Thinking Machines

Thinking Machines Lab is an artificial intelligence research and development company founded by a team with proven experience in some of the most influential AI projects of recent years, including ChatGPT, Character.ai, the Mistral language model and open-source projects such as PyTorch, Fairseq and Segment Anything.

The company focuses its activity on three main areas. The first is frontier model research, with particular attention to domains such as science and programming. The second is the development of multimodal systems oriented towards human-AI collaboration, rather than focusing exclusively on full model autonomy. The third is the open publication of results: technical articles, code and research posts under the name Connectionism, with the aim of broadening collective understanding of how current AI systems work.

Key technical lines of work include on-policy distillation, fine-tuning via LoRA, deterministic inference in large-scale language models and the exploration of modular architectures. These topics are addressed both from the perspective of fundamental research and from their application in concrete products.

Thinking Machines Lab also works on personalisation mechanisms that allow models to be adapted to specific needs, contexts and values, with the aim of expanding the range of applications beyond the usual use cases in programming and mathematics.

On the safety front, the company publishes recipes and best practices for building safe AI systems, conducts red-teaming exercises and post-deployment monitoring, and shares datasets and model specifications to support external alignment research.