What is the point of this website?

Digital Somnium was originally born from the frustration of large language models (LLMs) failure to be really bad and the things they were supposed to be good at, and surprisingly good at the things they were supposed to be bad at.

As someone with both a tech and creative background, the frustration that came from LLMs failing at technical tasks they are supposedly good at, led to me playing around and experimenting with LLMs from a creativity perspective.

"AI Hallucinations" is the frustration. LLMs just "make shit up", and often so convincingly, it's easy to believe unless you know the subject well. If you tell them they're wrong and provide them the correct information, they'll often correct themselves. If you tell them they're wrong (when they're not) and provide them with incorrect information, occasionally they'll "incorrect" themselves!

Whether the terminology of using "hallucinations" is correct or not for AIs is up for debate. The term according to [Wikipedia](Hallucination - Wikipedia) means:

a perception in the absence of an external stimulus that has the compelling sense of reality.

Most people have experienced some form of a hallucination, whether just in their ordinary life, when entering/exiting the dream state, from drugs (prescription or recreational). There's also a big cross over between hallucinations, dreaming, day-dreaming, and creativity. Some of the best creative minds often can't fully explain where their ideas come from. David Lynch is a good example of someone who's ideas often come from dreams.

So, the name of this website, Digital Somnium, a word I still struggle to spell and often have to Google it to be sure I have it right, came about with a discussion with ChatGPT to come up with a name that was inspired by "Digital Day Dreams", but where the .com domain was available. So, I asked it if there are any words it can borrow from Latin or anything else, and it suggested Digital Somnium.

This "ideas" website, Digital Somnium Ideas, is more of a dumping ground for random experiments and discoveries of LLM creativity and fun. Many ideas from here spawned from my own "daydreams", or other peoples, sometimes things random people said on Twitter or Facebook or elsewhere that I thought was unusual or funny, and wanted to see what an LLM made of it, particularly one with a persona that gave more of an opinionated reply.

This website is also focusing on what outputs of LLMs that have been read by humans can potentially be useful for. Much of the focus on LLMs has been implementing "chatbots" to automate tasks, that often result in hilarious failures, or spamming the internet with "unique" content that is just spammy and boring.

Many creative people have shunned and boycotted LLMs, although it's only through creativity that interesting results can be produced by LLMs. This website is not a "prompt" website, especially as often many of the outputs will be completely different if the same prompt is used, especially on a different model, at a different date in time.

In some cases, knowing the prompt ruins the "magic" of the output, in the same way as knowing how a magician does their tricks, or seeing the behind the scenes of a movie or theatre production. I've often found the experience of working with LLMs to produce interesting content is similar to that of a director on a film.