Thank you Karo. This transparency and opportunity for dialogue is what is sadly missing from many of the witch hunts that are starting to emerge elsewhere in this platform. 😢
I know. And I really worry that there will be even more people who'll think that not using AI e.v.e.r is a good thing. As if opting out of understanding technology was ever a good thing.
I want to try something. Someone commented that they ran Pangram on a thesis they wrote before AI existed and it was flagged as 100% AI 🤣 It'd be a great community experiment.
I have chosen the transparency path for myself too. Not using AI is like not using keyboard to write. As I am engineer, leveraging AI in our teams removed so many blockers it is hard to even describe. We should just remember to not delegate thinking to it, but use it to enchance our thinking & creativity
I love how you defined “AI assisted craft.” This is exactly how I think about AI. It should be a copilot that helps us refine ideas, think through our work, and get things done, not something that replaces the human thought, creativity, and lived experience behind it.
The human touch is what gives the work meaning. AI should support that, not become a substitute for it.
Checking for craft runs into the same question the AI-origin check does though, just one level up: who is doing the checking, and are they independent of whoever made the thing. A craft-checker built and graded by the same person who made the work has the identical blind spot Pangram has, just pointed at a different target.
Asolutely agree. I sell wooden art on Etsy and my sales have gone up a lot since I started using AI to help write my product descriptions and research what customers are actually looking for. It doesn’t make the pieces I make any less art. The craft is still mine. This whole Pangram thing is BS.
One thing I've been noticing is that readers often describe something as "AI-generated" when what they're may actually be sensing is a lack of care. Care has a way of leaving fingerprints; through judgment, specificity, lived experience, and thoughtful choices. That's why I love the distinction here. The presence of AI doesn't determine whether something feels meaningful. The presence of genuine human attention often does.
Karo, thank you for this. My brain feels friend from how misunderstood these terms are and what AI-assisted actually means. You've broken down the steps in such a simple way to explain the wide void between human > AI. This is such a great one:
The prompt test: This one is my favorite. Can I guess, just by looking at the work, what prompt could have produced it almost exactly as it appears? If I can, it isn’t personal or specific enough. More AI than craft.
As someone who appreciate the need to define these terms more specifically-thank you!
That's so wonderful to hear! Thank you for this comment! I felt it's important we have a proper name for this type of quality work, otherwise we'll just keep circling. And I see from some comments on Substack that people completely misunderstand what Pangram is for - they think it will detect slop. But it doesn't detect quality, it detects the tools used.
I love this point of view. AI is a great tool and should be leveraged. However, we need to stay in control. We need to make the decisions. In the end, we can use a hammer and build something great, or we can build something that isn’t sturdy. It’s how we use the hammer that matters
What I have found that work extremely well is to order bots to save execution plans in a form of checklist you can go through, track progrees and prevent bot of getting lost over time or across sessions.
Yes, I built a similar self-learning element, especially for writing email templates. It learns from previous emails based on open rates, so there's so much possible with AI nowadays.
Separating how work was made from how good it is clears up much of the AI argument.
A human can make slop. AI can help produce careful work. The two-axis model makes that distinction easy to see.
I would only push on the word care. Someone can care deeply and still make weak work. The audience cannot see the effort. They can only judge whether the result is accurate, specific, useful, and worth their time.
Naming the middle category is the useful bit here: the argument usually gets stuck at either end, when what actually separates the two is who made the consequential calls and whether they'd put their name to the result.
Spot on. This is how In use AI and ironically when done this way it comes back as 💯 human but there was definitely ai in the loop as in it was not all handwritten. My essay today is an example of this
You would have to be willfully blind to how people adapt and improve with AI to launch a witch hunt like this. Thank you, Karo, for such a thoughtful and timely article.
I like your general direction, but I think this statement is a bit oversimplified-
"Supporting, as in: it does not get to make the decisions."
There are countless decisions and executions a language model makes, just at a lower level of abstraction than your instructions or direction. Which is fine, that's the point. If it wasn't good at that, we'd have to hand-edit every line of an essay or codebase which defeats the purpose. The more you use AI, generally the handier you get and the more you are able to go up the ladder of abstraction which saves time, allows you to focus on the big stuff, go to the beach, whatever.
