Not so long ago, “Magic Mirror” builds were all the rage, and we have to admit getting out daily reminders and newsfeeds on an LCD display sitting behind a partially reflective mirror is not without its charms. But styles ebb and flow, so we don’t see too many of those builds anymore. This e-ink daily calendar reminder hearkens back to those Magic Mirrors, only with a double twist of AI.
This project is the work of [Ilkka Turunen], and right up front we’ll say the results are just gorgeous. A lot of that has to do with the 10.3″ e-ink display used, but more with the creative use of not one but two machine learning systems. The first is ChatGPT, which [Ilkka] uses to parse the day’s online calendar entries and grab the most significant events to generate a prompt for DALL-E. The generated DALL-E prompt has specific instructions that guide the style of the image, which honestly is where most of the artistry lies. [Ilkka]’s aesthetic choices, like suggesting that the images look like a 19th-century lithograph or a satirical comic from a turn-of-the-(last)-century newspaper. The prompt is then sent off to DALL-E for rendering, and the resulting image is displayed.
It has to be said that the prompts that ChatGPT generates based on the combination of [Ilkka]’s aesthetic preferences and the random events of the day are strikingly complex. The chatbot really seems to be showing some imagination these days; DALL-E is no slouch either in turning those words into images.
Like the idea of an e-ink daily reminder but prefer a less artistic presentation? This should help.
I do AI research as a hobby now. I’ve got too much OCD now to use any of the AI generated stuff – I keep seeing the flaws, they jump right out at me.
The left rear wheel on the car above, or the position of the sun relative to the car shadow, or the chimney shadow.
The image from the project site is like something out of a lovecraftian horror poster. Bowl od soup and cup of tea on the ground next to the table, man wearing rolled up trousers and knee high socks, boy wearing man’s hat, weird pumpkin flowers some of which are misshapen, variable numbers of fingers on various hands, weird things under the clock, ghoulish people in the shop front, woman in front of shop sitting astride a table.
Note the sun shade overhang on the shop: one end hangs straight down, the other end is curved.
The amount of information we receive through any of our senses is astonishingly little – on the order of a couple of bits per sense, with maybe 10 or 20 bits per second per sense. With these limitations, it’s easy for a program to fool a human into thinking that a synthesized image is OK – we don’t really look at the image in any detail.
https://en.wikipedia.org/wiki/The_Magical_Number_Seven%2C_Plus_or_Minus_Two
The same is true of ChatGPT: tell it to summarize a book or movie, and see how accurate it is: Depending on circumstance, the results it gives can be hilariously wrong.
And note that kids are using ChatGPT to help them with their homework even as we speak…
Nobody is suggesting you trust it with heart surgery.
It’s pretty good at providing code to do things you don’t normally do though, it’s imperfect, like stack overflow, but it can save a ton of time and you just need to validate it like a Jr dev wrote it.
😒 you’ve talked about human perception then linked to a Wiki article on Miller’s Law which is about short-term memory, a completely different issue. Memory and perception are obviously different concepts and someone that purports to be obsessive about precision should’ve noted that 🤔
It’s strange that you claim your OCD has such a profound affect on you but only list very obvious, easily spotted flaws in the AI artworks: any type of human made art with abstraction or inaccurate representation must be awful for you.
Stranger still is the idea that kids using ChatGPT to cheat on their homework with lacklustre results because of it’s limitations “even as we speak” is somehow a bad thing. Kids have been cheating on their homework since ten minutes after homework was invented, your argument seems to be that ChatGPT isn’t accurate enough so wouldn’t that mean kids can’t use it to cheat very well?
You should apply the same critique to human generated art. I think you will be in for a surprise.
You took the time to leave a reply on the tech write up…. And your entire reply was criticizing the artistic output.
You skipped the code.. the integration.. the awesome use of two distinct platforms to create a very unique calendar and instead simply pointed out how good YOU are at pointing out flawed things.
Congratulations, I guess?
The stink of envy is real.
Yep, that one is an awesome project… I wonder if you could use your 24h to run something like stable diffusion locally to avoid the monthly bill… anyone know if it would be possible on a rasp pi 5?
That is a great idea! It seems like people have made it work – you possibly could a small version of llama alongside it to generate the prompts. Might be another weekend coming my way
It would probably take a month to produce each image on a Pi, even if it were a 5.
I’ve run stable diffusion on my M1 Max MBP, it’s fast enough to be interesting.
Best to run it as a service on some hardware that’s moderately capable.
This is a great suggestion – I think in the time frames it’s doable. Seems like there’s a project https://github.com/vitoplantamura/OnnxStream that could achieve this. Combine it with a small implementation of llama and i think you might have a winner! I think you’ve just inspired another weekend project
Thank you!
You’ve perfectly explained what’s been bothering me about so many of these critical comments here on HaD– using criticism just to toot their own horn– which I’ve been unable to put into words myself.