Proven Research About AI Behaviors Are Rapidly Aging And Becoming Woefully Obsolete

In today’s column, I examine the unnerving realization that many of the ways we have been told to best prompt or use generative AI and large language models (LLMs) are no longer applicable. Why so? Because the research insights that identified those handy rules of thumb are rapidly becoming obsolete. AI is advancing so quickly that research analysis from a year ago is probably already out of date (and even research from a few months ago can be out of touch).

This is problematic for users and for researchers. For researchers, the problem is that each new AI model tends to make changes that end up rectifying prior shortcomings. Thus, an AI behavior that used to be a bone of contention no longer exists. Users who continue to use helpful tips or techniques to steer or coax AI are unknowingly relying on outdated insights. In brief, what we think we know about AI behavior has a surprisingly short shelf life, and prevailing approaches to dealing with AI no longer work. The added twist is that actions you take based on obsolete insights can be counterproductive nowadays and do more harm than good.

Let’s talk about it. This analysis of AI breakthroughs is part of my ongoing Forbes column coverage of the latest in AI, including identifying and explaining key AI complexities (see the link here).

AI And Human Psychology

As a quick background, I’ve extensively covered and analyzed many facets of the advent of modern-era AI, including its varied behaviors and its impact on human psychology, well-being, and mental health. For a handy list of over 120 of my analyses on this significant topic, see the link here and the link here. The growing use of AI has largely been spurred by rapid advances in generative AI and large language models (LLMs), including the widespread use of ChatGPT, GPT-5, Claude, Grok, Gemini, Copilot, and others.

When Tech Advances

Old Rules That Are Outdated

Classifying AI Approaches

I classify AI approaches into these three categories: