12The model's JSON output breaks our parser about 2% of the time. Why, and what should we do?▼mediumNewStripeShopifyUber◆ premiumA small-sounding problem that becomes a big one at volume. The good answer explains why free-form generation produces malformed output and which controls make it near-impossible.Practise this question →Read answer →
32We forced the model to answer in strict JSON, and accuracy on our reasoning-heavy task dropped. Why, and what would you change?▼mediumNewOpenAIAnthropic◆ premiumConstraining output format helps parsing and can hurt thinking. Decide where the reasoning happens, then test the structure on your own eval.Practise this question →Read answer →