Essentially the idea of a context window in modern LLM models, there is implicit domain knowledge to every task in which no matter how capable the model may be, if not in the context, the software will not be functional.

I think this is the cause for the division in the perception of how useful AI is.

If you work in a field with mostly proprietary implementations of solutions, the top model aren't going to be all that helpful. The models won't have the domain knowledge, because open source code doesn't exist for most domains, because there's very real competitive advantage in keepings code/processes, that aren't trivially implemented, a secret!

I think proprietary data is the new moat, because that's where the vast majority of useful domain knowledge exists.

Incredibly prescient given what’s happening now 25 years later. This message resonates with me quite strongly! Thanks for sharing.

There's quite some more of the kind, and of about that time.. 80ies-90-ies.

Like the 3 levels of audience (known for milennia as shu-ha-ri):

https://wiki.c2.com/?ThreeLevelsOfAudience

and 4 levels of knowledge:

https://wiki.c2.com/?FourLevelsOfCompetence

or this from Dijkstra, 1989:

Surely, our computers would unscramble all the secret code of all our enemies and guide our missiles with unfailing precision right on their targets. Robots would take over the tedium of production, guaranteeing a positive balance of trade for all nations. Office automation would multiply the productivity of the white-collar worker, information systems would enable management to avoid waste and to make the right strategic decisions, and finally, the giant brains would not only relieve us from the tedium but also from the obligation to think about hard problems and from the painful responsibility to take difficult decisions. In short: computers were tolerated because they promised a well protected and prosperous paradise for the lazy, the incompetent, and the cowardly.

https://www.cs.utexas.edu/~EWD/transcriptions/EWD10xx/EWD104...