> 40 months

Not counting from 1971s DARPA? Sorry I'm allegric when LLMs being called AI like nothing existed before it.

Could the "LLM" of 1971 DARPA produce working code that it translated from a legacy codebase to Java and this within a short timeframe? ;-)

Doesn’t it all look like child’s play though?

You would need to go back to McCullough, Pitts and von Neumann in the 1940s if you wanted to talk about where it really got started.

The 1970s were a long, dark AI winter, thanks to FUD spread by Minsky and Papert. A lot of recent work could have been done back then despite the lack of good hardware, as seen in the other HN story where the guy trained a transformer on a PDP-11. But the whole field was radioactive after their book came out.