Share
ChatGPT's artificial intelligence is back in the spotlight after users noticed strange behavior in its speech model. OpenAI has already issued an official statement on the matter.
Main theses:
- ChatGPT started mentioning “goblins” 175% more often than before, and “gremlins” by 52%.
- The OpenAI team admitted its guilt in what was happening.
What's wrong with ChatGPT again?
Many questions arise for users when working with GPT-5 language models.
The main problem is that in conversations they increasingly began to use metaphors with “goblins”, “gremlins” and other fictional creatures.
Against this background, users began to complain about the excessive “familiarity” of the chatbot's responses.
The OpenAI team finally decided to explain what really happened.
According to company representatives, the main reason for this behavior of the bot was the algorithm for training models for specific “personality” types, which users could then choose to communicate with ChatGPT.
The problem was particularly evident in the bot with the Nerdy personality type (slang for “nerd”). Although the Nerdy personality accounted for only 2.5% of ChatGPT responses, this version was responsible for 66.7% of all goblin mentions.
Share
As the developers note, when they trained the Nerdy version for a specific “nerdy” manner of communication, they unconsciously assigned it a particularly high priority for using “creature metaphors.”
From there, goblins spread, OpenAI officially acknowledged.
Share