The practical issue the George isn't addressing is that we are (currently) reliant in data centers hosting, and local frontier-esque models are currently price prohibitive in terms of RAM.
These companies absolutely can not and will not take on the legal liability of serving models that give consumers instructions on cooking meth.
That being said, I agree with what he is saying, it's just that we need the memory supply chain to catch up so we can run ablated GLM models locally.
These companies absolutely can not and will not take on the legal liability of serving models that give consumers instructions on cooking meth.
That being said, I agree with what he is saying, it's just that we need the memory supply chain to catch up so we can run ablated GLM models locally.
My machine, my AI.