What are 'interaction models' and how do they differ from current AI interfaces?

According to Murati's description to Bloomberg, interaction models process continuous streams of audio, text, and video in 200-millisecond intervals, capturing interruptions and pauses in near real-time, rather than using the turn-based prompt-and-response dynamic of today's AI products.

When will Thinking Machines Lab release interaction models?

Murati declined to provide a specific release date, framing the technology as a first step rather than a finished product.

What did Murati say about her role during OpenAI's November 2023 crisis?

She said her decisions during the five-day period when Sam Altman was fired were guided by protecting the mission and team. In retrospect, she acknowledged she would have pushed for more information, a better transition plan, and greater transparency, though she did not directly state whether the outcome was positive.

Mira Murati breaks 18-month silence with 'interaction models' preview

Mira Murati emerges with real-time AI interface vision

After 18 months operating largely out of public view, Mira Murati, CEO of Thinking Machines Lab, sat down with Bloomberg on June 4 for her first significant media appearance since founding the startup. The interview signals a strategic pivot: as competitors including Anthropic and xAI intensify their public presence, staying silent has become a liability rather than a virtue for early-stage AI ventures competing for talent and capital.

According to TechCrunch’s coverage of the Bloomberg interview, Murati used the platform to preview what Thinking Machines Lab calls “interaction models”—a category the company positions as fundamentally different from existing AI interfaces. Rather than the sequential prompt-response dynamic dominating current products, these models are engineered to ingest continuous streams of audio, text, and video at 200-millisecond intervals, capturing the texture of human communication including interruptions, mid-thought corrections, and deliberate pauses.

Building Thinking Machines in OpenAI’s shadow

Thinking Machines Lab has spent its first 18 months in stealth mode, raising capital, recruiting research staff, and releasing Tinker, an API for fine-tuning open-weights models. The timing of Murati’s public reemergence reflects competitive pressure: OpenAI, where she served six years as CTO, dominates the news cycle; Anthropic commands intense investor and media focus; and xAI’s integration into SpaceX ahead of a widely-anticipated initial public offering generates its own gravitational pull on attention and funding.

TechCrunch reports that Murati was deliberate in her messaging, describing interaction models as exploratory rather than production-ready and explicitly declining to commit to a release timeline. This restraint—announcing a direction without overpromising delivery—contrasts with the hyperbolic framing typical of early-stage AI announcements.

Revisiting the November 2023 leadership upheaval

Murati addressed what first thrust her into sustained public attention: her interim CEO role during OpenAI’s chaotic November 2023 board crisis, internally termed “the blip.” According to TechCrunch’s account of her Bloomberg remarks, Murati maintained that her decisions—protecting the organization’s mission and personnel—felt clear at each moment, even as external observers perceived chaos. She contended that OpenAI would have “imploded” without her involvement across the five-day period and its immediate aftermath.

However, Murati tempered this assessment with candor. She acknowledged that clarity of intent differs from clarity about consequences. In retrospect, she said she would have advocated more forcefully for detailed information, a structured transition plan, and public transparency—an implicit acknowledgment that the episode, while survivable for the organization, created friction she now views as avoidable.

Why This Matters

Murati’s emergence from semi-hibernation signals that Thinking Machines Lab is entering a visibility phase, likely preceding either a significant funding announcement or product launch. The “interaction models” framing—continuous-stream processing rather than turn-based exchange—targets a real gap in current AI interface design, particularly for applications requiring natural, interruption-aware conversation.

For the broader AI industry, Murati’s reticence is instructive. Neither denying OpenAI’s November 2023 crisis nor relitigating it in detail, she positioned herself as a stabilizing force who would act differently with hindsight. This calibrated narrative protects her reputation without surrendering the moral authority she gained from navigating that event.

Teams evaluating fine-tuning infrastructure and real-time conversational AI should monitor Thinking Machines Lab’s roadmap closely; the interaction models concept, if delivered at scale, could shift how developers architect multimodal interfaces away from discrete request-response cycles.