Engineering360: "Is AI actually coming for STEM careers?"

This essay reads like something I might have written myself—perhaps with some chatbot assistance. I’m retired from a software career but have stayed closely engaged with developments in STEM, and with AI in particular. Over the past few years, I’ve watched chatbot behavior evolve in ways that are unmistakable. Early on, even basic or “softball” questions could yield bizarre or unreliable responses. Today, queries about fairly niche technical topics are often answered accurately and coherently—something that simply wasn’t true two years ago. That is real and meaningful progress.

What has not changed, however, is the apparent inability of these systems to plan, anticipate, or offer genuine insight. When I ask for help with a truly unique project—one that is not already well represented in blogs, papers, or online discussions—the responses tend to flatten out. They read like the output of someone with limited creative experience: competent within familiar patterns, but unable to reason forward into uncharted territory. In that regard, the answers resemble those of an engineer constrained by outdated assumptions rather than one actively grappling with a new problem.

What gives me pause is not merely that this limitation exists, but that it appears largely unchanged despite intense effort. The last several years have brought dramatic improvements in knowledge synthesis, recall, and presentation. If LLMs were naturally trending toward genuine creativity or future-looking insight, one might expect at least some parallel progress there. Instead, the gap remains.

That raises an uncomfortable possibility: perhaps large language models are fundamentally incapable of the kind of anticipatory reasoning and creative adaptability that many hope for—and fear. Humans are notoriously susceptible to extrapolating long-term trajectories from short-term trends. It may be that recent gains reflect optimization within a narrow paradigm rather than movement toward general intelligence. If so, LLMs may ultimately fall well short of the transformative promise projected onto them—unable to do what humanity both hopes they can achieve, and worries they might.

(Full disclosure - this was written with the help of ChatGPT 5.2. The thoughts are mine, the wording is "AI")