Three-panel banner: coder at desk, neon AI brain

QT: illustrated haiku folks trust computers AI hallucinates facts garbage in, spewed out My technique to make the haiku image. I made 3 images in flux 2 pro using txt 2 img with a prompt including prompted text. Then in Img2Img, I put those 3 images in as start imges and added this prompt: Make the images 3 seperate horizontal layers with image 1 as the top layer, image 2 as the middle layer, and image 3 as the bottom layer. Show the text in each layer and a rendition of each image as much as possible I also did a blog post showing how to use the Img2Img https://budgetpixel.com/blog/using-imgtoimg-to-create-a-new-image @DavidP kicked of the illustrated haiku QT with this post https://budgetpixel.com/p/15602 the old adage "garbage in, garbage out" still holds for AI as well. In my opinion, to get the AI up to speed, I think they threw everything in. Now I think the different companies are trying to improve the quality of the input so that the output improves. Remember when we used to get 6 or 7 fingers all the time? Now we get 5 fingers almost all the time. With data, sometimes patterns present facts as true that are not true. There is still a lot to be done in that regards but I do think it is improving. In this image, I asked the AI to give me an image showing my vocabulary words : "nacreous antechinus embroidery". Oldest or most lightweight models only gave me an image. Some gave me an image and text. Few gave me an image and the definitions and they were the latest models. Spelling on some was bad and some was good. Spelling definitely improved with newer models. The same newer models can give a good image, and correctly spelled words giving a correct definition and also an image with an hallucinated false definition. Here is one hallucination. https://budgetpixel.com/p/15928 There's a saying, "Trust but verify." I'm thinking "Check, verify, then maybe trust." Yeah, I'm paranoid but that's ok.