Classic GPT-3 Fail: Mid-Sentence Aurora Explanation Erupts Into Profanity

flowersslop · x · 2026-08-13

A user shared a screenshot from the GPT-3 era: the model was seriously explaining the best time to see the aurora in Sweden, only to suddenly glitch mid-sentence (right at 'n-') and start violently repeating profanity (FUCK YOU). A classic, funny alignment failure case from early LLMs.

Original post →

More from Fun

Fun channel →