On Monday, OpenAI introduced its latest AI model, GPT4o, which responds fluently and without hesitation. It can even detect emotions in the user’s voice and respond accordingly. The capabilities are impressive and somewhat unsettling.

OpenAI’s ChatGPT creators showcased the newest version of their AI, GPT4o, on Monday evening. OpenAI CTO, Mina Murati, demonstrated how this AI, named GPT4o, not only responds faster in conversations but also provides higher quality answers. Notably, the new version integrates audio recognition and output directly into the model. This allows the OpenAI AI to conduct conversations without needing keyboard inputs or audio recognition programs.

The results, demonstrated by Murati alongside two senior developers, were astonishing: the AI responds fluidly and without stuttering. More impressively, it recognises emotions in the user’s voice and responds appropriately. If someone is scared, the AI shows empathy and can reply in a cheerful, calm, or dramatic tone.

Additionally, the AI can simulate emotions on demand. For instance, Murati showed how the AI could tell a bedtime story while considering the user’s preferences. The conversation with the AI was also impressive because it responded to interjections and interruptions like a human would, rather than delivering pre-recorded answers like an audio recording.

Of course, the demo at OpenAI’s headquarters was a well-rehearsed exercise, but Murati insists that the interaction was live, not pre-recorded. If OpenAI is not bluffing, then GPT4o’s ability to simulate emotions is particularly impressive and somewhat unsettling.

The AI is designed to perform this in not only English but also 49 other languages. “We aim to bring this experience to every user,” Murati explained, announcing that the new model would eventually be available to all users worldwide, not just paying customers.

OpenAI can afford this because the new AI manages computing resources more efficiently than its predecessors. This efficiency means OpenAI saves money by making this version available to all users.

Initially, however, OpenAI’s “Red Team” will test the AI and push its limits. The Red Team will ensure that the AI cannot be used for malicious purposes. They will also identify and close any backdoors or tricks that could be exploited to bypass built-in safeguards before general users can take advantage of them.