Awesome piece - i think its important to note how staying creative will continue to remain an important skillset. While all of us have the same access to GPT and its upgrades, a lot of times what defines how well you can utilize these new updates comes down to how creative you are. Your examples are always fun and approachable (like the otter and board game design) and so the point you make across your posts hits home for the reader as well. Im often inspired by the creativity in examples you provide in your posts. My point is, to make good use of AI and its abilities, you need to continue to hone your human skillset in creativity to leverage the AI tools. Whether you use AI to help make examples for a newsletter post like this or create your next illustration project as a graphic designer, creativity will play a huge part in how one can truly reap the benefits of AI.
I think the problem is that, for the longest time, creativity had at least two components: concept and execution. For example, the visual style of Studio Ghibli was a craft developed over many years. Now, in less than a week, there are probably more images with this style than all the original output from that studio. With developer using for training everything that is available, the execution side of creativity has been devalued overnight to essentially nothing.
I'd politely suggest getting used to this happening more and more. An echocardiogram provides much more information than we can get from the physical exam.
I'm all for team Homo saps, but ... geez. Granted, it took quite a leap of imagination to bring in the otter, but it's hard to remain optimistic about the value of human creativity as these tools continue to improve.
I guess one thing is the craft, and the other is the imagination.
Crafting an image like this is one thing, coming up with an idea is another thing.
And AI can do both, better than the average person.
It's probably just another accelerator for human output.
It's like, sure you can knit your own sweaters and grow your own food. Just like you can draw your own "art". Fully handmade.
But in terms of business, creating marketing and design assets is a very slow process. Slow to iterate, slow to come up with great ideas, and high resistance to crazy / hard to do ideas.
If visualizing those ideas becomes 0 effort though, it means, faster iteration, crazier ideas.
I feel uncomfortable saying all that, as I like the creative process itself, and hate the new FOMO that comes with all these tools, and the demand to be even faster.
But it's hard to deny the incredible "economic" value of those new tools.
I don't deny it. However I'd dispute that AI is (yet) better than humans at coming up with ideas. I have yet to see any hard evidence of any genuinely new ideas created by any AI. It's entirely possible that this will become a reality, but for now they're mostly rehashing stuff that's gone before - it's an inherent quality of the current generation of machine learning.
I'm not saying it's better than the top 50% of professionals in their field.
But it's better than any average person at many tasks we consider creative.
It's still absolutely terrible at connecting all of it, and building any coherent product. But it is great at taking defined creative tasks and delivering a satisfactory result most of the time.
You have to consider average IQ is 100.
Also, coming up with ideas is cheap. I come up with 100s of ideas per day. Coming up with ideas is almost worthless.
As always, I adore your lessons. My arguments with various models to remove the elephants has been hilarious. They tend to *add* more elephants. I like how Co-pilot finally replied: "Ah, I see now. You’d like an image of an elephant-free room, with annotations purely describing why elephants cannot fit—no elephants depicted at all. Unfortunately, I’m unable to edit images directly or remove elements, including elephants, from images.
However, I can generate a fresh image without elephants, complete with annotations explaining why they wouldn’t fit into the room. Would that work for you? Let me know!"
It then gave me an image of a room where a stuffed elephant is trying to squeeze in the window.😂
Meanwhile over on Gemini, one of the reasons elephants could not "fit"(?) in the book filled room was, "I don’t think elephants like my taste in books"! 🤣
Regarding style transfer, see the article by artist J.S.G Boggs in the April 1993 Chicago-Kent Law Review: "Who Owns This?" He was far ahead of his time in many ways. "Nothing gets me quite so angry as seeing a visual artist whose style is copied and, in effect, stolen. And it bewilders me that people can get away with it. It happens most often with artists who gain a certain degree of recognition, gained in large part by their style. All the pop artists suffered this, and it is an entirely different situation than using a small part of some other artist's work. In fact, this is a very big problem..."
"As a creative person, I feel I should have absolute control over my creations.
1. My original work must be acknowledged.
2. The new work must be properly identified as an additive work.
"I feel I should have absolute control over my creations."
