There are none. The shorter is 56 words. Pangram doesnt run the test on texts shorter than 50 words. But, yes, a ~60-word text still gives you a low-confidence verdict, as I discuss.
I don't think using Pangram to check a text's authenticity is equivalent to passing moral judgment on it, and ultimately it seems like the tool is doing exactly what I'd want it to do. If a human spits a bunch of ideas into a chatbot prompt for the AI to flesh out, I'd want Pangram to tell me that the text is AI generated, even if the ideas aren't. And if the ideas are AI generated, but a human goes and reworks the text to make it stylistically distinct, I'd still want Pangram to tell me that the text is essentially human generated. This seems... precisely the point of a tool like Pangram, no?
I don't think we'll ever get a magic wand that we can wave that tells us whether some text that is the combined effort of human and AI is "worth reading" or "fundamentally slop," but I think it's good to have tools that push back against the rising onslaught of large amounts of generated text. I've written about the problems inherent in this, where even if the text can seem compelling and good, there's this asymmetrical challenge in parsing whether the text is actually based on some fundamentally sound view of the world or whether it's just pattern matching to persuasive context.
Mine is not a critique of Pangram, which is very impressive. It is a discussion of (the absence of discussion about) why we care about Pangram. I think it's just a childhood disease of the AI era.
The reader should always be lazier than the writer. That's what makes writing worth reading for me.
I like this article, but you definitely should not have included any verdicts on text with less than 50 words.
There are none. The shorter is 56 words. Pangram doesnt run the test on texts shorter than 50 words. But, yes, a ~60-word text still gives you a low-confidence verdict, as I discuss.
I don't think using Pangram to check a text's authenticity is equivalent to passing moral judgment on it, and ultimately it seems like the tool is doing exactly what I'd want it to do. If a human spits a bunch of ideas into a chatbot prompt for the AI to flesh out, I'd want Pangram to tell me that the text is AI generated, even if the ideas aren't. And if the ideas are AI generated, but a human goes and reworks the text to make it stylistically distinct, I'd still want Pangram to tell me that the text is essentially human generated. This seems... precisely the point of a tool like Pangram, no?
I don't think we'll ever get a magic wand that we can wave that tells us whether some text that is the combined effort of human and AI is "worth reading" or "fundamentally slop," but I think it's good to have tools that push back against the rising onslaught of large amounts of generated text. I've written about the problems inherent in this, where even if the text can seem compelling and good, there's this asymmetrical challenge in parsing whether the text is actually based on some fundamentally sound view of the world or whether it's just pattern matching to persuasive context.
https://thebsdetector.substack.com/p/geopolitics-slop-is-ascendant
Mine is not a critique of Pangram, which is very impressive. It is a discussion of (the absence of discussion about) why we care about Pangram. I think it's just a childhood disease of the AI era.