Thank you for this. Great discussion on responsible AI use and output. The question at around the 30:00 mark is the right question to be asking. I take great lengths in skills, guardrails, maintaining brand voice and my thought directions as well as the research and experience that I put into my articles. It's real work.
When people use AI to think FOR them vs. directing AI ... that seems to be the disconnect.
Ultimately, wether or not AI was used... my name is attached... and that makes me responsible for what is being said. It should not be judged by what tool I used to say it.
First off, how I make this: with my mind and every tool I deem useful (AI, monitor, Google search, spectacles, chair) - it is just the responsible thing to do.
The observation: Substack’s Pangram integration is framed as “transparency for readers” but the way it’s scoped and implemented misallocates scrutiny, mismanages risk, and undermines both creative and business uses of AI. What could have been a genuinely helpful provenance signal is a one‑sided audit regime on Substack authors today.
Explicitly scoping it for posts and comments without a word on implementing it for advertisements furthers corporate capture while penalizing layman use. If claudefishing is a problem, where is the catfish detector? Chris Best says that Substack uses AI “all the time” internally (for writing software, research, and building features) so why is AI inside the factory treated as an implementation detail but AI at the tailpipe (author's text posts) treated as a moral object for scrutiny? And Substack repeats the same old corporate play: they get bonus points for protecting their platform from "AI slop" and leaving Pangram to own any mis-detections.
How they could've done this better (not perfect): declare AI-transparency site-wide, regardless of when anything was posted, but starting from a future date. Else, launch it for authors alone and keep the scores private to them, allow them to update honest "How I Make This" sections ...and launch site-wide after a test period. Publish a human and moral charter - state clearly what is against Substack's ethos (Pangram is not policy, it is as imperfect a tool as AI). Add a thumbs-down quality button, allow humans to flag poor content (not Pangram).
Here Chris Best (and separately Substack posts) talk about implementing in ways that do not match what Substack is doing - and this has a smell. Given the archives‑safe cutoff, non-advertisement scanning, reliance on an external imperfect detector, and lack of quantitative AI transparency data, this exercise looks less like real‑world testing and more like misguided policing and risk offloading.
I am glad people are thinking about measuring the value of content.
I asked A.I. to assess your conversation. It replied:
Metric Rating
Original Thinking 7.5/10
Quality of Conversation 8.5/10
Conceptual Entropy 8.5/10
Information Density 6.5/10
Novel Framing 9/10
Overall Intellectual Value 8/10
The conversation is not groundbreaking in the sense of producing entirely new theories, but it is substantially above average because it generates several useful framings ("denial-of-service against the public square," "thick wrappers around AI," "Pangram for ideas") and continually pushes toward the harder question: how to preserve valuable human discourse in a world where content generation becomes nearly free.
Great piece - I'm facing this issue in my creative spaces and struggling with it internally as well. If I use an LLM to make a graphic, then iterate with that many times - at which point is it my creation and I'm just using a tool? The publics spaces will argue that ANY use makes it not mine. Is AI really the issue that artists are whining about these days? It's learning art the same way they did (I did) by absorbing all it can and then imitating before branching out into creation - it just does it at an accelerated rate of speed compare to humans. Those that argue with me about it I ask "Where were they when scanners were getting better and digital camera that were 'forcing' my father and brothers out of work as photographers?" They say it's not the same - yet it is. It's a tool just the same as those. My brothers still make good money doing photography in the age of AI. Why? They are damn good at what they do and those that want that touch pay for it - those who are too cheap or feel any photograph is good enough never take up their time. I'd love to have a conversation with you about my blackbox AI programming company Nate - I think you'd find it an interesting conversation worth your time.
The Ideas Graph is a clear step up from Pangram, but both still sit in the detection family, which is third party. There’s a third signal that sits alongside them that’s already standardised. Let the writer declare: C2PA with CAWG identity assertions was built for exactly this, a cryptographically signed statement of who made what, with which tools, that travels with the work. Detection treats every writer as a suspect and produces an arms race. Attestation treats disclosure as a craft mark and produces a positive signal that honest writers will want to adopt. You said it yourself: this post scores AI-assisted, and the meaning is still yours. A signature saying so beats the scanner guessing. I’ve run a simple disclosure footer on my own essays for months. Nobody trusts them less. Substack is the one platform where creator identity is the product, so oddly, the natural home for signed provenance rather than just more slop scanning.
Written by Ian. The framing took shape in collaboration with Claude against a shared context harness; the thought and the words above are mine. Claude helped open it and stamps its close. 🌱🧵🔊
99% of posts on LinkedIN that start with “wild, insane, xyz just finished abc…” are just 💩. The whole thing is becoming unfortunately useless. Reddit is well advised to remove/govern/block that type of content. Thank you for bringing this topic up.
You might want to take a look at a history/sci method called the Daliahgram. In my field it’s used to pull disparate quant and qual factors into a single annalysis. Different space, but it really echos what I think you’re going for here & in the newsletter. Authors are Frachetti, Di Cosmo & Büntgen. (2023) “The dahliagram: An interdisciplinary tool for investigation, visualization, and communication of past human-environmental interaction”
Thank you for this. Great discussion on responsible AI use and output. The question at around the 30:00 mark is the right question to be asking. I take great lengths in skills, guardrails, maintaining brand voice and my thought directions as well as the research and experience that I put into my articles. It's real work.
