I really enjoyed this post. The poem about censorship should worry a lot of poets, but perhaps it will just set the bar higher like Tiger Woods did for golfers.
Could you share the prompt you used to generate the poem?
I'm working thru a book you might be interested in "Teaching with AI". by Bowen and Watson. I recommended it to my friend who is on the school board.
Here's a quote from the book:
"Those who can collaborate and think with Al will gradually replace those who can't. We will need to prepare students for this new era of human thinking and new definitions of the knowledge economy."
I don’t know HOW AI will change things. But it WILL change things. For the source of the poem, open X/Twitter. Over on the left, there’s a column of icons. Near the top is a square with a diagonal slash in it. That’s Grok. At the bottom of the screen, there’ll be a box where you enter a prompt. Use it like Google to ask questions. Or tell it to write a poem on censorship. Or tell it to produce a photo of George Washington with a big smiler on his face. Ask it for recipes. You can use it in lots of different ways. And is FAST!
Absolutely. I'm not an accomplished poet, but I could do a more nuanced prompt. I just typed, "Write a poem on censorship," and this instantly rolled out. I was surprised at what a passable job it did with such a spare query in so little time.
I thought it was surprisingly adept with metaphors. They're obviously imitations, but the same can be said of literature in general. It has been quite a while since the words, "There is nothing new under the sun" were committed to parchment.
I could certainly craft more elaborate queries, as I regularly do with my visual AI requests. Or, I could take this poem and do some editing to improve it considerably.
So, I asked Grok "How do you know how to create a poem?". The reply was very interesting. It explained how it's training made this possible.
I asked ChatGPT the same question, it was less forthcoming, but with a little prodding it related some of the same information.
Interesting, Chat may not know what data is was trained on, so it will synthesize the answer (because it is obsequious) to give you something.
You can tell it to answer questions without "synthesizing" information. This results in a limit on what it delivers. Depending on what you're after this could improve the reliability of it's answers. This also gives you some hint of how much data on the subject was in it's training.
These LLMs are statistical entities that are dependent on the data they train with. But it while more measurements improve statistical results, it doesn't appear that this is the case for LLMs.
I'm working on a Wright Brothers wind tunnel problem. From what I can tell so far there isn't enough data in the training about the wind tunnel, so Chat synthesizes the output. Which in this case is not useful.
LLMs are most useful to someone who knows a fair amount about the subject they are working on, so they can sort out the wheat from the chaff.
Ouch, time to play Santa and get the gifts out - don't tell the kids...
Fascinating. I'm working on a near-future piece and was generating illustrations of all 45 presidents. Grok was superb at doing 40 of them. Did splendid pictures of some of the obscure ones. But it has absolutely no idea what Zachary Taylor, Franklin Pierce, Rutherford B. Hayes, James Garfield, and Chester Arthur look like. As for Zachary Taylor, and it gives you young, thin, bearded officers. I was finally able to do a passable Hayes and Garfield by describing features. (He has a long beard. He has a high forehead and receding hairline, etc. The other three were hopeless. I kept giving instructions that it ignored. "Chester Arthur does not have a beard." "Zachary Taylor has gray hair." I'd give the instructions, and it would just give me the same ole pics that looked wrong. When I told it how bad they were, it both thanked me and got defensive, telling me that it has given me historically accurate images. Remarkably, in the case of Arthur, it named some specific archival photos and then gave me images that looked nothing like those photos.
I tried Chester Arthur in ChatGPT - no joy, even after prodding. Am image of Ronald Regan was just fine.
I'd suggest that the number of images available makes a difference. Humans have a section of the brain dedicated to facial recognition and but LLMs are dependent on enough images in the training.
Stunning how smiling Lenin and smiling Rasputin convey so much sinister, if not evil, contemplations... harbingers of things to come... stuff of nightmares.
I loved the selfies so much I stole them and posted them on my FB. (I of course credited you).
And it's the smiles on peole we know to be awful that make them look "uncanny" as if they're not really human.
This is a terrific Christmas gift.
As for poetry, I barely read any of it any longer because so much of it reads like the old "Mad-libs." It' has been a long while since I read any that was actually uplifting, or even very interesting. That's just sad.
Thank you for this! Loved Mondrian and Klimt, what would AI have done with Pablo Picasso? I keep trying the Adobe version of AI with automobiles and recognizable local landmarks, it has a long way to go.... Happy Holidays! When I had an Echo, my friends shared the SNL skit with me, my Echo was sold back to Amazon...
A wonderful Christmas gift! Thanks to both of you who clearly share one mind, the two fused by the magic of marriage.
Thanks! Our marriage is, indeed, magic. Always has been. Nearly a billion and a half seconds together, and not a bad one yet.
