Ask HN: Why are some innovations from previous AI cycles forgotten about?
Take for example Expert Systems (https://en.wikipedia.org/wiki/Expert_system) which are AI systems designed to make logical inferences. They had systems that could do accurate logical inferences in the 1970s, and even designed a VAX chip (SID) and made medical diagnosises (MYCIN) [apparently] to a level at or above many physicians.
Why are innovations like these (another example being genetic algorithms) ignored and forgotten about even though they work better in logical problems which LLMs suck at? Expert systems don't work better than LLMs - they are very brittle and completely fall apart when they encounter a novel situation outside the scenarios programmed into their if-then rules.