In context: The implications of today’s AI models are already remarkable, but adding a hyperrealistic human voice takes it to a whole new level. Sesame AI, founded by former Oculus CEO Brendan Iribe, introduces Miles and Maya, two AI voices that are making waves in the industry.
Sesame AI’s latest innovation is the Conversational Speech Model (CSM), which rivals the human-like qualities seen in Google’s Duplex and OpenAI’s Omni. While the realism of Miles and Maya has impressed many, accessing the technology is currently a challenge due to Sesame’s scaling efforts. Fortunately, a demo by Creator Magic on YouTube provides a glimpse of what the technology can do.
Using a multimodal approach, Sesame’s technology processes text and audio in a single model for more natural speech synthesis. Despite its human-like quality, the system still faces challenges in conversational context and flow. However, Brendan Iribe remains optimistic about closing these gaps in the future.
The societal impact of this groundbreaking technology has sparked mixed reactions, ranging from amazement to discomfort. The CSM’s ability to create natural conversations with imperfections like breath sounds and chuckles adds to its realism, prompting emotional connections from some users.
However, not all reactions have been positive. Some users, like PCWorld’s Mark Hachman, have felt uncomfortable interacting with the AI due to its eerily realistic qualities. This discomfort highlights the fine line between human-like AI and uncanny valley effects.
As AI technology becomes more personal and realistic, concerns about misuse and societal impact arise. The potential for scammers to exploit hyper-realistic voices for phishing scams is a significant risk. Voice cloning advancements could blur the lines between humans and AI, making identity verification more challenging.
While Sesame’s technology has legitimate uses, such as enhancing customer service and tech support, its potential misuse by cybercriminals poses a real threat. Open-source releases could make it easier for scammers to create convincing scambots, further complicating the landscape of AI ethics and security.
