A perfect Mona Lisa

I recently followed Neural.love’s guide to drawing the Mona Lisa with ChatGPT. Like most of things I tried in the days following the release of ChatGPT, the experience (further) opened my eyes to applicability of AI.

I chose the best (in my opinion) of the output Mona Lisas that ChatGPT produced in that fateful prompt session (working title: Limone-a-Lisa, a reference to its yellow color) to serve as my website’s unofficial mascot.

I’ve always been almost shamefully trusting of the proverbial algorithm in my personal life, especially in context of media consumption, like Youtube and Spotify (as are we all). It seems only fitting that the logo to my compendium of thoughts towards the world should also a product of the algorithm.

What is ChatGPT and how does it work?

By now, most people know this already. That being said, ChatGPT is a large language model trained by OpenAI. It’s based on the GPT-3 architecture, which uses a transformer network to process and generate natural language. ChatGPT is trained on a massive amount of text data, which allows it to generate human-like responses to a wide range of prompts.

What’s so great about the image?

There are several aspects of this particular image that made it stand out from the rest of the model’s output on that fateful afternoon last December.

• The framing, i.e. outer black box, outline of the face and yellow background, which were all provided by the user prompt, are all very well represented
• What I assume is a representation of Mona Lisa’s nose (similar to a rotated ‘3’) is reminniscant of a “cat face” smiley, i.e., “:3”.
• The other feature of the face, besides the eyes, is the interesting curve which almost looks like a misplaced mouth. I assume this is ChatGPT’s interpretation of a “misterious smile” as instructed by the prompt.

These aspects, and more, all contribute to the quality of this image.

Is it actually a representation of the Mona Lisa?

The outputs of large language models can be seen as unique and original interpretations of the input they are given, similar to the way a human artist might approach a subject. Does GPT-3 inherently know what the Mona Lisa is? Likely not. More likely, it know that the prompt is about the Mona Lisa, or rather, it does know that the prompt is about a painting, and it knows that the painting has some specific features it can replicate. It’s up to the view (us) to interpret the output as a representation of the Mona Lisa. In that sense, the output would be an original work, in a sense.

On the other hand, the argument still stands that the output of large language models is not truly original or creative, but rather a reflection of the data they were trained on. Large language models have no real understanding or knowledge of context which would mean that any output lacks meaning in the same sense that .

So it’s not a work of art?

Who knows. In any case, I like it.