Mira Murati reveals 'interaction models' that process conversations in real time

Mira Murati, former CTO of OpenAI, made his first public appearance in 18 months to present the technical vision of Thinking Machines Lab. The company is developing 'interaction models' that process continuous streams of audio, text and video at intervals of 200 milliseconds to capture the texture of conversations. There's still no launch date for technology.

How do real-time interaction models work?

Unlike traditional models that process static inputs (a text prompt, an image), Thinking Machines Lab 'interaction models' operate in real time. They receive continuous streams of audio, text and video, analyzing each modality simultaneously at intervals of 200 milliseconds. This allows capturing nuances such as pauses, tone of voice and facial expressions, something that current AI systems still struggle to replicate. The proposal is to create a more natural and fluid dialogue experience, like a conversation between humans.

Why do interaction models matter to Brazilian developers?

For the AI community in Brazil, this approach opens doors for applications in call centers, virtual assistants, distance education and even accessibility tools. Imagine an assistant who not only understands what you say, but as you say — detecting frustration, enthusiasm, or confusion. Companies that already invest in trading foundation models, such as banks and insurance companies that create their own intelligence, can find in the interaction models a natural evolution for more human customer care.

When will the interaction models be available?

Murati didn't release launch dates. The company, which still operates in a very private way, indicated that the models are in a research and development stage. Thinking Machines Lab was founded by former members of OpenAI and other big techs, and the current focus is to refine the continuous multimodal processing architecture before any commercial rollout. Meanwhile, the market observes with expectation, especially after the recent case in which open-weight AI created a corporate worm that spreads itself on corporate networks, showing that security and latency are key challenges.

What is lacking for large-scale adoption of interaction models?

For interaction models to become reality in products, Thinking Machines Lab needs to solve infrastructure and cost bottlenecks. Processing three simultaneous data streams every 200ms requires considerable computational power, and optimization to run on common devices (cell phones, laptops) is essential. In addition, data privacy is a concern: Continuous processing of audio and video raises questions about consent and storage. The company hasn't detailed its data policy yet.

Source: read the original ad in LetsDataScience

Frequently Asked Questions about interaction models

Will interaction models replace traditional chatbots?

Probably not in the short term, but they represent a significant advance by capturing the non-verbal context, making dialogues more natural in complex scenarios.

What practical applications can these models have in Brazil?

Customer service, telemedicine, online education and sales tools can benefit from real-time analysis of tone and expressions to improve user experience.

How does Thinking Machines Lab differ from OpenAI?

Founded by former OpenAI members, Thinking Machines Lab focuses on continuous multimodal interaction rather than sequential models, promising lower latency and higher contextual fidelity.