At the end quality is the defining standard (slop is slop, as you pointed out), and the user is the gatekeeper no matter what. If a piece of output is good or bad, AI generated or assisted or directed or not, the user decides whether to go forward with that output. Even if you enable OpenClaw to write blog posts or send emails on your behalf, you decided to open that gate.
That's very true. Also this point: "The more you use AI, generally the handier you get..." That's why I'm concerned that Pangram-like tools bet on the wrong side of what it means to be educated. I saw the comments under the original post, intelligent people proudly announcing they they've never used AI. That's nothing to be proud of, why would it be? Since when is opting out of understanding anything a good thing? And tools like Pangram will discourage even more people.
Thanks for your reply! Agreed completely on AI 'detection' tools. The ironic thing is that it's probably easier to develop a tool to rate how generally well-written something is (though that doesn't substitute for human taste), rather than assign an `AI Likelihood' score. These tools are even advertised as having AI humanizer tool detection, which is a clear sign we've jumped the shark here. Bad AI writing has certain patterns in its badness, the badness is the problem not the patterns or that it's AI generated. It's also a pretty obvious example of the pot calling the kettle black LOL.
Thank you Karo. This transparency and opportunity for dialogue is what is sadly missing from many of the witch hunts that are starting to emerge elsewhere in this platform. 😢
I know. And I really worry that there will be even more people who'll think that not using AI e.v.e.r is a good thing. As if opting out of understanding technology was ever a good thing.
And thank you for reading, Sam!
Exactly! I would also say there is a deep level of hypocrisy going on as well...
Definitely. Not to mention they use AI to detect AI. People will find a way to use AI to fool AI. Circus.
I want to try something. Someone commented that they ran Pangram on a thesis they wrote before AI existed and it was flagged as 100% AI 🤣 It'd be a great community experiment.
lol, I should test it too.
Let me know if you do!
I have chosen the transparency path for myself too. Not using AI is like not using keyboard to write. As I am engineer, leveraging AI in our teams removed so many blockers it is hard to even describe. We should just remember to not delegate thinking to it, but use it to enchance our thinking & creativity
I love how you defined “AI assisted craft.” This is exactly how I think about AI. It should be a copilot that helps us refine ideas, think through our work, and get things done, not something that replaces the human thought, creativity, and lived experience behind it.
The human touch is what gives the work meaning. AI should support that, not become a substitute for it.
I wish Pangram and other detection tools focused on checking for craft, not AI.
Checking for craft runs into the same question the AI-origin check does though, just one level up: who is doing the checking, and are they independent of whoever made the thing. A craft-checker built and graded by the same person who made the work has the identical blind spot Pangram has, just pointed at a different target.
Asolutely agree. I sell wooden art on Etsy and my sales have gone up a lot since I started using AI to help write my product descriptions and research what customers are actually looking for. It doesn’t make the pieces I make any less art. The craft is still mine. This whole Pangram thing is BS.
agreed!
One thing I've been noticing is that readers often describe something as "AI-generated" when what they're may actually be sensing is a lack of care. Care has a way of leaving fingerprints; through judgment, specificity, lived experience, and thoughtful choices. That's why I love the distinction here. The presence of AI doesn't determine whether something feels meaningful. The presence of genuine human attention often does.
I noticed that too. People treat them as synonyms.
Karo, thank you for this. My brain feels friend from how misunderstood these terms are and what AI-assisted actually means. You've broken down the steps in such a simple way to explain the wide void between human > AI. This is such a great one:
The prompt test: This one is my favorite. Can I guess, just by looking at the work, what prompt could have produced it almost exactly as it appears? If I can, it isn’t personal or specific enough. More AI than craft.
As someone who appreciate the need to define these terms more specifically-thank you!
That's so wonderful to hear! Thank you for this comment! I felt it's important we have a proper name for this type of quality work, otherwise we'll just keep circling. And I see from some comments on Substack that people completely misunderstand what Pangram is for - they think it will detect slop. But it doesn't detect quality, it detects the tools used.
👏🏽👏🏽👏🏽thank you, yes!!
This is excellent, and spot on. It's time to move past the absolutes on AI, and figure out how you want to use it.
