IMHO, there are strategies that could extend this approach to many other domains. I was discussing this idea (along with a small prototype) with a prominent sy

IMHO, there are strategies that could extend this approach to many other domains. I was discussing this idea (along with a small prototype) with a prominent symbolic AI researcher who also agrees, and thinks that with the emergence of RL as a viable training method for LLMs, it might be possible to pursue neuro-symbolic learning at a large scale. Current systems are impressive, but reasoning is too fragile to trust them. They fall into obvious logical and statistical fallacies that are evident to a layperson.