There's no single number. The trap is reaching for perplexity or BLEU because they're easy to compute, then optimizing a score that has nothing to do with whether the output is good. The move is to match the metric to the task and name where each one lies to you.
Updated Aug 2026 · Grounded in real Forward Deployed Engineer interview loops and written to a senior-engineer editorial bar.
There's no single number. The trap is reaching for perplexity or BLEU because they're easy to compute, then optimizing a score that has nothing to do with whether the output is good. The move is to match the metric to the task and name where each one lies to you.
The discriminating follow-up is 'your BLEU went up but users complain the answers got worse, what happened?' The strong answer names n-gram metrics as surface-overlap proxies that reward copying reference phrasing and punish a correct paraphrase, so they decouple from quality on open-ended generation. Candidates who treat 'higher BLEU = better' as a law fail here.
No comments yet — be the first to share your approach.
