roughly 2 years ago, I knew very little about ML/AI, outside of prompting sonnet 3.5 and copy pasting output into files. since then I’ve been able to pretty easily/quickly pivot to doing all sorts of ML work (mostly inference infra today).
the big secret is a lot of people in industry are just figuring things out as they go and reading the same, free, papers you have access to right now!
like seriously, go talk to them. everyone is just tinkering with ideas. sure, it helps to come from a strong math background, or a strong CS background, etc. BUT there are people I work with today that come from very untraditional backgrounds and have wonderful minds for this.
figure out what your are interested in, and do that. the rest will figure itself out. just being able to locally host models is a fantastic start.