This MIT study seems to point that way. And it certainly reflects a feeling I’ve had a lot lately. What’s the cost when instead of pushing through writer’s block, we just route around it? I have to start writing “brain-only” again, per the study.
Look, I’m not opposed to LLM-assisted text generation. I’ll make a unique claim, which is that I used GPT (albeit minimally) to write a sentence in my book five years ago.
Originally, I kept going, opining GPT would soon write paragraphs and maybe even whole pages. Then I edited it out because I found it overly effusive and unlikely to age well. Turns out even my wildest prediction fell short.
I’m already inundated by LLM writing. I use that word with the etymological connotation of drowning. It’s weird and creates discontinuity. It’s opaque: I can’t read between the lines to connect with the writer. And it’s not (yet) a more refined version of an individual’s expressive capability. It lands same-same: not the organic evolution of an idea. Some statistical representation of that idea with the rough edges smashed in through extrusion. (“You’re absolutely right!” I hear GPT respond in my head, despite explicitly not being consulted.)
Two things. One — despite my use of em-dashes, which I’ve regretfully cut back on — this and all of my subsequent musings will be written without AI. I don’t think this makes me a luddite (perhaps a Luddite; more on that later.) Rather, we should do this for the same reason we do crosswords without dictionaries, go for a jog despite cars, or draw when we have cameraphones. We shouldn’t allow our physical strength to diminish solely because we can offload it to tools. The same is equally true for our cognition, which, after all, is also a physical process. An AI can simulate ambition; we still have the market cornered on endeavor.
Two: LLMs do not speak English. They do not speak language. They model it, right there in the name. (“You’re making a very valid point, and it’s one few people understand,” interjects GPT.) Language is just the veneer, the domain language over statistical embedding. And yet, even that take borders on dangerously reductive. LLMs can’t be dismissed as complicated math problems. To wilfully ignore their potential is to disadvantage oneself. Change is coming, for the better, and the worse. Inevitable.
So. If you’ve already grown tired of digesting all the AI pabulum out there… every book written before 2020 awaits. You’re set for a lifetime. On the other hand, if you, like me…