Mislabeled image

[rajdeepint]

The second image in Transfer Learning Section in chapter1/4.mdx says it requires "$$$ Compute" to finetune a model, which is incorrect in comparison to the first image.
The first image claims that it is very expensive ($$$) to pretrain a language model from scratch and then the next section logically implies that is far less expensive to take a pretrained model an then finetune it using a smaller dataset, so the compute shouldn't be $$$ be expensive but probably a $ expensive.