neurosymbolic AI sounds like deep learning powered prolog

[Jake Waas]

I only caught Manas' talk today, I have to go back and listen to Luis'.

If I understood the use of structured knowledge with LLMs and the neurosymbolic rubric, structured knowledge (viz ontologies) appear in 

- user queries being externally structured and formally checked

- The LLM's text-generation is used to choose external structured objects, rather than directly text-generating an answer

- The query response objects are coerced into a formal structure and formally checked.

 This made me think of prolog except the search algorithm has been replaced by an llm's chain-of-thought reasoning strategy.

I imagine I got lots of that wrong, I am mostly not involved with LLMs. Clarification would be very welcome.

Jake 'screwlisp'