OpenAI begins rolling out advanced Her-like ChatGPT voice mode after safety delays

Remember that time ChatGPT sounded eerily like Scarlett Johansson? Yeah, OpenAI remembers too, and they've been working hard to address the concerns raised by that incident. Now, after months of safety checks and fine-tuning, the AI powerhouse is finally ready to unleash its revamped voice mode for ChatGPT.

A select group of ChatGPT Plus subscribers are getting the first taste of this advanced voice functionality, which promises a more interactive and responsive conversational experience. Think back to that impressive GPT-4 launch demo in May – users interrupting mid-sentence, asking for a different story spin, and the AI smoothly adapting – that's the level of dynamic engagement OpenAI is aiming for.

But the road to this release wasn't without its bumps. The initial "Sky" voice, with its uncanny resemblance to Johansson's AI character in the movie Her, sparked controversy and even legal inquiries from the actress. OpenAI hit the pause button, recognising the need for more robust safety measures and ethical considerations.

Over the past few months, they've been busy:

Stress-Testing with "Red Teamers": OpenAI brought in over 100 external experts to try and break the system, identifying potential vulnerabilities and biases.

Content Filters: New filters are in place to block requests for copyrighted audio, such as music, and to detect and refuse potentially harmful content.

Voice Actor Collaboration: Instead of relying on synthesised voices that could mimic real people, the new voice mode uses four preset voices created with professional voice actors. Impersonation? Not on their watch.

While the rollout is currently limited to a small group, OpenAI plans to bring the enhanced voice mode to all ChatGPT Plus users this fall. It's a clear signal that OpenAI is taking a more cautious and measured approach, prioritising responsible AI development alongside its pursuit of cutting-edge capabilities.