When people use AI to think FOR them vs. directing AI ... that seems to be the disconnect.
Ultimately, wether or not AI was used... my name is attached... and that makes me responsible for what is being said. It should not be judged by what tool I used to say it.
First off, how I make this: with my mind and every tool I deem useful (AI, monitor, Google search, spectacles, chair) - it is just the responsible thing to do.
The observation: Substack’s Pangram integration is framed as “transparency for readers” but the way it’s scoped and implemented misallocates scrutiny, mismanages risk, and undermines both creative and business uses of AI. What could have been a genuinely helpful provenance signal is a one‑sided audit regime on Substack authors today.
Explicitly scoping it for posts and comments without a word on implementing it for advertisements furthers corporate capture while penalizing layman use. If claudefishing is a problem, where is the catfish detector? Chris Best says that Substack uses AI “all the time” internally (for writing software, research, and building features) so why is AI inside the factory treated as an implementation detail but AI at the tailpipe (author's text posts) treated as a moral object for scrutiny? And Substack repeats the same old corporate play: they get bonus points for protecting their platform from "AI slop" and leaving Pangram to own any mis-detections.
How they could've done this better (not perfect): declare AI-transparency site-wide, regardless of when anything was posted, but starting from a future date. Else, launch it for authors alone and keep the scores private to them, allow them to update honest "How I Make This" sections ...and launch site-wide after a test period. Publish a human and moral charter - state clearly what is against Substack's ethos (Pangram is not policy, it is as imperfect a tool as AI). Add a thumbs-down quality button, allow humans to flag poor content (not Pangram).
Here Chris Best (and separately Substack posts) talk about implementing in ways that do not match what Substack is doing - and this has a smell. Given the archives‑safe cutoff, non-advertisement scanning, reliance on an external imperfect detector, and lack of quantitative AI transparency data, this exercise looks less like real‑world testing and more like misguided policing and risk offloading.
I am glad people are thinking about measuring the value of content.
I asked A.I. to assess your conversation. It replied:
Metric Rating
Original Thinking 7.5/10
Quality of Conversation 8.5/10
Conceptual Entropy 8.5/10
Information Density 6.5/10
Novel Framing 9/10
Overall Intellectual Value 8/10
The conversation is not groundbreaking in the sense of producing entirely new theories, but it is substantially above average because it generates several useful framings ("denial-of-service against the public square," "thick wrappers around AI," "Pangram for ideas") and continually pushes toward the harder question: how to preserve valuable human discourse in a world where content generation becomes nearly free.
I posted a few paragraphs of one of Nate’s pieces from early July. 100% AI!
Great piece - I'm facing this issue in my creative spaces and struggling with it internally as well. If I use an LLM to make a graphic, then iterate with that many times - at which point is it my creation and I'm just using a tool? The publics spaces will argue that ANY use makes it not mine. Is AI really the issue that artists are whining about these days? It's learning art the same way they did (I did) by absorbing all it can and then imitating before branching out into creation - it just does it at an accelerated rate of speed compare to humans. Those that argue with me about it I ask "Where were they when scanners were getting better and digital camera that were 'forcing' my father and brothers out of work as photographers?" They say it's not the same - yet it is. It's a tool just the same as those. My brothers still make good money doing photography in the age of AI. Why? They are damn good at what they do and those that want that touch pay for it - those who are too cheap or feel any photograph is good enough never take up their time. I'd love to have a conversation with you about my blackbox AI programming company Nate - I think you'd find it an interesting conversation worth your time.
The Ideas Graph is a clear step up from Pangram, but both still sit in the detection family, which is third party. There’s a third signal that sits alongside them that’s already standardised. Let the writer declare: C2PA with CAWG identity assertions was built for exactly this, a cryptographically signed statement of who made what, with which tools, that travels with the work. Detection treats every writer as a suspect and produces an arms race. Attestation treats disclosure as a craft mark and produces a positive signal that honest writers will want to adopt. You said it yourself: this post scores AI-assisted, and the meaning is still yours. A signature saying so beats the scanner guessing. I’ve run a simple disclosure footer on my own essays for months. Nobody trusts them less. Substack is the one platform where creator identity is the product, so oddly, the natural home for signed provenance rather than just more slop scanning.
Written by Ian. The framing took shape in collaboration with Claude against a shared context harness; the thought and the words above are mine. Claude helped open it and stamps its close. 🌱🧵🔊
99% of posts on LinkedIN that start with “wild, insane, xyz just finished abc…” are just 💩. The whole thing is becoming unfortunately useless. Reddit is well advised to remove/govern/block that type of content. Thank you for bringing this topic up.
You might want to take a look at a history/sci method called the Daliahgram. In my field it’s used to pull disparate quant and qual factors into a single annalysis. Different space, but it really echos what I think you’re going for here & in the newsletter. Authors are Frachetti, Di Cosmo & Büntgen. (2023) “The dahliagram: An interdisciplinary tool for investigation, visualization, and communication of past human-environmental interaction”