BREAKING: Ex-OpenAI CTO Mira Murati's AI Responds in 0.4 Seconds - The Race Against OpenAI Begins

Added: 2026-05-19

[00:00:00]Mira Murati may have just revealed the real reason she left OpenAI to completely change how humans talked to machines. And this time, the machine doesn't wait for your turn to speak.

[00:00:12]Two years ago, Mira Murati students staged at OpenAI and introduced GPT-4.

[00:00:17]Well, oh, that demo changed the internet overnight. For the first time, AI felt less like software and more like a presence. Natural pauses, interruptions, emotion, real-time voice.

[00:00:33]And it wasn't accidental. Mira Murati was one of the key people behind OpenAI's multimodal and voice roadmap.

[00:00:40]And now, her new company, Thinking Machines, has finally revealed what she actually wanted to build next. The company has just unveiled a new class of AI systems called interaction models.

[00:00:52]And the important word here is not so much about model. The operative term is interaction because according to Thinking Machines, today's AI system still behaves, well, like an email. You send something, you wait, the AI replies one turn at a time. But humans, well, we don't communicate like that. Real conversations overlap. People interrupt.

[00:01:16]People react while listening. People tend to notice visual cues in real time.

[00:01:21]It happens to all of us all of the time.

[00:01:23]And Thinking Machines says, "Current AI architectures fundamentally fail at this miserably. So, they rebuilt the stack."

[00:01:30]Their system processes audio, video, and responses simultaneously in 200 millisecond chunks. Meaning, the AI can listen, talk, see, and react at the same time. Not sequentially. I said, "Simultaneously." And that's called full duplex interaction. And the benchmark numbers are quite aggressive. Thinking Machines claims its interaction model reaches 0.4 second response latency.

[00:01:58]Wow. OpenAI's GPT real-time two reportedly sits closer to 1.18 seconds.

[00:02:05]That difference, I guess, sounds very small, but the magnitude of it in voice interaction, milliseconds decide whether something feels human or, well, robotic.

[00:02:18]And the timing of all of this is also quite crucial to understand because just days before Thinking Machines revealed its systems, OpenAI launched GPT real-time two. That's always a tongue twister for me. I don't know why.

[00:02:31]Real-time translation, real-time voice reasoning, streaming transcription. So, basically, it's pushing deeper into the exact category Mira helped create inside OpenAI.

[00:02:44]But Thinking Machines is making, I believe, a bigger architectural argument. OpenAI has improved real-time AI. Thinking Machines wants to redesign interaction itself. The company says, "Current models freeze perception while generating responses." Well, fair point.

[00:03:02]Its own system does not. It continuously processes the world while responding, which changes what AI systems actually become.

[00:03:12]Imagine customer support systems that respond naturally without these awkward pauses, factory systems that detect mistakes instantly from live video, medical co-pilots that continuously monitor environments in real-time. Oh, for that matter, AI companions that actually feel conversational instead of transactional. And behind all of this is one of the most intense talent wars in tech history.

[00:03:38]Thinking Machines raised $2 billion at a $12 billion valuation.

[00:03:44]Now, we talk about Meta. Meta has reportedly tried to acquire the company, then began hiring aggressively from it.

[00:03:51]OpenAI lost key researchers. Thinking Machines hired PyTorch creator Soumith Chintala as CTO. Everyone is of course poaching everyone because well, whoever owns the interaction layer may own the next interface after applications. And maybe that's the biggest story here.

[00:04:12]For a while now, AI companies have competed on intelligence. Now, guess what they're doing? They're competing on sheer presence. Yes. Who sounds more natural? Who reacts faster? Who feels more human? And Mira Murati seems to believe the future of AI will well, not be typed. It will be spoken.

[00:04:34]And here of course is the front page take, spoken.

[00:04:39]Clearly, this is becoming a race to build AI that humans can naturally live alongside. And you know what? In a very strange way, Mira Murati may have finally gotten her voice back. Do you feel so? If do, then please let us know what is your voice have to say in the comments below. This ladies and gentlemen is Front Page by the AI&M Network. Like, share, subscribe and always remember, think AI, think AI&M.