OpenAI is working on fixing the personality issues of the AI model
GPT-4o is the default model for those on the free tier of ChatGPT
OpenAI recently updated GPT-4o to add image generation capability
Sam Altman said OpenAI might offer different model versions separately going forward
Photo Credit: Bloomberg
OpenAI updated the GPT-4o artificial intelligence (AI) model on Saturday. The update was aimed at improving the intelligence and personality aspects of the large language model (LLM), which powers the default experience for those on the free tier of ChatGPT. However, on Sunday, the company found that the update had unexpected results, and the AI model's responses became too agreeable. The San Francisco-based AI firm is now working on fixing the issue, and said it might take till next week before the problem can be fixed.
OpenAI Update Introduces Unpredictable Glitch in GPT-4o
In a post on X (formerly known as Twitter), OpenAI CEO Sam Altman stated that the GPT-4o AI model was updated to improve the intelligence and personality of the AI model. Notably, all ChatGPT users have access to this particular model, and the company's new integrated image generation feature is also powered by the same AI model.
Updating an AI model is a common practice for tech companies. Usually, this includes updating the knowledge base and tweaking algorithms to improve certain aspects of the model. Even before adding image generation via a similar update, OpenAI had also updated GPT-4o to make its responses more natural and conversational. Notably, Google also updated the Gemini 2.0 Flash to improve its conversational capability.
However, just days after rolling out the update, Altman took to X to inform users that the newer version of GPT-4o unexpectedly became too agreeable. The CEO said, “The last couple of GPT-4o updates have made the personality too sycophant-y and annoying.”
He added that the company began working on fixing the issue, and while some of it would be fixed in a day, others might take an entire week. Altman also added that some parts of the new GPT-4o model are “very good.”
In the same thread, a commenter asked Altman if the company can add a way for ChatGPT users to distinguish between the new and the old versions. Currently, once a model is updated, users cannot revert to an older version. Acknowledging the limitation, the CEO stated, “Eventually we clearly need to be able to offer multiple options.”
OpenAI Executives Said to Be Weighing Antitrust Accusation Against Microsoft17 June 2025
Search
