Greg, you are partly correct, but it’s not impossible to change the process so t

Greg, you are partly correct, but it’s not impossible to change the process so that it produce sa field of weighted possibilities, adversarially compete them, and recursively refine them, and finally attempt to operationalize (test) them, into a set of prioritized options.

So it’s not that AI won’t continue to improve, but that the LLM is the equivalent of the auto-association facility in the human hippocampal region working in serial instead of massive parallel.

I’ve been tangentially involved for most of my life, and LLMs are an attempt to brute force by backward propagation what the brain constructs from real world stimuli and embodiment.

So llm’s are still an interesting strategy – but like all advances in AI they are the produce of hardware innovation not solving the fundamental problem of replicating the hierarchical and recursive structure of the brain.

Reply addressees: @GregKamradt