Taste begins where measurement ends.
Taste Is All That's Left
581–587 of 587 posts
Re: Taste Is All That's Left
#582But this is basically reinventing the term "Aesthetics" which is a field that had several centuries of philosophy research about it. Not all of them will agree AI cannot have it. Actually I think it is possible to reconcile Kant's third critique with the statement "LLM is possible to make aesthetic judgement" (with certain generous application of principle of charity, naturally).
AI does have something like Aesthetics and taste, but it's the average of all the inputs and so it's almost by definition the answer to 'what is the most average design possible based on all human prior data inputs?' That's why AI prototypes of front end UIs tend to look the same without massaging them with your own design tokens, design language, and injecting your own taste.
read a book.
Re: Taste Is All That's Left
#583Earlier quoted context omitted.
Running Llama 3 between your ears? Don't you and all the people in this thread claiming AI slop know that AI was trained on human writing ?
>Don't you and all the people in this thread claiming AI slop know that AI was trained on human writing? Do you think this implies that it is impossible to distinguish LLM-generated text from human text? I don't think that follows at all. >Running Llama 3 between your ears? I'm always happy for a chance to improve my classifier! Can you provide some long-form writing from before 2024 that uses this style?
No, I dont think there is any reliable data to be able reliably classify AI text. So far what I've seen is humans actually fail this task reliably
Re: Taste Is All That's Left
#584Earlier quoted context omitted.
I'm sceptical, where are all these essays he was supposed to have learned this very distinct style from? I certainly don't remember seeing them.
Yeah if you look back at earlier posts on the same blog, definitely not the same style. I'm guessing the author is either lying to save face or has spent so much time with Claude that they cannot distinguish Claude's voice from their own (we're all vulnerable to this).
it's more horrifying to think that an entire generation's "taste" might be built on the constant output of LLMs
it's like a new generation of chefs that only knows of Chili's, and thinks it's good.
shudder
Re: Taste Is All That's Left
#585Earlier quoted context omitted.
Are you serious? Pangram is notorios for false positives!
All ML classifiers (and algorithms) have a non-zero false positive rate. Having an error rate is baked into every ML classifier and algorithm. And its always non-zero in practice. In fact, hitting every test in some sort of test suite is likely a sign of a less accurate classifier, not a more accurate one.
Well no shit, but that doesn't mean you just trust them just because they say it works. The cost of a false accusation can be very high.
Re: Taste Is All That's Left
#586Earlier quoted context omitted.
>Don't you and all the people in this thread claiming AI slop know that AI was trained on human writing? Do you think this implies that it is impossible to distinguish LLM-generated text from human text? I don't think that follows at all. >Running Llama 3 between your ears? I'm always happy for a chance to improve my classifier! Can you provide some long-form writing from before 2024 that uses this style?
No, I dont think there is any reliable data to be able reliably classify AI text. So far what I've seen is humans actually fail this task reliably
Interesting hypothesis! I find that my internal sense of whether the article is AI almost always agrees with Pangram when I check it, and almost never triggers on pre-2024 text, even if I didn't know beforehand when it was written. I think this cuts against your hypothesis -- if I was just recognizing a generic style of writing, I would expect it to fire on pre-2024 text, and if I was just recognizing text written in 2026, I would expect it to fire much more often when reading blog posts of already-established writers who I trust.
Re: Taste Is All That's Left
#587People in this thread are using the phrase "AI slop" as if everything made by AI is automatically slop. If you believe this, fair enough, but then you must support the abolition of AI. If you believe that AI is capable of making good things, then it doesn't matter whether this article, or anything, was AI generated. Whatever it is -- software, prose, images -- if you have taste, you don't have to figure out if it was…
I disagree. Using TechJudge from your blog, I'd put AI into the category of technologies that cause more harm than benefit for humanity.
You once said we're like children in a forest, running around and putting every berry we find in our mouths. I believe generative AI (in the form of being generally available chatbots, writing and coding tools, image and music generators, etc) is a berry that's poisonous. Certainly so if we eat much of it. Making them generally available feels analogous to making the technology of nuclear fission or the invention of gunpowder generally available and shove it into every aspect of our lives. There are clearly some specific use cases where these technologies can be beneficial (in case of AI cancer pattern recognition from CT images sounds like one to me). But we need to be careful what we're using them for.
So yes, in this sense their use (as a generally available technology) should be abolished or at least heavily restricted. It would also limit the externalities it's mindless use generates in environmental cost.