Bonkers compute only in the beginning. Over time it'll reduce as models are made more efficient.
Or it will stay the same as the efficiency gains will be eaten up by bigger models
alt Hacker News
Or it will stay the same as the efficiency gains will be eaten up by bigger models