3 ms·

https://www.pangram.com/history/c837bbcb-4aed-43e1-987c-afe25ad4dde2?ucc=O8HdtLqflRu https://www.pangram.com/history/c837bbcb-4aed-43e1-987c-afe2... I know you

https://www.pangram.com/history/c837bbcb-4aed-43e1-987c-afe25ad4dde2?ucc=O8HdtLqflRu https://www.pangram.com/history/c837bbcb-4aed-43e1-987c-afe2... I know you're being downvoted, because people are tired of hearing "this is AI". But I think your comment "this is AI" is warranted. This content, explaining something of cultural, personal significance (instead of the typical technical explainer) should really be written by a human.

asgraham 2mo ago
How good are these detectors? (Serious question— I’ve never had a reason to use one)
- no_multitudes 2mo ago
  Pangram's actual classifier model is really strong. It has an incredibly low false positive rate. Whenever people complain about it, I ask them to provide verifiable non-adversarial long-form false positives, and nobody has done so yet. False negatives are more common (as they should be -- it's better to let an AI article through than to false accuse someone.) Their frontend is often misleading, though, for these reasons: - 100% means that it thinks 100% of the article is AI, not that it's 100% confident the article is AI. It's very easy for people to get this confused. - It is much less accurate on shorter text, and doesn't advertise this fact clearly enough. - It works on chunks of text (I think they're like 100 tokens) and can't identify within that chunk which parts are AI and which parts are human.) This should be more clear in the UI, I think.
astrobe_ 2mo ago
Possible false positive because of the text was translated from Japanese to English with AI? I don't see the point of starting a newsletter written by AI.. But I don't understand the point of cheating with a computer on lichess.org either.
- twister2920 2mo ago
  that's a true positive
- wavemode 2mo ago
  > Possible false positive because of the text was translated from Japanese to English with AI? People say this a lot, but I've never seen Claude or any other LLM insert clichéd LLM-isms into translated text. There's obviously a degree of interpretation and "smoothing" that goes into translation, but it doesn't rise to the level of making the entire thing sound like an AI's voice. This blog post was prompted, not translated.
- StilesCrisis 2mo ago
  A Japanese blog wouldn't discuss this. It'd be like an American blog with an article about "what does thank you mean." We already know.
teo_zero 2mo ago
> But I think your comment "this is AI" is warranted. Clue: it's never warranted. Nobody wants to hear it. Even when it's true. Even when it's evident (actually less so when it's evident).
- weiliddat 2mo ago
  Why?
  - teo_zero 2mo ago
    If I have read the article, then reached the comments sections, it's because I have found the subject interesting. It's like watching a movie and someone trying to tell me unsolicited news about the director's private life. I don't care: tell me about the movie, instead. Comment about the content, not the form.
    - seanhunter 2mo ago
      In the case of a blog that purports to be personal reflections, commenting that it is actually AI output is a comment on the content, not merely the form.
      - teo_zero 2mo ago
        The topic is the expression "itadakimasu". What the blog purports to be, belongs in the boundary. Who wrote it belongs in the boundary. Whether the syntax is correct, contains em-dashes, uses nice colors and fonts, belong in the boundary. Comment about "itadakimasu" are on topic, others are off topic.
        weiliddat 2mo ago
        I think someone else wrote it better than I can: https://news.ycombinator.com/item?id=49080703 https://news.ycombinator.com/item?id=49080703 This is worthy feedback to the author (and others), that writing with clear AI tells/register is so distracting that it takes away from the spirit and content of the writing. Perhaps this is just the zeitgeist and in 20 years this is irrelevant. It’s relevant for whoever is writing now.

asgraham 2mo ago

How good are these detectors? (Serious question— I’ve never had a reason to use one)

no_multitudes 2mo ago
Pangram's actual classifier model is really strong. It has an incredibly low false positive rate. Whenever people complain about it, I ask them to provide verifiable non-adversarial long-form false positives, and nobody has done so yet. False negatives are more common (as they should be -- it's better to let an AI article through than to false accuse someone.) Their frontend is often misleading, though, for these reasons: - 100% means that it thinks 100% of the article is AI, not that it's 100% confident the article is AI. It's very easy for people to get this confused. - It is much less accurate on shorter text, and doesn't advertise this fact clearly enough. - It works on chunks of text (I think they're like 100 tokens) and can't identify within that chunk which parts are AI and which parts are human.) This should be more clear in the UI, I think.

astrobe_ 2mo ago

Possible false positive because of the text was translated from Japanese to English with AI? I don't see the point of starting a newsletter written by AI.. But I don't understand the point of cheating with a computer on lichess.org either.

twister2920 2mo ago
that's a true positive
wavemode 2mo ago
> Possible false positive because of the text was translated from Japanese to English with AI? People say this a lot, but I've never seen Claude or any other LLM insert clichéd LLM-isms into translated text. There's obviously a degree of interpretation and "smoothing" that goes into translation, but it doesn't rise to the level of making the entire thing sound like an AI's voice. This blog post was prompted, not translated.
StilesCrisis 2mo ago
A Japanese blog wouldn't discuss this. It'd be like an American blog with an article about "what does thank you mean." We already know.

teo_zero 2mo ago

> But I think your comment "this is AI" is warranted. Clue: it's never warranted. Nobody wants to hear it. Even when it's true. Even when it's evident (actually less so when it's evident).

weiliddat 2mo ago
Why?
- teo_zero 2mo ago
  If I have read the article, then reached the comments sections, it's because I have found the subject interesting. It's like watching a movie and someone trying to tell me unsolicited news about the director's private life. I don't care: tell me about the movie, instead. Comment about the content, not the form.
  - seanhunter 2mo ago
    In the case of a blog that purports to be personal reflections, commenting that it is actually AI output is a comment on the content, not merely the form.
    - teo_zero 2mo ago
      The topic is the expression "itadakimasu". What the blog purports to be, belongs in the boundary. Who wrote it belongs in the boundary. Whether the syntax is correct, contains em-dashes, uses nice colors and fonts, belong in the boundary. Comment about "itadakimasu" are on topic, others are off topic.
      - weiliddat 2mo ago
        I think someone else wrote it better than I can: https://news.ycombinator.com/item?id=49080703 https://news.ycombinator.com/item?id=49080703 This is worthy feedback to the author (and others), that writing with clear AI tells/register is so distracting that it takes away from the spirit and content of the writing. Perhaps this is just the zeitgeist and in 20 years this is irrelevant. It’s relevant for whoever is writing now.