Thought — 2 Min Read

ChatGPT

by Case Greenfield, December 8th, 2022

Thought — 2 Min Read

ChatGPT

by Case Greenfield

December 8th, 2022

It is happening, as we speak. 2022 seems to be the breakthrough year. Artificial intelligence has entered the artistic realm, as a supporting tool. There’s no way back.

In an earlier story I told you that I have tried DALL-E 2 (by OpenAI) and StableDiffusion (by Stability AI) for idea generation. Now, OpenAI has published a new AI algorithm, ChatGPT. It can be used in a similar manner, but it does not generate images, it generates text.

Here’s an example:

ChatGPT in Art

And another example:

ChatGPT on Mind Models

Both DALL-E 2, StableDiffusion and ChatGPT work well enough for me to definitely start using them as artistic tools, next to brushes, canvases etc.

What is the mind model of ChatGPT?

But there’s something else. I’m wondering, what is the mental model of ChatGPT? Or mind models, for that matter. I read it is build upon 1,3 billion parameters, and that it is unclear what data exactly were used for training the algorithm. Only that they spent about a billion dollar to train it, so the dataset must have been unbelievably huge.

Makes me wonder even more, what ist the mind model of ChatGPT and other LLM’s (Large Language Models)? How does that even work, the combination of AI and Mind Model? Do AI algorithms have something that can be identified as a mind model, at all? An underlying guiding principle, a model of the world?

— (UPDATE December 12, 2022) —

There is a lot of rumor around Generative AI and its impact on creative work these days, and there is some growing critisism on applications such as ChatGPT, and I tend to agree.

Berlin-based professor of Computer Science Dagmar Monett, paraphrasing Richard Dawkins, called ChatGPT “the vacuous rhetoric of mountebanks and charlatans; it is the Neo-Postmodernism Generator, a literally infinite source of randomly generated syntactically correct nonsense, distinguishable from the real thing only in being more fun to read.”

The criticism is not on the way how I use these AI-tools, ie. for idea generation. No, the criticism is on the veracity of what the tool produces. Sometimes it is true and reflects reality, but also very often it is totally untrue what the tool produces. Always syntactically correct – hence creating  the impression of being correct – but often semantically a little up to totally incorrect. Be aware!

It has been said often by others, but again. It is quintessential to keep in mind that IT IS JUST A TOOL, A PIECE OF SOFTWARE, BASED ON DATA. It has no conscience, it does not perceive or experience the world and it does not understand the world, it has no feelings. It just generates words in the statistically most likely order, based on the input words and dataset for training the ML algorithm. That’s it. “The software uses complex machine learning models to predict the next word based on previous word sequences, or the next image based on words describing previous images.”

So. Using these tools as a support for idea creation (eg. ChatGPT) of generating variants (eg. DALL-E2) is at least helpful, although it would definitely be worth finding out what the generated ideas or variants come from ultimately. In other words, in what frame are you being pushed as an artist? What is – or are – the underlying mind model(s)?

And related to that, the really big question is the following.

How ‘creative’ will the outcomes of these tools really be? Will the variation be limited? Or will it really develop a broad and diverse set of new suggestions? I said before, that my first experiences are not unanimously positive. The varivariations of my Study for Poppies tended to converge into these “halucinatory mushrooms”. More of the same.

— (UPDATE January 6, 2023) —

Today, I learned that there is also an AI 3D model generator by Open AI, called Point-E. It isn’t perfect yet, but still. It is a good start. And it will definitely evolve.

So, what does this mean for artists? Well, you can now type a phrase like “a yellow rubber duck” and the system will generate a 3D model of a yellow rubber duck. You can try it yourself. I did.

Here’s a very simple example of what I did. I typed “woman”. and this is what I got:

Piont-E Woman

I’m not saying it is perfect, but, you know … it will improve, for sure.

So, for artists, this means that you can now (or soon) simply write a text, and the system will create a 3D model that you can enter into your 3D design software. You can then adjust and improve the model. Next, you create a real 3D object, like a statue.

Share This Story: