Thinking Machines Lab

And obviously we are seeing that some of those patterns play out and actually retaining that sense of agency is so important for me.

There's a lovely quote from Miro Murati, who used to be the CTO of OpenAI, the celebration company, Thinking Machines.

talking about how important it is that actually all these kind of prophecies of what is about to happen in the next three, five, ten years, they may or may not be right, but really importantly, they seem to remove our sense of agency that we have some control over this.

We have some control over our own lives.

The same is true of many other AI startups too.

VC investors recently determined that Thinking Machines was four times more valuable than Snap, despite generating a 60th of its revenue.

They priced another AI company at $32 billion, despite not having a product.

Private markets investors have been guzzling AI Kool-Aid through a beer bong without even knowing what's in it.

Australian co-founder Elise Jenkins and her startup, Coherence Neuro, raised US$45 million from Peter Thiel's Founders Fund and the OpenAI Startup Fund for their work developing brain implants to track and treat cancer.

Firmus bounces back with new Thinking Machines contract.

Following their failed IPO, the Australian data centre firm has inked a multi-year agreement with AI Lab Thinking Machines, which will train and deploy models using Firmus data centres in Asia.

Mervac house sales fall by almost a third.

The real estate giant said its new home sales fell 31% in the September quarter as the housing market downturn deepens.

So one of the biggest questions is, like, why? Why does China, have they been coming with this open source or open weight approach, um, and the US just really hasn't? Well, first, uh, you know, some recent models from the US have done fairly okay, um, on the open source scheme.

So, uh, Thinking Machines Lab's new Inkling model, really good.

Um, you know, NVIDIA's, uh, open model's fairly good.

But no one's been able to touch the, uh, Chinese models.

The spec ruled out sycophancy and optimizing on short-term user feedback produced it anyway.

Stress testing model specs reveals character differences among language models, anthropic and thinking machines, October 2025.

Scenarios that force trade-offs between spec principles reveal distinct characters across frontier models.

System cards, anthropic.

The method behind many model organisms, including the alignment faking and reward hacking setups.

RL.

Tinker, Thinking Machines.

A training API for SFT, DPO and RL, GRPO, PPO on open weight models via LoRa with an open cookbook of recipes.