OpenAI is holding back new facial recognition features of large language model GPT-4. This due to concerns about recognizing specific individuals.
This is according to a report by The New York Times. Since OpenAI released its GPT-4-based version of ChatGPT, the company has been testing the new functionality. One of the key new capabilities is that ChatGPT recognizes images and can provide relevant information based on them.
The functionality appears to be so powerful that ChatGPT can now also provide details about individuals via facial recognition. This could raise major concerns around privacy. OpenAI reportedly fears that the functionality will tell more than just faces, such as the person’s gender or emotional status.
OpenAI is therefore ensuring that this functionality will not be available to the general public in the new GPT-4 version of ChatGPT, but only to a selected group of end users. The general public must continue to make do with text queries for now.
According to a researcher at OpenAI, facial recognition within AI solutions is not new, but a long-standing fact. Additionally, OpenAI’s AI model would not recognize all faces, although it might recognize images of famous people with a Wikipedia page.
Moreover, the facial recognition within OpenAI’s AI model would still have many hallucinations and therefore would not provide correct descriptions to images.