I wonder if the issue is that AIs just have no idea how to draw a clock. Or, is it that they’ve been trained on papers where doctors talk about the various issues patients sometimes have when drawing clocks.
I suspect it’s probably the first one. AIs seem to have a real problem with anything visually complicated. One of the easiest ways to spot AI slop is to look at the logos on t-shirts.
AI run on statistical probability. The more options there are in something, the harder it is to get right. A clock face becomes hard because practically the entire clock changes from moment to moment, and logos on shirts are similar in that they’re largely all in the same place but vary dramatically in shape to be as distinct from each other as possible, which is basically the exact opposite of what an AI wants. When your whole thing is basically averaging stuff out on a probability curve that’s weighted towards specific keywords, having massive variation in your data points is bad.
Yeah and some clocks have a second hand and some don’t sometimes clocks use roman numerals sometimes they’re arabic numerals, and that’s if it can understand based on context if someone saying just “clock” in the data the scraped is referring to a digital clock or an analog clock.
In general LLMs don’t understand logic, though I suspect they have given some of them ability to run some code validation logic (that’s not actually AI) when you tell it to generate code in some languages. I say this because I’ve had it produce some code that could compile, but it seemingly put some example code into a function and had some other example code that needed to call that function with another parameter so it just created a third function that accepts the additional parameter and calls the first function (throwing away that parameter). It compiles but doesn’t have any understanding of how stupid that is on a logical level. So it seems like it’s just trying stuff until it’s capable of compiling without there being any understanding of how anything works.
I’ll just leave this here: https://clocks.brianmoore.com/
I wonder if the issue is that AIs just have no idea how to draw a clock. Or, is it that they’ve been trained on papers where doctors talk about the various issues patients sometimes have when drawing clocks.
I suspect it’s probably the first one. AIs seem to have a real problem with anything visually complicated. One of the easiest ways to spot AI slop is to look at the logos on t-shirts.
AI run on statistical probability. The more options there are in something, the harder it is to get right. A clock face becomes hard because practically the entire clock changes from moment to moment, and logos on shirts are similar in that they’re largely all in the same place but vary dramatically in shape to be as distinct from each other as possible, which is basically the exact opposite of what an AI wants. When your whole thing is basically averaging stuff out on a probability curve that’s weighted towards specific keywords, having massive variation in your data points is bad.
Yeah and some clocks have a second hand and some don’t sometimes clocks use roman numerals sometimes they’re arabic numerals, and that’s if it can understand based on context if someone saying just “clock” in the data the scraped is referring to a digital clock or an analog clock.
In general LLMs don’t understand logic, though I suspect they have given some of them ability to run some code validation logic (that’s not actually AI) when you tell it to generate code in some languages. I say this because I’ve had it produce some code that could compile, but it seemingly put some example code into a function and had some other example code that needed to call that function with another parameter so it just created a third function that accepts the additional parameter and calls the first function (throwing away that parameter). It compiles but doesn’t have any understanding of how stupid that is on a logical level. So it seems like it’s just trying stuff until it’s capable of compiling without there being any understanding of how anything works.
Entertaining mess. Nine new borked clocks a minute. A couple readable.
Oh, those AIs just really love that 2013 Hannibal Series!
I remember studying this (the illness, not the show) in psychology. Something like wernickes aphasisa? Or maybe brochas aphasisa?
Your brain fails to process like the left half of what it’s seeing, so they draw clocks only showing the left half with so the numbers there.
I was assuming they saw the post here a couple of days ago and screenshotted or got the idea from there
LGTM! (probably on a Friday)
Ship it for beta testing
deleted by creator