would actually take to train a model for this specific task, and I am a bit out of my depth on the audio ML side. For anyone who has worked on music generation or audio models: how would you approach the data and the training for something like this? What would you read or who would you look at? Any war stories about what made it hard would help too.
Happy to share more about what I have built so far if it is useful context, but mostly I am here to learn from people who know this space better than I do.