Hey fam what’s up. As a researcher, here’s how I’d go about looking into this.

First, go to scholar.google.com and type in what you’re interested in: in this case “neuro-symbolic AI.” A bunch of papers show up with titles like “Neuro-symbolic artificial intelligence: a survey” and “Neuro-symbolic approaches in artificial intelligence”. Some of them are a couple years old but don’t be worried about this.

Start looking the papers over. Don’t be freaked out if they look complicated, at this stage we’re not going to read them end-to-end. Basically you want to start learning things like: (1) key concepts and terminology, (2) active research areas, (3) people/labs investigating this, (4) journals/conferences that focus on this, (5) commonly used software packages and toolkits, or implementations associated with a paper/result. If you’re serious about this, you will start taking notes.

Here’s how to read a research paper: look very closely at the Abstract, Introduction, and Conclusions. Skim the rest to see what’s there (and whether it’s worth reading closely later.) Do this to a bunch of research papers. Take notes about numbers 1-5 above. Number 1 and 2 will help you understand the field, number 3 and 4 will help you find the very latest research (look up the web pages of people/labs/conferences), and number 5 will give you the tools to reproduce their work and extend it. If you don’t understand the presupposed knowledge behind concepts and terminology, look it up on the web or in a textbook - this may take a while at first, but it’s an investment. Look for people to talk to about this, either online or in person.

As you do this reading, look for questions, directions, or implementations that particularly interest you. Then start looking for more papers focused on that. i.e. do more google scholar searches, or look up what researchers or conferences are dealing with that topic of interest. Start playing around with software packages cited, and implementations associated with research. If you keep pushing on this, you will start asking questions that no-one has worked on yet - at this point it’s a great idea to write things down and upload them to arxiv, ideally partnering with someone already in the field.

If you do the above, you will be doing research in the same way as an advanced Master’s student or beginning PhD student. Good luck. Please don’t become just another corporate drone, but wtf it’s up to you.

Haven’t heard that term before, but one of the things that’s been obvious to me to experiment with – not that I’ve actually gotten around to it yet – is to have an LLM try using Prolog, Z3, and/or SQL as tools to overcome some of its weak points. That’s naively what I’d expect “neuro-symbolic AI” techniques to be (assuming you’re looking at current topics rather than stuff from ~20 years ago), but again, shot in the dark here.

I usually start with the Wikipedia Article when I’m interested in new things. It’ll have many references at the bottom to read more about a concept.

Interestingly enough, there’s zero mention of Claude in there. And when I google it, there’s many very convoluted blog posts. And I can’t tell whether it’s above my head or hallucinated stories. They go on for like 20 pages but don’t really explain anything with all those words. Or what they actually found in Claude’s code.

Symbolic-AI in itself isn’t too hard. That’s stuff from the 1980s and in every computer science textbook. Just no clue how something like an expert system is supposed to be connected to a Chatbot or programming agent.

Don’t know either. So far I’ve just archived ‘symbolic’ NN’s, as those that doesn’t use text, or normal modalities as the unit, so not sure what they work on, really.