The machines are fine. I'm worried about us
15 points by edsu
15 points by edsu
I think this essay was written by AI. There's a Dune quote:
Frank Herbert (yeah, I know I'm a nerd), in God Emperor of Dune, has a character observe: "What do such machines really do? They increase the number of things we can do without thinking. Things we do without thinking; there's the real danger." Herbert was writing science fiction. I'm writing about my office. The distance between those two things has gotten uncomfortably small.
The original text in the book uses an em-dash, but this quote uses a semicolon instead. A plausible explanation is that at least some of the text in this article was generated by an LLM, which being trained on quality writing, used a few em-dashes, that were then replaced by semicolons to make the text seem "less AI".
It is deeply ironic and disturbing for an AI to be writing "the machines are fine, I'm worried about us [humans]" type blog posts. Like fuck you skinwalker, you are not one of us.
I am in agreement with the essence of the argument: productive work requires that some person understand the system; using LLMs to replace understanding the system defeats the ability to supervise them.
I will just note that in my judgment, this is almost certainly LLM-written. This is based both on a subjective reading and on a check with Pangram (which is a remarkably good, conservative LLM text detector). I'm not saying you need to hate it, but maybe calibrate the effort you're spending on reading and arguing with the article with the effort the author probably did not put into writing it.
More generally, I'm finding that 50%+ of pro- and anti-AI think pieces in my feed appear to be LLM-written. People obviously have strong opinions about the topic, and it's easy to get them to upvote low-effort articles that say the things they want to hear.
re: substance, i think i agree broadly with the thrust.
re: the provenance of this article, i must admit that i didn’t detect any LLM tells on my first, not so close reading of it. but then, i haven’t really focused on my distinguishing/discriminating abilities all that much. the commentary on bluesky is largely positive, although i see one commenter say it read like Claude’s writing.
re: the overlap of substance and provenance, is it possible this author wrote some parts of the text themselves and delegated others? i’m curious if their (potential) usage of LLMs in writing this article is something they see as separate from their usage of LLMs in doing physics work, likely reducing their qualms about relying very heavily on Claude, or connected to they way they use it in their physics work—is it possible this article was heavily edited with Claude, rather than generated out of whole cloth? i am tempted to reach out to the author and ask about their writing process and thoughts.
I don't mind when people use LLMs as editors. But when I encounter a suspiciously-looking article, I ask myself what's the minimum amount of text you'd need to enter to have an an LLM produce a functionally identical post.
A lot of LLM-generated blog posts can be boiled down to really short prompts ("generate an article about the importance of Immanuel Kant in the age of AI", etc). It's not a proof of AI slop, but if I see a combination of LLM output + shallow thesis + no prior human-like content from the author... you know, life's too short to engage?
And remember that the article talks about the subtle, negative consequences of relying on LLMs to do the grunt work, so even if it's just heavily AI-edited, what's that supposed to mean?
The trajectory, from the outside, was identical.
Ahh, the 'ol Chinese cubicle thought experiment.