Definitely interesting approach. I think neurosymbolic approaches are what we ultimately need so I'm probably biased. At least it means using LLMs for what they're good at (language skills) and only that. Then rely on proper code symbolic models which do the reasoning heavy lifting. I'd expect it can give nice output with smaller models.

Wondering how to implement your own inference engine? Here is a nice simple blueprint to get started.

This is definitely an interesting declarative language. Looking forward to more such neurosymbolic approaches.

Finally a path forward for logic programming? An opportunity to evolve beyond Prolog and its variants? Good food for thought.

Nice resource to get started with Prolog.