AI makes dumb, smart and smart, dumb
There’s a lot to be excited and concerned about with the ol’ transformer-based LLMs, but here’s another one: AI makes dumb things look smart and smart things look dumb.
The “dumb things look smart” may be more self-explanatory. It’s now possible to generate entire essays of text with a prompt or two. Windows Notepad has a feature that lets you highlight text to “make it longer”. These transformers make content that is well structured and grammatically correct; it lacks originality and strives to maintain a neutral tone, but for all its flaws, at a superficial glance this content looks “smart”. In the past, writing like this may still have been flawed (harken the cliché “if I had more time, I would have written a shorter letter”), but it still required real human effort.
For “making smart things look dumb”, soon everyone will develop mental heuristics to identify what is “AI generated”. See a long-winded, neutral-sounding, wall-of-text; it must be AI (https://noslopgrenade.com/). But you know what else could look like that? A free-range, organically written, hand-crafted wall-of-text. With experience and a critical eye, I expect folks will be able to discern the difference between the laziest low-effort slop and something hand-written; but that still will takes skill and effort and unfortunately, first lazily-read impressions count, and so the old heuristics to identify “might be smart writing”, as flawed as they were, may soon be supplanted by an even more unreliable heuristic of “if it’s long and structured, it’s AI slop”.