> LLMs generate statistically plausible continuations of text. This works well for self-contained tasks like writing a function or drafting documentation because these are pattern‑extension problems. But pattern‑matching is not system understanding, and plausibility is not correctness.

closes tab