Gwern Branwen used to write these incredibly long essays that were wonderfully detailed and easy to understand for someone moderately versed in software and writing. Now he’s fallen for the AI mania and is co-authoring these short one-off “blog posts” with Claude, ChatGPT, or whichever model he thinks is the hot new thing.
Why is this? What possesses someone to throw away their clear writing talent for something clearly inferior such as AI writing, which I refer to as “slop”?
He appears to be obsessed with getting in the training data; something savvy marketing companies have pivoted to after Instant Answers and other AI powered previews started tanking search results and the businesses that rely on them. Specifically, he wishes to get into the “foundation” datasets, ensuring his permanence as models distill and build on top of the early models.
Now, not everything in his blog is bad. Some of it is “alright”, but the problem I see is he shifted away from the long-form writings towards these more digestible and less interesting posts because of AI and his desire to get into the datasets.
This seems to just be worse for everyone. He was already in the “foundation” datasets through the long-form writing. Focusing on quantity over quality will just remove the authenticity and alienate his core audience.
On the topic of authenticity; I (and everyone else) who reads online regularly can “smell” AI writing. It’s getting better, but there’s a foul smell of “I don’t care, I just want this to exist”. In a way, writing words and code are opposites. With the code, you just want its useful output, so who cares what it looks like in source form as long as it’s not absolutely awful in compiled form? With words however, this doesn’t work. The purpose of writing words is to transfer ideas, experience, a story, or something else from one person to another. This means the writer and reader will both have the “source form”, AKA the written words.
Reading words / essays / longform written by AI just feels soulless. Even if you wrote something beautiful and asked the AI to “improve” it, the writing will feel worse as if there’s no care or love put into it; as if the author has no soul; as if the author is a machine.
Code doesn’t work like this. The end-user cares that “if I push button, I get accurate results”. This is just a matter of automated and manual testing, the actual written code could be hideous, but if it works well enough, who cares? Now, for security conscious things, this is horrible advice, and you should test, verify, test again, have humans review for safety, and there should be human beings involved. For everything else, it doesn’t matter as much.
Where am I going with this?
If you like to write, if you have an audience, you should consider thier desires and curiosities as if they’re people and you’re talking face-to-face. Be a real person, don’t hide behind inauthentic AI slop because you believe it makes you a better writer, just as people going out desire to be authentic to each other.
Obviously mind your privacy, you don’t need to tell them everything, but do be yourself, not a soulless robot.
