LLM capabilities and limits (overview)

At their best, LLMs are remarkably versatile: they draft and edit writing, translate between languages, explain ideas at any level, write and debug code, summarize long documents, and brainstorm. Because the interface is plain language, a non-expert can get useful results in seconds — that accessibility is much of why they spread so fast.

But the same machine has hard limits. It can state false things with total confidence, a failure called hallucination. Its knowledge is frozen at training time, so it misses recent events unless given tools. It is weak at exact arithmetic and long multi-step logic, sensitive to how you phrase things, and unable to truly verify its own claims. And it reflects the biases in its data.

The healthy mental model is a brilliant, fast, widely-read assistant that is also confidently unreliable. Treat its output as a strong draft to check, not a final authority — especially when the stakes, the facts, or the math actually matter. Used that way, the strengths are real and the limits are manageable.