Andrew Ng and Yann LeCun: US Is Losing AI Race Due to Closed Models

6 months ago (haebom.dev)

It's incredible China is opening their weights and half our companies aren't. It's a complete inversion of how things have worked for so long. Very disappointing.

Open-source AI models are a big deal. In robotics and CV open models like Segment Anything are what made object segmentation way more accessible. This is a strong force pushing Physical AI forward!

  • Which models release the full training dataset, the training code, the inference code, and any random seeds used? Without all of those, you can't "compile" the project. If you can't compile it yourself, then it's not open source.

    Compiling the training or inference code does not give you a model.

Is it really a race? Western Ai companies are releasing some dumbed down open weights too. Maybe the closed source chinese models are so much better, they are kept under wraps by the chinese military.

In any scientific or technological domain, the mentality of us vs. them is poisonous. Humanity is in it together.

Whether a model is open or closed is beside the point. How much value are they generating?