Grumpy opinions about everything.

Tag: AI

Even More Fun With AI

Since I’m easily amused, I’m still playing with the AI image generator DALL-E. I wanted to take some time to show you just how easy this is to use. All these images were generated on the requests shown below. The only ones I’ve modified or edited are the ones in the series to demonstrate an edit. You will usually get two versions with each request. You can save either one or both. You can generate up to 50 images a day. I’ve hit my limit more than once.

I generated the image above using the following command. In a cartoon style, generate images of a man dressed as an artist, wearing a beret and a smock, and holding an artist palette and brush. On the easel, have a computer screen with a picture of the same man.

It only takes a little bit of change to get a completely different feel to your pictures. I created the image below basically by saying in a realistic style as opposed to in a cartoon style.

Generate an image in a realistic style of a man dressed as an artist with a smock and beret, holding a palette and a paintbrush and standing in front of an easel with a computer screen in a messy studio.

Now, let your imagination soar and see what you can come up with. I’m going to make up a few commands now and see what develops.

In a surrealistic style generate images of a young boy with pale green skin and feathers on his arms soaring with fantastical eagles over a fantasy landscape of canyons and forests.

In a realistic style, generate images of an African American woman in a sophisticated business suit and carrying a briefcase. Place her in a whimsical background with fantasy creatures, such as elves, fairies, and unicorns.

In a steampunk style, generate images of a doctor examining a patient with a stethoscope. Have the patient have a third eye in the middle of his forehead.

In an abstract style, generate images of a man and a woman sitting in chairs facing each other. Have the man smoking a cigar and have the woman with purple hair.

In a style reminiscent of the French impressionists, generate images of young people in colorful outfits, dancing on an outside patio.

In a cubist style generate images of baseball players.

DALL-E offers the capability to edit your pictures. However, as you’ll see from the series below it actually generates new pictures with slight variations based on your requests for edit.

In a cartoon style generate images of a young man walking down a sidewalk on a sunny day. Have him in a suburban neighborhood with trees and flowers and well tended lawns and houses.

Change the neighborhood to urban with cars parked on the street and the sky overcast.

So far, the changes seem to be fairly consistent and go along with the general style we requested. Let’s make a couple of big changes and see what happens.

Have him look disheveled and scruffy. Make the neighborhood run down with trash in the streets. Have him carry a whiskey bottle and a cigarette.

I think it’s right on with this one.  It got the feel I was looking for.

Now for my last picture, I’m going to let my imagination run wild and try to come up with these strangest and most bizarre images I can imagine and see if DALL-E matches or exceeds my ideas.

In a dystopian style with a post-apocalyptic background, generate images of an emaciated Santa Claus jumping rope while talking to an Easter Bunny with vampire fangs and a leprechaun with claws. Add any other menacing characters you think may be appropriate.

I couldn’t decide which one I liked best, so I decided to use them both.  I’m not sure what it says about me that I like these images.

I couldn’t resist trying one more edit.  I asked DALL-E to generate the above image in cartoon style.  It sent me this response: I encountered some issues while trying to generate the cartoon-style images. Please let me know if you would like to try a different request or if there’s anything else I can assist you with.  Perhaps it thought cartoon style would attract children and the subject matter was a little too ghoulish for kids.  After tweaking the prompt a couple of times, I got the following image:

Maybe they were right the first time.

Give it a try.  I think you’ll have as much fun as I do.

Doctor’s orders!

More Fun With AI

In my previous post Fun With AI, I talked about how much I enjoy DALL-E, the ChatGPT image generator. But that’s only a small part of ChatGPT, although it is the most fun part.

So, what is ChatGPT? I’ll let ChatGPT explain in its own words: “ChatGPT is an advanced artificial intelligence language model developed by OpenAI, based on the GPT (Generative Pre-trained Transformer) architecture. It processes and generates text by predicting the likelihood of a word sequence, enabling it to perform a wide range of tasks, from holding conversations to writing articles and answering queries. Its training involved analyzing vast amounts of text data, which helps it understand context and nuances in language.”

I hope that’s clearer to you than it is to me. But what I do know is it’s been very helpful to me. I’m just going to go over a few of the things for which I use ChatGPT.

I first started using ChatGPT as a research assistant. I asked it to find me articles about specific topics and suggest references and lines of inquiry. It’s not a substitute for detailed individual research but it is a head start in the right direction. Although you do have to check some of its research because it occasionally gets it wrong or cites outdated material.

One of the things where I have found ChatGPT to be very good is finding quotations on multiple topics. Sometimes it can be difficult to locate specific applicable quotes. (Does anyone else remember Bartlett’s Familiar Quotations?) Again, you must double check the results. Occasionally it will attribute the quote to someone who was referencing the original author.

