Advanced Machine Intelligence

Advanced Machine Intelligence

AMI is researching JEPA-based world models so AI can learn abstract representations of the world rather than rely only on text statistics.

Visit website

What is Advanced Machine Intelligence?

AMI, short for Advanced Machine Intelligence, is not a chat app you can open today. It is a research company built around world models, with Yann LeCun as executive chairman. Its chosen route is JEPA: instead of learning mainly from word statistics and predicting the next token, the model is meant to learn abstract representations of how the world changes. Think of an AI watching a scene or a stream of observations, picking up objects, positions, and relationships, then forming an idea of what might happen next. That work matters for robotics, video understanding, and systems that need to make decisions in real environments.

What people can follow now is the research itself, especially papers and the code the company says it will open-source, not a packaged product. For developers, the useful question is whether JEPA can become reproducible, editable training and evaluation tools: reading the code, running experiments, and seeing which tasks produce useful representations. It still takes suitable data, compute, and engineering setup; this is not a drop-in replacement for a general chatbot.

When comparing language-model products with world-model research, FindAIGo helps keep both kinds of entries in view so it is clearer what can be tried today and what is still taking shape in the lab. The papers and open code are the part to watch. The real test is not the slogan about replacing LLMs, but whether the work helps AI understand more than text.

加载中 / Loading

Discussion

Loading comments…