the barrier to entry for training good embedding models is 100x lower than for training good LLMs – requires significantly less compute, expertise, and data. seems like a good area for building a business

→ View original post on X — @jxmnop