Exactly. Thank you so much for reading, Mike!
I love this point of view. AI is a great tool and should be leveraged. However, we need to stay in control. We need to make the decisions. In the end, we can use a hammer and build something great, or we can build something that isn’t sturdy. It’s how we use the hammer that matters
Exactly. Thank you for reading, Mike!
I never blamed digital photographers for using Adobe photoshop when it first came out, nor did I think their final output was less worthy of praise.
AI assisted craft falls in a similar bucket.
Exactly! Thank you for reading, Anita 🤗
Love the bring your own opinion!
I’m a 51 year old trying to navigate the land of AI.
Giving context and boundaries is my biggest revelation after almost joining the Anti AI camp.
Tell the robot what you don’t want to do is equally important as asking what you want. Well written and thank you for sharing.
What I have found that work extremely well is to order bots to save execution plans in a form of checklist you can go through, track progrees and prevent bot of getting lost over time or across sessions.
Yes, I built a similar self-learning element, especially for writing email templates. It learns from previous emails based on open rates, so there's so much possible with AI nowadays.
This is inspiring!
Passion should come through, refinement, editing etc, all with AI to save time. Thanks for posting.
My pleasure, thank you so much for taking the time to read and comment, Jeanette 🤗
I love this definition, Karo. I'm proudly in the AI assisted craft camp. Thanks for defining it so specifically.
Someone called us AI-assisted Artisans 🤗
Love that!
Separating how work was made from how good it is clears up much of the AI argument.
A human can make slop. AI can help produce careful work. The two-axis model makes that distinction easy to see.
I would only push on the word care. Someone can care deeply and still make weak work. The audience cannot see the effort. They can only judge whether the result is accurate, specific, useful, and worth their time.
Good point. It should be a mix of care and skill.
Naming the middle category is the useful bit here: the argument usually gets stuck at either end, when what actually separates the two is who made the consequential calls and whether they'd put their name to the result.
Yes, we're so early, some parts of the culture related to AI are simply not named yet. Thank you for reading Mel 🤗
Spot on. This is how In use AI and ironically when done this way it comes back as 💯 human but there was definitely ai in the loop as in it was not all handwritten. My essay today is an example of this
I will check it out. Thank you Phil!
You would have to be willfully blind to how people adapt and improve with AI to launch a witch hunt like this. Thank you, Karo, for such a thoughtful and timely article.
My pleasure, thank you very much for reading, Mike.
I like your general direction, but I think this statement is a bit oversimplified-
"Supporting, as in: it does not get to make the decisions."
There are countless decisions and executions a language model makes, just at a lower level of abstraction than your instructions or direction. Which is fine, that's the point. If it wasn't good at that, we'd have to hand-edit every line of an essay or codebase which defeats the purpose. The more you use AI, generally the handier you get and the more you are able to go up the ladder of abstraction which saves time, allows you to focus on the big stuff, go to the beach, whatever.
At the end quality is the defining standard (slop is slop, as you pointed out), and the user is the gatekeeper no matter what. If a piece of output is good or bad, AI generated or assisted or directed or not, the user decides whether to go forward with that output. Even if you enable OpenClaw to write blog posts or send emails on your behalf, you decided to open that gate.
That's very true. Also this point: "The more you use AI, generally the handier you get..." That's why I'm concerned that Pangram-like tools bet on the wrong side of what it means to be educated. I saw the comments under the original post, intelligent people proudly announcing they they've never used AI. That's nothing to be proud of, why would it be? Since when is opting out of understanding anything a good thing? And tools like Pangram will discourage even more people.
Thanks for your reply! Agreed completely on AI 'detection' tools. The ironic thing is that it's probably easier to develop a tool to rate how generally well-written something is (though that doesn't substitute for human taste), rather than assign an `AI Likelihood' score. These tools are even advertised as having AI humanizer tool detection, which is a clear sign we've jumped the shark here. Bad AI writing has certain patterns in its badness, the badness is the problem not the patterns or that it's AI generated. It's also a pretty obvious example of the pot calling the kettle black LOL.
It is, heheh. Indeed! Thank YOU for reading and taking the time to comment!