AI needs an editor too: how we moderate responses in CareerBot
AI is having a huge hype moment right now, but in production a model without filters is like an intern without onboarding: it might deliver solid work, or it might suddenly say something very strange. In AI CareerBot, we added output moderation.
AI is having a huge hype moment right now, but in production a model without filters is like an intern without onboarding: it might deliver solid work, or it might suddenly say something very strange. In AI CareerBot, we added output moderation: we block profanity, irrelevant answers, and potentially harmful advice.
If the AI still responds poorly, we fall back to either a safe templated reply or asking the user to rephrase the question. No magic, but no production cringe either.
We also tightened the HR scenarios: for topics related to resumes and interviews, we added another layer of checks for the ethics of recommendations. Because advice like “embellish your experience” is not career growth — it’s a fast track to becoming a meme.
In short, we’re making AI not just smart, but also sane.