Which is a common failure mode of prompt-based systems?

• A. Hallucination.
• B. Perfect compliance every time.
• C. Inability to generate any output at all due to fixed length.
• D. Consistently optimal results without errors.

Answer: (A)

Hallucination is a common failure mode in prompt-based systems. It happens when the model outputs information that sounds credible but isn’t grounded in the prompt or real data. These systems generate text by predicting the next word based on patterns learned from vast training data, not by verifying facts against a reliable source. When specifics are uncertain or data is sparse, the model may fill in with invented details that fit what it thinks the user expects, such as fabricated facts, misquoted statistics, or fictional citations. This is why strategies like fact-checking, requesting sources, or using retrieval-augmented generation are important to keep outputs anchored to reality.

The other options describe outcomes that aren’t representative failure modes you typically need to guard against. Expecting perfect compliance every time is an unrealistic expectation, since models can misinterpret constraints or produce unintended results. An inability to generate any output due to fixed length isn’t a standard failure mode; token limits can truncate longer responses, but the model usually still produces some content. And consistently optimal results without errors is not achievable with current systems, so the idea of always flawless performance doesn’t reflect how these models behave.