You don't bet in achieving a pipe dream in the near future when developing something (at least I hope you don't). The actual AI devs are also rather humble about what can be realistically achieved in the near future if they're allowed to speak openly, it's the techbros and investors who're blowing everything up so phenomenally that the only way for this bubble - which by now is big enough to crash the US economy on its own - to proceed is to achieve the literally impossible, which is General Artificial Intelligence within at least the next 2 to 3 years (they can only pump so much money into it). I haven't heard a single developer or scientist saying this to even be remotely realistic.

This is more about economics than programming, AI experts and grifters just collect the money as long as the bubble persist. Also has to do with tech-solutionism and latestage capitalism. In the end the science of Machine Learning will prevail (although a lot of devs might need to find other specialisations; perhaps COBOL?), but this economical house of cards will fall apart. Let's hope it will burry US-neolibertarianism and -fascism under it this time.

Usually people know better than blowing up the hype for one iteration this much and act like it's the second coming. There will be diminishing returns but instead a stairway to heaven is promised.

It starts to feel like one last attempt to hang on to the old system, and if this fails the house of cards crumbles and China takes the lead in tech.

For most software, iteration starts getting diminishing returns only if it's approaching feature completeness and no bugs. LLMs are plateauing well before they became super genius job stealers like they were supposed to, and it's going to take a major breakthrough to see any significant improvement.