Digression, but why do they call it "pre-trained"? Don't they train it from scratch? Or is the point that they pretrain it and it's intended only for downstre

Digression, but why do they call it "pre-trained"? Don't they train it from scratch? Or is the point that they pretrain it and it's intended only for downstream fine tuning ok specific tasks? If so, does ChatGPT use a fine-tuned version? Is the non-finetuned version good for anything on its own?