cadamsdotcom
a day ago
My suggestion: upskill to match a growing need.
No formal education really exists in this field yet. So don’t get scammed. Instead, look at everything Andrew Ng, Andrew Karpathy, and HuggingFace are doing. The recently published HF book on training from scratch looks interesting, go learn to train some models from scratch. Go put some cash down on OpenRouter and the A/B blind test websites, and play with open models enough to have an overly excited conversation in a bar, waving your beer and pontificating about your favorites and why China is or isn’t winning — in your opinion.
You say you’ve tried side projects and such, but their exact nature will either color you as a normie, or someone who could help a lab make forward progress. Bad: “I typed a prompt and look what I got!” Better: “I used a vibecoding tool to get started then iterated with Claude Code / Codex, pulled in some stuff I did in Figma Make, and got ChatGPT to walk me through setting up the infra to get the app deployed”. Even better: “I wanted to understand nanochat so had Gemini help me walk through it, can I explain it on a whiteboard?” Even better: “I ported nanochat to Rust with the help of Gemini, that was really cool but man the borrow checker gave me grief with X and Y and Z.”
There are people making amazing things at every level of the stack which demands net new hiring (as opposed to hiring to backfill, which is going to zero).
Find where you have the most fun, play there, and match it to some net new hiring need in the industry.
Welcome to an incredible journey!
mattfrommars
a day ago
Thank you for the detail reply. I feel less overwhelmed now and sort of idea for next step to take. I've noticed Andrew NG deep learning specialization to be frequently mentioned along with Karpthy Zero To Hero free course.
I have toyed around free open source models, gemma3n, and running locally, etc.
As per project I did, it was to load up vector database and built a 'mini' agentic chat bot. Nothing fancy like fine tuning a model.
Ambitious project I've been thinking is to build an inference on top of established model, for example inference on deep seek model or llama in Python or Rust. But I run into self doubt since I don't have any formal ML background, so I'm not sure how I can pull it off.
I do keep an eye on trending github project, I'm amazed on project people make. My recent interest has been coding agent that operates within CLI and coding agent that work off git workspace.
This brings back a point I should have mentioned, since I don't have a picture, I am unable to prioritize on which step to take first and from there, branch off.