CONSTELLATION: Lattice

The pilot drew the line: fluency is not authority. Episode two stays with the landing. Lattice treats verification as a craft — slower than generation, less glamorous than a demo, and the difference between assistance and abdication. We open on invented citations, then the public record: Mata v. Avianca (S.D.N.Y. 2023), where ChatGPT-fabricated judicial opinions reached a federal docket and Rule 11 sanctions followed the humans who failed to check. A clearly labelled composite workplace scene shows how the same pattern arrives in ordinary drafts. Four claim kinds get different checks; evals that score polish can miss “does the source exist.” Takeaway: open the source yourself, or name the primary — or you still have a draft. Light cross-link to Circuit’s DNS episode. Narrated with an AI voice and edited by a human. Follow CONSTELLATION: Lattice for every Lattice episode.

A model can finish your sentence. It cannot owe anyone an apology. This Lattice pilot draws a hard line between fluency and authority: treat the model as a co-pilot whose proposals you verify, not an oracle whose answers you receive. Language models are useful for drafting, transforming, and exploring ideas, but fluency is not authority, and the risk is that we stop checking whether the answer is right. Transparency about assistance, data dignity, and authenticity over mere fluency. Fluency proposes. You dispose. Narrated with an AI voice and edited by a human.

Follow CONSTELLATION: Lattice for every Lattice episode.