We are very early stage in the AI evolution.
LLMs provide us with an opportunity to glimpse into what the future will look like.
It's like seeing a blueprint.
We get a sense of what's possible.
Really quite impressed with Step-3.7 so far.
Seems like a solid model for its size.
The MoE architecture also allows it to decode rapidly, even on more modest amounts of spare compute.
This may become a new daily driver for me.
More to come soon.
@pmarca Yes, it's a crucial innovation to level the playing field.
I think given that LLMs are primarily language-driven, that law is a natural extension of their functionality.
They just need proper training and reinforcement with RAG.