surprised i haven’t heard of more people using this approach to AI: symbolic backbone + neural nets called as functions for perceptual identification: (theory starts around the 20 minute mark)

Cool-looking framework. Many big AI systems are already made that way (e.g. AlphaGo has a tree search backbone that uses ML to perceive promising moves), but I doubt many have such a concise DSL. Is training integrated into the language somehow? I didn't catch that while skimming.

i also only skimmed but i will get back to you when i revisit