The article is a very good review of Symbolic AI, in general, not just Cyc.
I have spent a lot of time with OpenCyc in the past, but haven’t touched it in ten years.
I believe tif there is a productive future for symbolic AI that it will involve using LLMs to construct knowledge graphs, symbolic relatikns, etc. from unstructured data.
>> The article is a very good review of Symbolic AI, in general, not just Cyc.
As a "review of symbolic AI" I found it uninformed and superficial and felt that it rehashed the same old points about how symbolic AI "failed", which are disputed by the facts; specifically the fact that major symbolic AI fields like SAT solving, automated theorem proving and planning and scheduling are still going strong and have produced real-world results, so much so that e.g. SAT solving, Planning, program verification, and automated theorem proving aren't even considered "AI" anymore because they now actually work, and work very well indeed.
Technically, you are probably correct. I did find it a good walk ‘down memory lane.’ I have been working in the field since 1982 and the article made me nostalgic.
With LLM itself using those graphs to validate its answers, etc. And at some point it will be internalized into architecture as a graph attention layer.