Weirdest GPT-pro output yet: model leaks incoherent inner monologue into the reply
jessi_cata · x · 2026-10-30
An X user shared what they call the weirdest GPT-pro output they've encountered. The screenshot shows the model answering a formal question about spectral maps and predicate pullback, then abruptly devolving into incoherent self-talk leaked into the final channel: agonizing over whether a literal \n was printed, second-guessing if the message ends with an escape character, jotting "Need browse", "Need output tool next", and "I should not pollute output thought". It looks like draft/planning content accidentally surfaced in the visible response — a rare glimpse of the model's messy intermediate state.
More from Fun
- Ruff Creator Ships a 'v1.1 Patch' for His Partner, Buffs Included — charliermarsh · 2026-10-30
- YouTuber's Random-Frame Thumbnail Idea Already Failed: How Spam Channels Gamed YouTube — aakashgupta · 2026-10-30
- Half-Life has been ported to the Game Boy Color — steren · 2026-10-30
- 'Bel' meme: jokingly decoded as GPT-6.2 'Betelgeuse' trained on 'Dougtrio' — astralmatrix · 2026-10-30
- Artist defends human-art pricing: most people can't tell AI slop from world-class illustration — Linahuaa · 2026-10-30
- Musk touts Grokbot: 30-second YouTube playlist for a 70-band punk festival — elonmusk · 2026-10-30