Technical

Srinivasa Reddy Kandi: Tackling AI Randomness with Deterministic Models at Thinking Machines Lab

September, 11, 2025-05:02

Share: Facebook | Twitter | Whatsapp | Linkedin | Visits: 37669 | :2821

---

Tackling AI Randomness with Deterministic Models at Thinking Machines Lab:

Thinking Machines Lab has been drawing intense attention since its launch, backed by $2 billion in seed funding and staffed with a team of former OpenAI researchers. This week, the lab offered its first real look at what it’s working on: AI models capable of delivering reproducible responses.

In a blog post titled “Defeating Nondeterminism in LLM Inference,” researcher Horace He explores why today’s AI systems often give different answers to the same question. The culprit, he argues, lies in the way GPU kernels—the small programs running inside Nvidia chips—are orchestrated during inference. By exerting tighter control over this process, He believes AI models can be made far more deterministic and reliable.

Such progress wouldn’t just benefit enterprises and researchers looking for consistency; it could also enhance reinforcement learning (RL). RL trains models by rewarding correct outputs, but noisy, inconsistent responses muddy the training process. If models generate steadier answers, RL could become significantly more effective—something Thinking Machines Lab reportedly plans to use in building customized AI models for businesses.

OpenAI’s former CTO, has promised the lab’s first product launch in the coming months, aimed at researchers and startups developing custom models. It remains to be seen if this determinism-focused work will underpin that release.

The blog post is also the first entry in a new series, “Connectionism,” which the company says will regularly share research, code, and insights with the public. Unlike OpenAI—which has grown more secretive over time—Thinking Machines Lab has pledged to maintain an open research culture.

For now, the post offers a rare glimpse inside one of Silicon Valley’s most secretive AI startups. The big question: can Thinking Machines Lab turn these ideas into breakthroughs that justify its $12 billion valuation?

Author: Kandi Srinivasa Reddy, Srinivasa Reddy Kandi, #KandiSrinivasaReddy, #SrinivasaReddyKandi