Funny. There is only one necessary code change and it’s to backpropagation, the

Funny.
There is only one necessary code change and it’s to backpropagation, the rest is just training. I clearly failed to make the point that the LLMs are capable of reasoning by their existing configurations – and that present attempts to shove questions of marginal indifference into a frame of cardinal inequality.
I’m not worried about code. There is so little of it in LLMs in the first place. I’m worried about how those working on them do not understand foundations necessary to produce reasoning outside of internal-closure grammars. (math, programming)