Mira Murati's First AI Model Is Built on China's Blueprint... Wild

Mira Murati's First AI Model Is Built on China's Blueprint... Wild

by AI Revolution · View the channel on PSI

Humanities, Social Sciences & Thought AI — Models & Research UYQArtificial intelligence (AI)UYQFGenerative artificial intelligence
🎙 AI Revolution 👥 568K 📅 July 16, 2026 ⏱ 16 min 👁 42K 📄 news review 🧭 2026-09-14
Available in: English (current) Français

Keywords

InklingMixture of ExpertsDeepSeekOpen WeightsAI Geopolitics

Summary

The video reports on the release of Inkling, the first AI model from Thinking Machines Lab, founded by former OpenAI CTO Mira Murati. Inkling is a mixture-of-experts transformer with 975 billion total parameters, 41 billion active, and a 1M token context window. It is fully open-weight under Apache 2.0, supporting text, images, audio, and video. The model was trained in under 9 months on synthetic data partially generated by China’s Kimi model and its architecture largely follows DeepSeek V3. Benchmarks show it trails leading models like GLM 5.2 and Claude Fable 5, but it excels in efficiency and balanced performance across tasks. The video highlights a controllable reasoning dial and training techniques that improve calibration and reduce hallucination. It demonstrates self-fine-tuning and agentic capabilities. The core narrative emphasizes that an American lab built on Chinese blueprints, highlighting geopolitical double standards in AI discourse. The business strategy centers on the Tinker fine-tuning platform rather than the model itself, aiming to fill a Western open-source void. The video concludes with details on Inkling Small and availability across major platforms.

177 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video offers significant value by providing a timely, detailed analysis of a major AI release from a company with high-profile leadership. It combines technical specifications, benchmark comparisons, and business strategy, placing them in a geopolitical context that is often overlooked. The argumentation is solid: specific numbers are cited, sources are reputable, and the video acknowledges the model’s limitations while emphasizing its strategic importance. The analysis of the double standard regarding Chinese vs. American distillation is compelling and well-supported. However, the video also contains a promotional segment for a commercial guide, and its tone is somewhat opinionated, which may color the objectivity of the presentation.

Scientific Rigor, Source Quality, Title Accuracy

The scientific rigor is high for a YouTube commentary: the video draws on official announcements and established tech media, providing direct links in the description. It cites concrete benchmark scores and training details, and even mentions external audits by Cognition and Fortress. The sources are diverse and credible, though the channel’s own editorial perspective is evident. The title accurately captures the central theme of the video—the use of Chinese blueprints—and does not overpromise. Overall, the content is well-researched and the title matches the substance.

204 words

Title / Content Match

The title accurately reflects the core narrative: Inkling is built on Chinese open-source foundations, which is presented as surprising and geopolitically significant. The content delivers on this claim.

Quality & Reliability

8/10

The video relies on primary sources (Thinking Machines announcement) and reputable media (Financial Times, Reuters, Wired, Axios, TechCrunch), provides specific benchmark figures and technical details. However, it presents an editorialized analysis with a geopolitical angle and includes a promotional segment.

External References

Radar Profile

Reliability 8/10