A community-theater desk claims a customized model writes better playbill synopses but never scored the unmodified pretrained checkpoint on the same gold synopses. What is missing from the experiment?
Select an answer to reveal the explanation.
Short Explanation
“Better playbills” needs a baseline: the unmodified pretrained checkpoint, or the prior-best recipe, scored on the same gold synopses. Without that run, customization cannot be credited. Occupancy kernels and leaked test peeks do not replace it.
Full Explanation
A treatment is “better” only relative to a comparable baseline on the same gold set. The unmodified pretrained checkpoint — or the prior-best recipe — is that baseline and must be scored with the same metric script. Without it, customization cannot be credited. Occupancy kernels, repeated test peeks, and datacenter rack plans do not replace the missing control run.