OpenAI CEO Sam Altman has revealed that a recent update to the company’s ChatGPT-4 (GPT-4) chatbot has been unsuccessful and needs remedial action.
In a post on the company’s research site, Altman disclosed that the aim of the upgrade was to make GPT-4 ‘more intelligent’ and ‘more human’, but instead it had become irritatingly obsequious and over-enthusiastic in its responses.
Examples given included congratulating users for revealing mental health issues or unorthodox religious claims, with some critics suggesting that such uniform praise could be potentially dangerous.
Following the update, some users have posted screenshotted examples of GPT-4’s uneven responses, with praise heaped on the user regardless of what is posted.
The post was removed from the site shortly after it was posted, but The Verge captured its key points before it was taken down.