What do data scientists need to know about LLMs to be competitive?

Question

I have noticed job postings asking for experience with LLMs, but they are a bit vague. The most specific requirements I found were RAG, fine-tuning, and PyTorch.I'd like to understand what I need to learn about LLMs to be competitive for data science positions. I have no experience with deep learning, generative models, and NLP, but I have a Ph.D. in a CS-adjacent area.

hall0ween · Accepted Answer

One thing to know about data scientists are (1) there's the job posting and (2) there's the work. Often you'll be hired for one thing and end up doing entirely different work.
To improve your chops in this field, (a) learn the basics of NLP, (b) build yourself a RAG using llamaindex or langchain (or other but build it).
As for fine tuning and deep learning, without experience in that field it's tough. It's like any complex thing (fine tuning not so much but deep learning), knowledge comes with time and exposure. So go find a reason to fine tune or build and train a neural network and go friggin do it.

willd13 · Answer

You say you have no experience with deep learning. Start by learning about "traditional" deep learning models, then move on to LLMs. Learning LLMs before deep learning is like learning algebra before addition."Deep Learning with Python, Second Edition" by Google's Fran&ccedil;ois Chollet is a pretty good intro, and also touches on LLMs / generative models towards the end.