This is a good idea, though one problem is that Einsum notation (as realized in Numpy and Pytorch) doesn't support the notion of co-contravariance, and the site is based on their Einsum notation. I could potentially add the variances for the examples, though that would move away from how the site currently works (where the information about the reduction comes only from the einsum input).
o1 is an application of the Bitter Less. To quote Sutton: "The two methods that seem to scale arbitrarily in this way are search and learning." (emphasis mine -- in the original Sutton also emphasized learning).
OpenAI and others have previously pushed the learning side, while neglecting search. Now that gains from adding compute at training time have started to level off, they're adding compute at inference time.
In my experience NaCl is not a viable platform at this time.
My friend and I spent a couple weeks just trying to get the developer tools to work. We tried on Ubuntu, OS X, Windows, and Arch Linux. We weren't able to get started on a single platform. In the process we came across bugs that were several months old that haven't been fixed, even on their supported platforms.
Another thing to worry about is whether NaCl will take off. We can't know how committed Google is and how many users will install the plugin. It was supposed to be enabled by default in Chrome 6 (I believe) but it still isn't enabled in 7.
NaCl is still promising (especially Portable Native Client) but Google has to get it working first.
The link in the paper to their Java implementation is now broken: does anyone have a current link?