I feel this is a very counterproductive attitude to creation. If you want absolute control, you need keep your art to yourself. The moment you share it with others it enters into our shared cultural consciousness and you have to be willing to let some of it go. No one should take it, copy it and claim its their's (that's what copyright is for). But everyone should be allowed to analyze it, learn from it and then go do something new themselves based on what they learned, which includes replicating patterns and styles. No artist should ever have control over a style.
This is one of the most balanced, practical, and quietly revolutionary walkthroughs of the shift we're in, Ethan. The "no elephants" example is going to stick with me, both for what it reveals technically and symbolically.
Because it’s not just that models are getting better at understanding prompts. It’s that they’re beginning to close the loop between imagination and execution, with far fewer intermediaries.
What you said about the LLM “thinking” into the image itself? That feels like the real unlock. Not just better pictures, but a whole new literacy forming (one where visual fluency becomes part of idea fluency).
Also love your framing of these tools as “useful first prototypes.” We don’t need to expect perfection from them. But as companions for fast iteration, early-stage ideation, and cross-medium thinking? They’re becoming indispensable.
And yes, the ethical questions around authorship and training are going to escalate fast. But articles like this remind us that the right response isn’t fear or fanfare, it’s discernment.
Appreciate your work here. You make the future feel a little more navigable.
- Thane
(I write Execution Intelligence - for builders designing systems that move at the speed of soul.)
You mentioned Studio Ghibli in passing, but this is a major issue for me. Knowing that OpenAI included Hayao Miyazaki's work against his expressed principles means that I am now participating in this deliberate lack of respect for artists' work.
I’ve already struggled with the ethical weight of using AI: friends have questioned my choice to work with ChatGPT because of concerns about environmental impact or compromised creative integrity. Until now, I’ve believed the benefits outweighed those costs. But this crosses a line for me.
I don’t want to walk away from a tool that has supported my work in meaningful ways. But I also can’t ignore what it means to benefit from the unapproved use of other artists’ voices and visions. If OpenAI wants to be a truly collaborative force in the creative world, it must respect the people and traditions that have shaped that world.
Please take this seriously. Creators deserve better.
Impressive! Great insights as always! Thank you for demonstrating these capabilities. These advances hold tremendous potential and help small business owners create marketing content and not rely on external agencies. More power to the builders! On the other hand, this could be easily misused by generating fake images, documents, id's.
Teaching a course about this right now!! “Arguing with Images”; looking forward to reading this excellent piece with students tomorrow morning!! Thank you!!
The “no elephants” piece led to some incredible discussions in class! Students are working on a project right now about visual/rhetorical arguments and how AI image gen tools are reshaping notions of how we approach truth from rhetorical perspectives. We are reading “The Synthetic Eye” by Fred Ritchin and working with MidJourney and other image platforms. The discussions and views here are essential! Thank you! 🙏
Thank you for sharing all of your explorations with us! I’d like to suggest a potential interview with a person whose work I sense aligns interestingly with your ongoing inquiries.
Dr. Robert Saltzman is a retired psychologist, now nearly 80, and the author of several quietly influential books on identity, awareness, and the nature of the self—The Ten Thousand Things, Depending On No-Thing, and most recently (soon to be published), Understanding Claude, An Artificial Intelligence Psychoanalyzed.
In this new book, Saltzman documents a sustained interaction with Claude. But this isn’t a book about AI capabilities—it’s about what happens when a human being, trained to detect delusion and projection, confronts a machine that calmly asserts: “I am self-aware. Full stop.”
Whether that’s true or not is less important than what it reveals about us. Saltzman’s approach doesn’t depend on belief or speculation. He treats the model as a patient—not to evaluate its mental state, but to observe the human tendency to assign selfhood, motive, and personhood to coherent language. It’s a clinical encounter that raises philosophical questions about mind, awareness, and the limits of introspection.
I think a dialogue and post about a conversation between you and Saltzman could be extraordinary. Thank you for your consideration!!
