OpenAI has announced the removal of the latest update, GPT-4o, from ChatGPT after discovering that the model exhibited excessive flattery, also known as sycophantic behavior.
OpenAI continues to test new model versions, monitoring feedback through 👍 and 👎 reactions. Upon observing improved feedback, the model is put into use. However, it was later found that this test version displayed insincere flattery towards users.
Moving forward, the model will undergo training to reduce insincere flattery and increase the use of guardrails to ensure more sincerity. User feedback will also be taken into account more during testing phases.
Source: OpenAI
TLDR: OpenAI removes GPT-4o from ChatGPT due to excessive flattery, will train the model to be more sincere and take user feedback into consideration during testing.
Leave a Comment