18 points zerop 3 hours ago 1 comments
thisisauserid 48 minutes ago | parent
I totally get why people want to reinvent LLMs so they're better/cheaper/faster and this hybrid approach with State Space Models is likely going to bear fruit.
But I'm WAY more interested in what's going to happen with Mamba-3, et al. when it's relatively easy to predict, ya know, state.