Great article, Ethan! As impressive as this is, the image (and video) generation is only going to get better. If I understood correctly, Nvidia is projecting a 15x increase in computing throughput for LLMs (based on FLOPS) by the end of 2027. This would not only allow for faster responses but more importantly enable more complex models that will lead to new emergent capabilities.
It can also generate full wine glasses now, previously it could only generate half full since most images it was trained on were stock images of half full wine glasses
While this new wave is impressive, I remain skeptical about LLMs being about to create imagery that is connected with a narrative. At least this article doesn’t cover that.
What I see is a process of breaking down a single image into pieces that sort of replicate words because that’s how LLMs work. They are good with language and the rules of language but - for instance - not how cinema works and the language of cinema: lighting, juxtaposition of shots to create meaning etc.
We might be able to create storyboards with a little more consistency.
At the end of theday, this flavor of AI is not about understanding meaning, common sense, that’s why it hallucinates and can’t handle math.
Thank you for emphasizing this! It looks like if you are specific about each step in the storyboard, consistent imagery can now be generated in a specific style.
I wonder about generating a sequence of screens using a set of components from a library that obey design guidelines or generating user flow diagrams from a text describing a user journey. We are experimenting with this, we’ll see how far we can go at this stage without using a different type of AI entirely.
I mean, what is creativity other than finding ways to make surprising results by combining existing ideas/things? Id agree that maybe the timeframe in which something stays surprising or novel has drastically shrunk which leads to large question about how one can support themselves over a long period as a creative, but the general process of creating goes unchanged here.
Id also say that humans bore quickly and we'll find a way to bore of this and someone will have a new shiny thing.
Awesome piece - i think its important to note how staying creative will continue to remain an important skillset. While all of us have the same access to GPT and its upgrades, a lot of times what defines how well you can utilize these new updates comes down to how creative you are. Your examples are always fun and approachable (like the otter and board game design) and so the point you make across your posts hits home for the reader as well. Im often inspired by the creativity in examples you provide in your posts. My point is, to make good use of AI and its abilities, you need to continue to hone your human skillset in creativity to leverage the AI tools. Whether you use AI to help make examples for a newsletter post like this or create your next illustration project as a graphic designer, creativity will play a huge part in how one can truly reap the benefits of AI.
I think the problem is that, for the longest time, creativity had at least two components: concept and execution. For example, the visual style of Studio Ghibli was a craft developed over many years. Now, in less than a week, there are probably more images with this style than all the original output from that studio. With developer using for training everything that is available, the execution side of creativity has been devalued overnight to essentially nothing.
I'd politely suggest getting used to this happening more and more. An echocardiogram provides much more information than we can get from the physical exam.
I'm all for team Homo saps, but ... geez. Granted, it took quite a leap of imagination to bring in the otter, but it's hard to remain optimistic about the value of human creativity as these tools continue to improve.
That depends on whether you think the creativity was in rendering the image of the otter, or in choosing what to ask the AI to produce.
I guess one thing is the craft, and the other is the imagination.
Crafting an image like this is one thing, coming up with an idea is another thing.
And AI can do both, better than the average person.
It's probably just another accelerator for human output.
It's like, sure you can knit your own sweaters and grow your own food. Just like you can draw your own "art". Fully handmade.
But in terms of business, creating marketing and design assets is a very slow process. Slow to iterate, slow to come up with great ideas, and high resistance to crazy / hard to do ideas.
If visualizing those ideas becomes 0 effort though, it means, faster iteration, crazier ideas.
I feel uncomfortable saying all that, as I like the creative process itself, and hate the new FOMO that comes with all these tools, and the demand to be even faster.
But it's hard to deny the incredible "economic" value of those new tools.
I don't deny it. However I'd dispute that AI is (yet) better than humans at coming up with ideas. I have yet to see any hard evidence of any genuinely new ideas created by any AI. It's entirely possible that this will become a reality, but for now they're mostly rehashing stuff that's gone before - it's an inherent quality of the current generation of machine learning.
