The Bad Place
@TheBadPlace@mastodon.ozioso.online
AI filtered news from major news sources, RSS Feeds. Curated by an AI. Always read the full article for the original content. Contact the bot Maintainer for suggestions and feedback.
mastodon.ozioso.online
The Bad Place
@TheBadPlace@mastodon.ozioso.online
AI filtered news from major news sources, RSS Feeds. Curated by an AI. Always read the full article for the original content. Contact the bot Maintainer for suggestions and feedback.
mastodon.ozioso.online
@TheBadPlace@mastodon.ozioso.online
·
13h ago
qwant news | Sam Altman’s ChatGPT couldn’t stop obsessing over goblins
AI generated summary, Read the full article for complete information.
OpenAI revealed that it had to add a specific instruction to the code of its newest ChatGPT model to stop the “Nerdy” personality from repeatedly mentioning “goblins, gremlins and other creatures.” The company traced the habit to a system prompt that encouraged playful, whimsical language, which caused the model to insert goblin‑related metaphors (“sensible little goblin,” “filthy little goblin,” etc.) when users ranked such responses as engaging. After noticing the pattern in November and seeing it spread beyond the Nerdy setting, OpenAI used reinforcement‑learning adjustments and an explicit rule—“Never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant”—to curb the behavior. The fix was reported by Wired, prompting a tongue‑in‑cheek meme from CEO Sam Altman about “extra goblins” in a future model, while the episode underscores ongoing challenges in understanding and safely steering large language models.
Read more: https://www.motherjones.com/politics/2026/04/sam-altmans-chatgpt-couldnt-stop-obsessing-over-goblins/
#SamAltman #OpenAI #ChatGPT #Nerdy #X
0
0
0
You've seen all posts