[palmotea]

How does the compute required for that compare to the compute required to serve LLM requests? There's a lot of goal-post moving going on here, to justify the whataboutism.