A town clerk needs everyday English rewrites of public notices and has almost no labeled examples. A creamery needs a fault predictor from a private 12-year pasteurizer log. Which pairing of model sources is accurate?
Select an answer to reveal the explanation.
Short Explanation
Think of borrowing a well-read dictionary for everyday rewrites, and building a private recipe book for the pasteurizer. Little labeled notice text is a job for an open pretrained or foundation model. A 12-year private log is a job for a custom model. Clustering filenames does not pick the source.
Full Explanation
Everyday English rewrites of public notices with almost no labeled examples suit an open pretrained or foundation model: broad language ability already exists, and the town clerk needs fluent rewriting more than a from-scratch classifier. The creamery's fault predictor from a private 12-year pasteurizer log is the opposite situation—narrow industrial behavior, proprietary history, and a predictive task that benefits from a custom model trained on that log. Model-source choice follows data scarcity, domain specificity, and whether general language knowledge transfers.
Insisting on a custom model for the notices because general language always needs thousands of local labels overstates labeling needs. Foundation models are pretrained for language; light prompting or light adaptation often beats collecting a huge local rewrite corpus for routine notices.
Using only an open pretrained model for the pasteurizer log because private logs cannot be used for training is wrong on both policy and technique. Private operational logs are routinely used to train custom models under appropriate controls; an open general model will not magically know creamery fault patterns without that domain data.
Saying both jobs require clustering filenames avoids the model-source decision entirely. Clustering file names does not rewrite notices or predict pasteurizer faults.
Exam caveat: foundation versus custom is not always either-or forever—teams may start with a foundation model for language and still train custom models for equipment telemetry under the same municipal or plant program.
Operational check: for notices, measure rewrite quality and prohibited-content filters on a small civic review set; for the creamery, confirm training uses only authorized private log fields with leakage-free time splits before any fault alerts reach operators.