Comment by jpalomaki

Isn't training material the biggest problem for truly open source LLMs (such that could compete with top tier models)? The computation part can be solved with money, but compiling a comprehensive training set that could be freely shared and free of copyright issues is pretty much impossible.