No not the opposite – they just looked at a particular subsets of tasks, datasets, and metrics. There are some things that LoRA is sufficient for.

→ View original post on X — @jeremyphoward,