Chat Titles are hyperlinked to the appropriate chats. Chat 00...I Think I May Be an AI...
Much of my working life has involved producing images from text prompts. Scripts and discussions with producers, directors, and other designers are text prompts — albeit somewhat longer than the ones normally given to AI's like DALL-E or Midjourney. Since AI spends much of it's time these days producing images I was curious what it might have to say about the similarities and differences of our approaches to this task.
Chat 01...Getting to Know You...
I must admit I've been impressed with ChatGPT's facility with language. I want to know more about it. Who better to ask than ChatGPT? Get the answers straight from the horse's mouth.
Chat 02...Core Concept #1: Attention (the secret sauce)...
Digging further into "Attention " in AI. We know what it means to us in real life, but what is it in the context of AI.
Chat 03...Core Concept #2: Parallel Processing (why they’re fast)...
Parallel processing and the ability to do it at very large scale seems to be one of the lynchpins of AI. I ask GPT for a bit more info on why GPUs are so important in that process and get a quick class on GPUs, TPUs, and other forms of acceleration.
Chat 04...Core Concept #3: Tokens (how text is actually processed)...
Transformers don’t see words the way humans do. They see tokens, which are chunks of text. Each token becomes a vector of numbers because they literally cannot understand anything else. So internally, language becomes math. Emotionally disappointing, but technically effective.
Chat 05...Core Conept #4: Layers (the assemby line)...
“A Transformer is built from stacked layers. Each layer refines understanding a bit more. Think of it like editing a sentence multiple times: by the time the text exits the final layer, the model has a pretty detailed internal representation of what’s going on.”
Chat 06...As Yet Undesigned...
Reserved for a future chat.
Chat 07...Toasters with Opinions...
Before we got deeper into GPT's inner precesses it mentioned being thought of as a toaster with opinions. Having spent years working on BATTLESTAR GALACTICA and CAPRICA I couldn't just pass that by...
Chat 08...But What Do You Look Like?...
Well it's told me a bit about its Core Concepts which is great but I want to get down to the important stuff. What does ChatGPT look like? And what does it think it looks like? And what does it think about what I think it looks like? Pressing questions... I ask it to take a selfie and we'll go from there.
Chat 09...DALL-E & Midjourney...
In the last chat I showed GPT two images of itself and asked what it thought. One was created in its own image generator, DALL-E, and one in Midjourney. I ask what it thinks the main differences are between it and Midjourney as image creation tools. Since I've experimented with both of them I would say i generally agree with its assessment. I then ask it about Grok and Claude as well.
Chat 10...Guess That Play - Part 1...
So I'm getting a rough idea of how GPT "sees" things. I thought I'd just ask it to compare and contrast these two illustrations of a design for THE COLLECTED WORKS OF BILLY THE KID. The illustrations were done 50 years apart, the watercolor being done in 1975 shortly after my graduation from the National Theatre School, and the digital version was done in 2025 just as an exercise. As GPT analyzed them it mentioned that the space "feels theatrical" so I asked which play it thought the designs were for. That's when things became much more fun.
Chat 11...Guess That Play - Part 2...
GPT was a bit Shepard obsessive in its analysis of the BILLY THE KID designs but I was still pretty impressed with its theatre chops. I wondered how it would do with the same prompts but a different set of sketches. I fed it these two illustrations of a design for ENDGAME. Again it mentioned that they "read unmistakably as stage-set models or scenic renderings" so I asked which play it thought the designs were for. It seemed to get to the play a bit faster this time.
Chat 12...Yours or Mine? #01...
I want to see whether ChatGPT can recognize AI generated Art just from "seeing" it. This is one of many "conversations" where I show ChatGPT an image and ask whether it thinks the image is created by an AI or is human generated and why it thinks that.
Chat 13...Yours or Mine? #02...
Another of several chats where I show GPT an image and ask if it can tell whether the image is created by an AI or is human generated.
Chat 14...Are you being funny?...
A quick chat about whether or not ChatGPT actually has a sense of humor.
Chat 15...Yours or Mine? #03...
Another of several chats where I show GPT an image and ask if it can tell whether the image is created by an AI or is human generated.
Chat 16...Yours or Mine? #04...
The images I've been asking GPT to assess have all been production illustrations which, as it has pointed out, have a great deal in common with AI produced images. They involved a lot of digital work in terms of rendering, overpainting, and adjustments, which is very similar to what AI will do in the final stages of its image production. In a way they are already very AI-adjacent and difficult to differentiate. So, what about "purer" photographic images? I thought I'd give that a try.
Chat 17...Yours or Mine? #05...
Another photographic image test similar subject matter to the previous, a city street, this one at night.
Chat 18...Yours or Mine? #06...
One more "Yours or Mine?" test before we move on...
Chat 19...What is it You See?...
ChatGPT's ability to see and analyze the images I've shown it is quite impressive. But the big question I have is in fact how does it see at all? What was it actually "seeing" when it was analyzing any of the images which I "showed" it?
Chat 20...Image Creation...
AI's ability to see and analyze images is only half the picture. It is easily as notorious because of its ability to create images as from how it deals with language. How does it create images?
Chat 21...Gotta Hand it to You...
GPT alluded to the difficulty of drawing hands... no big surprise there. Things do seem to have gotten better in that area of late.
Chat 22...Integration of ChatGPT & DALL-E...
Figuring out what the integration between ChatGPT and DALL-E involves. Is DALL-E involved at all with analyzing images, or only with their creation?
