OpenAI delays ChatGPT voice mode citing security assessments

OpenAI had initially deliberate to supply the brand new options to some paying subscribers in late June however is suspending that preliminary launch by a month, the corporate stated in a statement on X. The options will likely be accessible to all paying customers within the fall, the corporate stated, although it added the caveat that “actual timelines rely upon assembly our excessive security and reliability bar.”

OpenAI first added the flexibility for ChatGPT to talk in a one in every of a number of artificial voices, or “personas,” late final yr. The demo in Might used a kind of voices to point out off a more moderen, extra succesful AI system referred to as GPT-4o that noticed the chatbot communicate in expressive tones, reply to an individual’s tone of voice and facial expressions, and have extra complicated conversations. One of many voices, which OpenAI referred to as Sky, resembles the voice of an AI bot performed by Johansson within the 2013 film “Her,” a couple of lonely man who falls in love together with his AI assistant.

OpenAI CEO Sam Altman has denied that the corporate skilled the bot on Johansson’s voice. The Washington Put up reported last month that the corporate employed a distinct actor to supply coaching audio, based on inner data and interviews casting administrators and the actor’s agent.

Because the world’s largest tech firms and upstarts like OpenAI race to compete in generative AI, a number of initiatives have run into surprising hurdles. Final month, Google scaled again how typically it reveals AI-generated solutions on high of search outcomes, after the software made unusual errors similar to telling individuals to place glue on their pizza. In February, the search firm withdrew an AI picture generator criticized for creating images such as a female pope. Microsoft final yr made adjustments to its personal AI chatbot after it typically supplied weird and aggressive answers.

OpenAI stated Tuesday that it wanted further time to make the brand new voice model of its chatbot higher at noticing and blocking some content material, with out disclosing specifics. Many AI instruments have been criticized for making up false data, spouting racist or sexist content material, or displaying biases in their outputs. Designing a chatbot that makes an attempt to interpret and mimic feelings will increase the complexity of its interactions, opening new methods for issues to go improper.

“ChatGPT’s superior Voice Mode can perceive and reply with feelings and nonverbal cues, transferring us nearer to real-time, pure conversations with AI,” OpenAI stated in its assertion. “Our mission is to deliver these new experiences to you thoughtfully.”