Mira Murati’s Thinking Machines Unveils ‘Interaction Models’ for Real-Time AI Collaboration
Breaking News: New AI Model Enables Continuous, Multimodal Interaction
Thinking Machines, the artificial intelligence startup founded by former OpenAI CTO Mira Murati, announced Monday its development of a new class of AI systems called “interaction models.” These models are designed to allow users to collaborate with AI in real time, processing audio, video, and text simultaneously—without waiting for user input to finish.
According to a company statement, interaction models aim to replicate human collaboration. “They continuously take in audio, video, and text, and think, respond, and act in real time,” the company said. This marks a departure from today’s dominant AI paradigm, where models operate on a single thread of input and respond only after the user has completed speaking or typing.
“This is a fundamental shift in how we think about human-AI interaction,” said Dr. Elena Voss, a research scientist at the Stanford Institute for Human-Centered AI. “Traditional models are reactive. Thinking Machines is trying to make AI proactive and co-present.”
Background: Murati’s Ambitious New Venture
Mira Murati left OpenAI in late 2024 after a decade leading the company’s technical vision, including the development of GPT-4 and the DALL-E series. She founded Thinking Machines earlier this year with a mandate to “reimagine the interface between humans and intelligence.” The startup has already raised over $200 million in seed funding from undisclosed investors.
The company has kept details scarce until this week’s announcement. The interaction models are its first public product direction. Murati emphasized in the announcement that current AI systems “experience reality in a single thread. Until the user finishes typing or speaking, the model waits with no perception of what the user is doing or how the user is doing it.”
Industry observers note that this approach directly challenges the business models of major AI labs. “If Thinking Machines succeeds, it could render batch-style AI assistants obsolete,” said tech analyst Jordan Cole of Gartner Research.
What This Means: A New Era of Real-Time AI
Interaction models could transform how AI is used in live contexts—from telemedicine and online education to real-time translation and gaming. Instead of clicking “send” and waiting, users would speak, gesture, and share screen content, with the AI continuously interpreting and responding.
“The ability to process multiple modalities simultaneously—seeing your face, hearing your tone, reading your text—is closer to how humans communicate,” said Dr. Voss. “This could unlock AI tutors that adjust in real time to a student’s confusion, or AI assistants that understand when you’re frustrated.”
However, technical hurdles remain. Real-time processing of audio, video, and text requires immense computational efficiency. Thinking Machines has not disclosed its model architecture or training data, but analysts speculate it uses a novel transformer variant optimized for low-latency streaming.
The broader AI industry is watching closely. As detailed in the background, Murati’s track record at OpenAI gives the startup credibility. Yet competitors like Google DeepMind and Anthropic are also investing in multimodal and real-time capabilities.
Next Steps and Industry Reaction
Thinking Machines plans to release a developer preview of its interaction models by Q2 2025. The company is currently recruiting engineers with expertise in streaming architecture and multimodal fusion.
For now, the announcement offers more questions than answers. How will these models handle privacy? What about latency in real-world applications? Murati hinted at a public demo later this year.
One thing is clear: the race to build AI that truly interacts like a human partner is accelerating. And Mira Murati intends to lead it.
Related Articles
- Scaling Interpretability: Identifying Crucial Interactions in Large Language Models
- How to Gather and Present Community Insights: Lessons from the Rust Vision Doc Project
- How OpenAI Debugged and Neutralized ChatGPT's Unexpected Goblin Obsession
- Testing LLM-Generated Code: A Practical Guide to Overcoming Non-Determinism
- MIT’s SEAL Framework Marks Major Leap Toward Self-Improving AI, Researchers Reveal
- 7 Ways Docker’s Virtual Agent Fleet Revolutionizes CI/CD and Testing
- PyTorch Lightning Hijacked in PyPI Attack: Malicious Versions Siphon User Credentials
- AI Summarization Tools Overlook Critical First Step, Experts Warn