OpenAI is sharing early insights from a trial of a feature that can read text in a way that sounds remarkably human-like. This represents a big step forward in artificial intelligence, but it’s also raising concerns about potential deepfake risks.
The company is showing off some initial demonstrations and practical uses from a small-scale test of its text-to-speech model called Voice Engine. This model has been given to about 10 developers so far, according to a spokesperson. Even though OpenAI talked about the feature with reporters earlier, they’ve decided not to release it widely.
They made this decision based on feedback from various groups like policymakers, industry experts, teachers, and creative professionals. Originally, they planned to offer the tool to up to 100 developers, but they changed their plans.
In a blog post, OpenAI acknowledged the serious risks of creating speech that sounds so human-like, especially during an election year. They mentioned that they’re working with different partners, including governments, media, entertainment, and education, to get their input and make sure they’re moving forward carefully.