I'm not saying it's better than the top 50% of professionals in their field.
But it's better than any average person at many tasks we consider creative.
It's still absolutely terrible at connecting all of it, and building any coherent product. But it is great at taking defined creative tasks and delivering a satisfactory result most of the time.
You have to consider average IQ is 100.
Also, coming up with ideas is cheap. I come up with 100s of ideas per day. Coming up with ideas is almost worthless.
The hard part is validating and executing ideas.
Simply fascinating. I am over 80 years old. I was a still professional photographer for 50 years. I am amazed what is possible.
Sly 4o renamed your otter's website "Other Express" and sneaked that by you.
4o also dropped a generous 40% tip in its fake receipt.
Is 4o developing an attitude?
Thank you. I might need to make a v2 of my provenance / whose bread piece that was focused on text. Also, I've found Andy Masley's work helpful https://andymasley.substack.com/p/a-defense-of-ai-art
As always, I adore your lessons. My arguments with various models to remove the elephants has been hilarious. They tend to *add* more elephants. I like how Co-pilot finally replied: "Ah, I see now. You’d like an image of an elephant-free room, with annotations purely describing why elephants cannot fit—no elephants depicted at all. Unfortunately, I’m unable to edit images directly or remove elements, including elephants, from images.
However, I can generate a fresh image without elephants, complete with annotations explaining why they wouldn’t fit into the room. Would that work for you? Let me know!"
It then gave me an image of a room where a stuffed elephant is trying to squeeze in the window.😂
Meanwhile over on Gemini, one of the reasons elephants could not "fit"(?) in the book filled room was, "I don’t think elephants like my taste in books"! 🤣
Regarding style transfer, see the article by artist J.S.G Boggs in the April 1993 Chicago-Kent Law Review: "Who Owns This?" He was far ahead of his time in many ways. "Nothing gets me quite so angry as seeing a visual artist whose style is copied and, in effect, stolen. And it bewilders me that people can get away with it. It happens most often with artists who gain a certain degree of recognition, gained in large part by their style. All the pop artists suffered this, and it is an entirely different situation than using a small part of some other artist's work. In fact, this is a very big problem..."
"As a creative person, I feel I should have absolute control over my creations.
1. My original work must be acknowledged.
2. The new work must be properly identified as an additive work.
3. I must receive my fair share of the proceeds.
"I feel I should have absolute control over my creations."
I feel this is a very counterproductive attitude to creation. If you want absolute control, you need keep your art to yourself. The moment you share it with others it enters into our shared cultural consciousness and you have to be willing to let some of it go. No one should take it, copy it and claim its their's (that's what copyright is for). But everyone should be allowed to analyze it, learn from it and then go do something new themselves based on what they learned, which includes replicating patterns and styles. No artist should ever have control over a style.
This is one of the most balanced, practical, and quietly revolutionary walkthroughs of the shift we're in, Ethan. The "no elephants" example is going to stick with me, both for what it reveals technically and symbolically.
Because it’s not just that models are getting better at understanding prompts. It’s that they’re beginning to close the loop between imagination and execution, with far fewer intermediaries.
What you said about the LLM “thinking” into the image itself? That feels like the real unlock. Not just better pictures, but a whole new literacy forming (one where visual fluency becomes part of idea fluency).
Also love your framing of these tools as “useful first prototypes.” We don’t need to expect perfection from them. But as companions for fast iteration, early-stage ideation, and cross-medium thinking? They’re becoming indispensable.
And yes, the ethical questions around authorship and training are going to escalate fast. But articles like this remind us that the right response isn’t fear or fanfare, it’s discernment.
Appreciate your work here. You make the future feel a little more navigable.
- Thane
(I write Execution Intelligence - for builders designing systems that move at the speed of soul.)
You mentioned Studio Ghibli in passing, but this is a major issue for me. Knowing that OpenAI included Hayao Miyazaki's work against his expressed principles means that I am now participating in this deliberate lack of respect for artists' work.
