> you've failed to recognise the limitations as the article points out Fixed weights don't preclude in-context learning and out-of-the-loop weight updates. And

> you've failed to recognise the limitations as the article points out Fixed weights don't preclude in-context learning and out-of-the-loop weight updates. And that's the only principled limitation mentioned in the post.