Yup - well this is where my suggestion is to change the GPT architecture. You can think of having a logic program function as the 'frontal lobe' of the general

Yup - well this is where my suggestion is to change the GPT architecture. You can think of having a logic program function as the 'frontal lobe' of the general pre-trained auto-associative model. This 'frontal lobe' region would try to come up with logical sequences to go along with it's internal auto-associative representations. Of course - the logic programming piece is just one approach - maybe chain of though or chain of reason prompting could work here too as many humans I think use this chain-of-reasoning approach themselves. Logic programming to me would function as a suggested shortcut.