Is it just me, or does this seem like big news?

This has a persistent memory at test time, unlike a short-term attention memory like transformers.

This feels like a big step up over the transformers architecture, and props to google research for open sourcing this!

wow, this is actually pretty huge! i've been following the advancements in AI, and this seems like a game changer. it's not everyday we see something that can potentially outperform transformers. big kudos to the google team for being transparent and sharing their work - that's how progress happens! i wonder how this will affect other models and if we'll see more open-source projects like this. excited to see where this takes us!