35 upvotes
Paper argues AGI scaling hinges on System-1/System-2 co-design
What is Does Scaling Toward AGI Depend on What System-1 Lets System-2 Execute??
The authors examine whether the current strategy of making ever larger, compute-hungry ‘System-2’ models is the only path to artificial general intelligence. They suggest that pairing a lightweight, fast ‘System-1’ with the reasoning engine could achieve more progress with less compute.
Who it is for
Researchers and engineers focused on efficient AI should read it to rethink scaling strategies and explore hybrid architectures.