The evergreen questions of tech are: is $CURRENT_TREND a bubble, and if so, when will it burst, what will its legacy be, and what comes next. I'm finding surp

The evergreen questions of tech are: is $CURRENT_TREND a bubble, and if so, when will it burst, what will its legacy be, and what comes next. I'm finding surprisingly little discussion of what trends might follow on AI (GPTs suggest several extant / prior fads: AR/VR, wearables, cryptocurrency/blockchain, quantum computing). OTOH, AI seems to have sprung from nowhere (at least so far as popular awareness is concerned) in 2023. (There was an early GPT-2 submission, though with little discussion, on 14 February 2019: <https://news.ycombinator.com/item?id=19166089 https://news.ycombinator.com/item?id=19166089>. The first mentions of large language models appear about 5 years ago. The first mention of large language models dates 14 years ago to 5 August 2010: "Large Language Models in Machine Translation (Google's way of doing it)" (upenn.edu) <https://aclanthology.org/D07-1090.pdf https://aclanthology.org/D07-1090.pdf> (PDF) <https://news.ycombinator.com/item?id=1578444 https://news.ycombinator.com/item?id=1578444>, but was entirely ignored at the time.) We're already seeing signs of public fatigue with AI hype (and perhaps overly-smart, overly-prying technology generally), e.g. a current HN post, "Study shows that tacking the “AI” label on products may drive people away" (CNN) <https://news.ycombinator.com/item?id=41231731 https://news.ycombinator.com/item?id=41231731>. An observation of mine in the past has been that technology tends to extend through a set of phases, roughly: - Proof of concept. Works for the creator. - Refinement: Incremental improvement of concept in function, reliability, efficiency, and utility. - Embellishment: Layering complexity on top of an initial design. Though initially well intentioned, the end result is fragility, impediments to usability, and eventually, rot. A baroque era. - Recaptitulation or replacement: An overly complex design may find itself refined back to a more functional one (e.g.: the car tailfin era recaptitulated into today's econo-box and performance sedan market, though SUVs perpetuate the meme of ostentatious nonfunctional embellishment). Or it may find itself replaced entirely by a more appropriate technology, as may be happening now with PCs being supplanted by tablets and mobile devices offering much of the functionality required by users. Strikes against LLM ML AI GPTs are that their data, compute, and energy requirements are huge, and corporate entities show alarming tendency toward monopolistic dominance. The possibility of delivering equivalent or at least sufficient results with far fewer resources and competitive risk might drive a countertrend. There's also the accelerating trend of unintended data disclosures, hacking attacks, and disabling entire industrial sectors (ChangeHealthcare (2024), shipping with Petya/Maersk (2017), Colonial Pipeline ransomware attack (2021), WannaCry (UK NHS, FedEx, Telefonica, 2017), Equifax (2017), Target (2013), etc., etc.) which may generate push-back.