Model Blatantly Copies the Answer Key
goodside · x · 2026-07-16
The author complained about posting that a certain LLM couldn't solve a large crossword puzzle, only for someone to reply, "It actually can."
Upon checking the chat history, the author discovered the model answered correctly by directly quoting the answer key from his own thread and even linked back to the original post in its response.
More from Fun
- Reddit shares an AI-generated mini movie called The Lunar Ship — Ermajean12 · 2026-07-21
- An AI first encounter that broke a physics engine in GDevelop — tristanbob · 2026-07-21
- “This is Java, not JavaScript,” says the Codex user with the 50-character function name — zeeg · 2026-07-21
- A meme says AI slop sounds much better when you call it technical debt — rudrank · 2026-07-21
- Engineer says an LLM could profile teammates from git history and chat logs — remoteDev1 · 2026-07-21
- Nathan Benaich turns AI hype cycles into a one-line meme — nathanbenaich · 2026-07-21