GPT-pro glitches leak internal monologue and end-of-text tokens

Multiple users reported bizarre GPT-pro outputs: one screenshot showed the model suddenly breaking into self-monologue about whether to print literal newlines, while a developer's misconfigured stop sequence leaked <|endoftext|> tokens into the visible response.

2026-10-30 ~ 2026-10-30 · 2 related posts

• Weirdest GPT-pro output yet: model leaks incoherent inner monologue into the reply — jessi_cata · 2026-10-30
• Dev mixes up stop sequence, gets GPT-Pro output polluted with <|endoftext|> — voooooogel · 2026-10-30