I’ve also found ChatGPT to be useful in generating title ideas. I’ll ask it to give me 10 suggested titles on a specific subject. While I’ve not used any of the titles exactly, they have formed the basis for developing my own specific titles.

I also use ChatGPT in my genealogy work. Occasionally when researching old records, I’ll find documents that are written in foreign languages or church records written in Latin. Previously I’d used Google translate but found that to be somewhat onerous. ChatGPT translation is quick and easy. I just cut and paste the text into ChatGPT and ask for a translation. You don’t even need to tell it the original language.

I also use ChatGPT as an editing aid. I ask it to review my articles for grammar, readability, and content. ChatGPT has let me know that I frequently use too many adverbs and don’t always get in enough commas. This is a simple process. I take my completed article and paste it in the ChatGPT and ask it to give me an edit.

There are other features that I haven’t used in ChatGPT because I don’t really understand them. One of them is helping with search engine optimization. Apparently, this is a way of ensuring that people searching for topics on the internet can find your website instead of others. While I haven’t taken time to investigate this, I guess I could just ask ChatGPT to explain it to me.

ChatGPT also has a strong program for developing interactive content. This hasn’t been something I’ve been interested in, so I haven’t explored it.

It has the capability to analyze and summarize large amounts of data. I’m hoping to use this at some point to help me with census analysis in my genealogy work. I just haven’t taken the time to develop that ability yet.

I found ChatGPT to be especially useful when I work on this blog. I spend a lot less time on Google. I get more focused searches with ChatGPT, but I do need to be more careful in reviewing the results. The image generation function has expanded my ability to illustrate my blogs. I don’t have to go searching for images in the public domain or worry about licensing fees. I just generate my own images.

There have been a lot of concerns recently about the “dark side” of artificial intelligence. There are plenty of articles in the popular press and I’m not going to dwell on it here. (ChatGPT can generate you a list of as many articles as you would like.)

I just want to express one concern that I have. I worry that it may make me lazy. I may decide not to do my own research, not to read and analyze the available information, and not to spend the time organizing and writing my articles.

I may just tell ChatGPT to generate an article about the use of ChatGPT for writing a blog. Or maybe I already did!?!

Just because I have so much fun with the image generator, I want to share a few of the other images I generated to illustrate this post.

Fun with AI

Depending on what you read, Artificial Intelligence will either be the greatest advancement in the history of humanity or it will be the end of civilization as we know it. I’m not sure if it will be either of those, but I do know I’ve been having a lot of fun with it. I’ve been using chatGPT, a popular form of AI for a little over a year. I first used it to help me research blog ideas and, for the first time, I used it to edit my most recent post. The text feature is what we most often think about when considering chatGPT, and I’ll talk about that in a future post, but today I want to tell you about DALL-E, a feature available on the enhanced version.

DALL-E is an AI powered image generator that creates original images based on user input. I have always thought the banner illustration for this blog was boring and wanted to create a new one. I tried using stock images but you always run into copyright and fee problems. When I discovered DALL-E, I realized it was the answer to my problem. I just didn’t know how much fun it would be.

My first attempt was to simply enter the command: “Generate images for the home page of a blog titled The Grumpy Doc”. These were the responses I got.

I thought they looked pretty good, but obviously I don’t have a beard so I tried a few more. I entered for him to have gray hair and glasses and be clean shaven. I got what looked like a gray haired 14 year-old with a lab coat and a stethoscope. When I said make him look older, it gave him a gray beard. When I said make the beard go away, it turned it brown. I went through a variety of commands for images in a realistic style and a cartoon style with many various other instructions. The more detailed the instructions you give it, the better images you will get. One caveat, it will not generate images of actual people or of copyrighted material. I probably generated almost 100 images for this project. Here are few of those images.

DALL-E has some trouble depicting human emotion. It doesn’t seem to distinguish between grumpy, angry and just plain mean.

When I asked DALL-E for a whimsical or sly smile, I got this down right goofy look.

Being a hefty guy myself, I asked for The Grumpy Doc to be a little heavier:

Please, no wisecracks about how that looks just like me or you may be in my next blog.

Sometimes the images just don’t make sense. In the image below you will see a stethoscope tube that doesn’t connect to any thing, book titles in no known language and a dial with unrecognizable symbols. Look at some of the above images and you will find many of the same types of mistakes. It is almost like a built-in game.

When asking for text or a title you have to check closely. Sometimes it is fine; other times there are subtle mistakes. One of the pictures above has one of those. See if you can find it. Other times you get something like this:

The one human emotion that DALL-E seems to be fine with is confusion or indecision.

Now what’s that all about?

Powered by WordPress & Theme by Anders Norén