For me, at a macro (societal) level, I'm kind of a mild "doomer". My worry is triggered by way our world is becoming "software-defined" at the same time AI hype is driving lots of people into insane delusions of productivity and quality. I do not look forward to a "vibe-defined" technological environment in the future. I really hope this hype cycle comes crashing down before it damages our shared world, having displaced properly engineered systems that had disciplined human oversight. The way AI products (not just LLM) are being hyped disturbs me. For example, I do not like the trend to alpha/beta test safety critical car control systems on public roads. I do not like the way the public is embracing these things and blithely trusting them, and rapdily lowering their vigilance. To me, this seems like a latent disaster, where global socio-technological systems may collapse once this feedback loop proceeds far enough.

At a micro (personal career) level, I just recoil at the whole LLM paradigm. I am seriously looking at my finances and contemplating a hard "no" to retire. I know myself pretty well. I have spent a career becoming a very senior "individual contributor" in an academic R&D environment. I am not a manager, and would not want a managerial job. I do not enjoy nor am I good at delegating work to lower-skilled juniors. I have had excellent collaborations with true peers over the years. I reject the mindset that an LLM can be such a peer. It lacks the hysteresis, learning capability, and actual agency to form a trustworthy working relationship. I refuse to spend my energy on pretending it is more than it is.

Part of my issue is that, despite doing R&D prototyping, I am a very top-down thinker. I do a lot of ruminating through all the levels of a system before I build it. I know I depend on this to get aspects of quality out of anything I build. I accept that I am not a good code reviewer, e.g. to take random code and find flaws or rework it. I need to trust my partner not to be gas lighting me with fiction nor sneaky flaws. When in doubt, I will "nuke it from orbit" and start over in a green field manner with top-down designs where I can gauge the risk.

I would much rather write from scratch than take someone's janky code and rework it. I would never take such code and then stake my own reputation on it being functional. Doing so with a stateless generator is nightmare fuel. I would rather walk away become a janitor or auditor who has to monitor the output of the LLM crack monkey...