Image: OpenAI
OpenAI’s new advanced voice mode for ChatGPT is starting to roll out to a small number of people who subscribe to ChatGPT Plus. The feature, which OpenAI showed off at its GPT-4o launch event in May, was criticized for sounding similar to Scarlett Johansson and was later delayed for safety reasons.
At OpenAI’s event, the new voice mode appeared to be noticeably more capable than ChatGPT’s current voice mode. Onstage, OpenAI employees were able to interrupt and ask the chatbot to tell a story in different ways, and the chatbot took their interruptions in stride to adjust its responses.
The advanced mode was set to release in alpha in late June, but OpenAI delayed the rollout by one month to “reach our bar to launch.” As part of that delay, the company said it was “improving the model’s ability to detect and refuse certain content.” OpenAI spokesperson Taya Christianson says the company tested the voice model’s capabilities with more than 100 external red teamers (people who try to attack technologies to find weaknesses). The company has recently faced a lot of scrutiny about its safety policies, so this pause may have been the right choice. OpenAI has also “added new filters that will recognize and block certain requests to generate music or other copyrighted audio,” Christianson says.
During OpenAI’s event, one of the key criticisms of the new mode was how much the onstage voice, dubbed “Sky,” sounded like Johansson, who played an AI personality in the movie Her. While the voice had been in ChatGPT before OpenAI’s spring demo, the company pulled it shortly before Johansson revealed she had sent letters to OpenAI asking how the voice was made. Christianson says that ChatGPT’s new mode will only use four preset voices it made with voice actors, adding, “We’ve made it so that ChatGPT cannot impersonate other people’s voices, both individuals and public figures, and will block outputs that differ from one of these preset voices.”
OpenAI plans to bring the new mode to all ChatGPT Plus users in the fall, according to Christianson.
Leave a Reply