A mayor’s office must evaluate both safety and tone quality of a public town-hall summary generator before wider release. Which approach best fits that combined evaluation?
Select an answer to reveal the explanation.
Short Explanation
Rolling out a town-hall summarizer without checking safety and tone is like opening city hall doors without a metal detector or a greeter. Paired evaluations catch harmful output and off-key tone before residents see it. You'll miss the real risk if you're pairing just jailbreak with tone, protected material with tone, or hate with tone, since it's broader harmful content plus tone you've got to catch together.
Full Explanation
A mayor's office town-hall summary generator must be evaluated for both safety and tone before wider civic release, because harmful content and off-key municipal voice are distinct failure modes that separate quality scoring and safety-log checks do not catch together. Paired safety and quality evaluations that score harmful content and summary tone together give the office a joint go/no-go signal aligned with public-communications risk: fabrications, toxicity, and inappropriate register are measured on the same release candidate. That combined evaluation fits Foundry generative-app readiness for public summaries. Running paired safety and quality evaluations that score jailbreak risk and summary tone together gets tone right but misses the safety gate for this release, so prompt-attack vulnerability and municipal tone are measured while broader harmful content in the output goes unscored on the same release candidate. Running paired safety and quality evaluations that score protected material and summary tone together also misses the combined gate, so copyrighted-text reproduction and municipal tone are measured while broader harmful content goes unscored on the release candidate. Running paired safety and quality evaluations that score hate content and summary tone together is too narrow for this gate, so only hateful output and municipal tone are measured while broader harmful content goes unscored on the release candidate. On the exam, combined safety-plus-quality evaluation is the release gate; distractors pair jailbreak risk with tone, protected material with tone, or hate content with tone without scoring harmful content and summary tone together. Caveat: paired scores still need representative town-hall fixtures (heated sessions, multilingual remarks, corrections) or the evaluation understates real municipal risk. Operational check: run the paired safety and tone suite on a held-out set of town-hall transcripts, block release on failing thresholds, and sample failing summaries with communications staff before expanding access.
Prefer to study offline? Printable PDF study book + test engine ($7.50, unofficial). These questions stay free here.