The various iterations on GPT style architecture make me think that this is it for now, until some other breakthroughs there will only be optimisations but in its current form GenAI is not the sentient Singularity that some are praising.
The models, even with optimisation are very GPU/ power hungry so it is inevitable to compare to blockchain and how it basically got us nowhere - a solution looking for a problem.
All of the major labs are experimenting with incorporating classic reinforcement learning into their architectures. If it works things will become very, interesting, very quickly.
The various iterations on GPT style architecture make me think that this is it for now, until some other breakthroughs there will only be optimisations but in its current form GenAI is not the sentient Singularity that some are praising. The models, even with optimisation are very GPU/ power hungry so it is inevitable to compare to blockchain and how it basically got us nowhere - a solution looking for a problem.