Google Titans architecture, helping AI have long-term memory

I wrote about that a while ago: https://paxamans.github.io/blog/titans/

Are there any pretrained models with this architecture yet or is it all still completely theoretical beyond Google's unverifiable claims? They published the original Titans paper last year and nobody seems to have built on the idea.

https://github.com/lucidrains/titans-pytorch - is the only public iteration.

But no one appears to have taken the risk/time to properly validate it.

The fundamental ideas in the paper aren't particularly novel. They will probably work as advertised.