Coming back to this 5+ years later, this post seems to have been pretty prescient. Now we live in a paradigm where contemporary AI systems have at their core a massively optimized predictive core with a comparatively small amount of RL slapped on top.
Coming back to this 5+ years later, this post seems to have been pretty prescient. Now we live in a paradigm where contemporary AI systems have at their core a massively optimized predictive core with a comparatively small amount of RL slapped on top.