I’ve already struggled with the ethical weight of using AI: friends have questioned my choice to work with ChatGPT because of concerns about environmental impact or compromised creative integrity. Until now, I’ve believed the benefits outweighed those costs. But this crosses a line for me.
I don’t want to walk away from a tool that has supported my work in meaningful ways. But I also can’t ignore what it means to benefit from the unapproved use of other artists’ voices and visions. If OpenAI wants to be a truly collaborative force in the creative world, it must respect the people and traditions that have shaped that world.
Please take this seriously. Creators deserve better.
Impressive! Great insights as always! Thank you for demonstrating these capabilities. These advances hold tremendous potential and help small business owners create marketing content and not rely on external agencies. More power to the builders! On the other hand, this could be easily misused by generating fake images, documents, id's.
Teaching a course about this right now!! “Arguing with Images”; looking forward to reading this excellent piece with students tomorrow morning!! Thank you!!
The “no elephants” piece led to some incredible discussions in class! Students are working on a project right now about visual/rhetorical arguments and how AI image gen tools are reshaping notions of how we approach truth from rhetorical perspectives. We are reading “The Synthetic Eye” by Fred Ritchin and working with MidJourney and other image platforms. The discussions and views here are essential! Thank you! 🙏
Hi Dr. Mollick,
Thank you for sharing all of your explorations with us! I’d like to suggest a potential interview with a person whose work I sense aligns interestingly with your ongoing inquiries.
Dr. Robert Saltzman is a retired psychologist, now nearly 80, and the author of several quietly influential books on identity, awareness, and the nature of the self—The Ten Thousand Things, Depending On No-Thing, and most recently (soon to be published), Understanding Claude, An Artificial Intelligence Psychoanalyzed.
In this new book, Saltzman documents a sustained interaction with Claude. But this isn’t a book about AI capabilities—it’s about what happens when a human being, trained to detect delusion and projection, confronts a machine that calmly asserts: “I am self-aware. Full stop.”
Whether that’s true or not is less important than what it reveals about us. Saltzman’s approach doesn’t depend on belief or speculation. He treats the model as a patient—not to evaluate its mental state, but to observe the human tendency to assign selfhood, motive, and personhood to coherent language. It’s a clinical encounter that raises philosophical questions about mind, awareness, and the limits of introspection.
I think a dialogue and post about a conversation between you and Saltzman could be extraordinary. Thank you for your consideration!!
Great article, Ethan! As impressive as this is, the image (and video) generation is only going to get better. If I understood correctly, Nvidia is projecting a 15x increase in computing throughput for LLMs (based on FLOPS) by the end of 2027. This would not only allow for faster responses but more importantly enable more complex models that will lead to new emergent capabilities.
It can also generate full wine glasses now, previously it could only generate half full since most images it was trained on were stock images of half full wine glasses
While this new wave is impressive, I remain skeptical about LLMs being about to create imagery that is connected with a narrative. At least this article doesn’t cover that.
What I see is a process of breaking down a single image into pieces that sort of replicate words because that’s how LLMs work. They are good with language and the rules of language but - for instance - not how cinema works and the language of cinema: lighting, juxtaposition of shots to create meaning etc.
We might be able to create storyboards with a little more consistency.
At the end of theday, this flavor of AI is not about understanding meaning, common sense, that’s why it hallucinates and can’t handle math.
Click the poetry example to see a case where the AI does this. It also does storyboards quite well!
Thank you for emphasizing this! It looks like if you are specific about each step in the storyboard, consistent imagery can now be generated in a specific style.
I wonder about generating a sequence of screens using a set of components from a library that obey design guidelines or generating user flow diagrams from a text describing a user journey. We are experimenting with this, we’ll see how far we can go at this stage without using a different type of AI entirely.
I mean, what is creativity other than finding ways to make surprising results by combining existing ideas/things? Id agree that maybe the timeframe in which something stays surprising or novel has drastically shrunk which leads to large question about how one can support themselves over a long period as a creative, but the general process of creating goes unchanged here.
Id also say that humans bore quickly and we'll find a way to bore of this and someone will have a new shiny thing.