I really enjoyed this post. The poem about censorship should worry a lot of poets, but perhaps it will just set the bar higher like Tiger Woods did for golfers.
Could you share the prompt you used to generate the poem?
I'm working thru a book you might be interested in "Teaching with AI". by Bowen and Watson. I recommended it to my friend who is on the school board.
Here's a quote from the book:
"Those who can collaborate and think with Al will gradually replace those who can't. We will need to prepare students for this new era of human thinking and new definitions of the knowledge economy."
I don’t know HOW AI will change things. But it WILL change things. For the source of the poem, open X/Twitter. Over on the left, there’s a column of icons. Near the top is a square with a diagonal slash in it. That’s Grok. At the bottom of the screen, there’ll be a box where you enter a prompt. Use it like Google to ask questions. Or tell it to write a poem on censorship. Or tell it to produce a photo of George Washington with a big smiler on his face. Ask it for recipes. You can use it in lots of different ways. And is FAST!
The key is the prompt you used. Knowing what to ask for and how to ask for it is the key.
It's like an artist knowing which brush, which color and which technique.
Could an accomplished poet create a better more nuanced prompt?
Absolutely. I'm not an accomplished poet, but I could do a more nuanced prompt. I just typed, "Write a poem on censorship," and this instantly rolled out. I was surprised at what a passable job it did with such a spare query in so little time.
I thought it was surprisingly adept with metaphors. They're obviously imitations, but the same can be said of literature in general. It has been quite a while since the words, "There is nothing new under the sun" were committed to parchment.
I could certainly craft more elaborate queries, as I regularly do with my visual AI requests. Or, I could take this poem and do some editing to improve it considerably.
So, I asked Grok "How do you know how to create a poem?". The reply was very interesting. It explained how it's training made this possible.
I asked ChatGPT the same question, it was less forthcoming, but with a little prodding it related some of the same information.
Interesting, Chat may not know what data is was trained on, so it will synthesize the answer (because it is obsequious) to give you something.
You can tell it to answer questions without "synthesizing" information. This results in a limit on what it delivers. Depending on what you're after this could improve the reliability of it's answers. This also gives you some hint of how much data on the subject was in it's training.
These LLMs are statistical entities that are dependent on the data they train with. But it while more measurements improve statistical results, it doesn't appear that this is the case for LLMs.
I'm working on a Wright Brothers wind tunnel problem. From what I can tell so far there isn't enough data in the training about the wind tunnel, so Chat synthesizes the output. Which in this case is not useful.
LLMs are most useful to someone who knows a fair amount about the subject they are working on, so they can sort out the wheat from the chaff.
Ouch, time to play Santa and get the gifts out - don't tell the kids...
Merry Christmas to ALL and to ALL a GoodNIGHT!
Fascinating. I'm working on a near-future piece and was generating illustrations of all 45 presidents. Grok was superb at doing 40 of them. Did splendid pictures of some of the obscure ones. But it has absolutely no idea what Zachary Taylor, Franklin Pierce, Rutherford B. Hayes, James Garfield, and Chester Arthur look like. As for Zachary Taylor, and it gives you young, thin, bearded officers. I was finally able to do a passable Hayes and Garfield by describing features. (He has a long beard. He has a high forehead and receding hairline, etc. The other three were hopeless. I kept giving instructions that it ignored. "Chester Arthur does not have a beard." "Zachary Taylor has gray hair." I'd give the instructions, and it would just give me the same ole pics that looked wrong. When I told it how bad they were, it both thanked me and got defensive, telling me that it has given me historically accurate images. Remarkably, in the case of Arthur, it named some specific archival photos and then gave me images that looked nothing like those photos.
Merry Christmas to you, too.
I tried Chester Arthur in ChatGPT - no joy, even after prodding. Am image of Ronald Regan was just fine.
I'd suggest that the number of images available makes a difference. Humans have a section of the brain dedicated to facial recognition and but LLMs are dependent on enough images in the training.
The Great Alexa parodies. Having lived in Minneapolis for three years, this is one of my favorites: Midwest Siri, also known as “Sheryl Lipinski.”
https://youtu.be/l_8n2q5iI4E?si=qfs5jpbEtwguwRIK
You betcha! That was great. Especially the opening, where she urges him to call 911. :)
I really liked those! Thank you to you and Alanna!
You’re welcome!
Stunning how smiling Lenin and smiling Rasputin convey so much sinister, if not evil, contemplations... harbingers of things to come... stuff of nightmares.
I loved the selfies so much I stole them and posted them on my FB. (I of course credited you).
And it's the smiles on peole we know to be awful that make them look "uncanny" as if they're not really human.
This is a terrific Christmas gift.
As for poetry, I barely read any of it any longer because so much of it reads like the old "Mad-libs." It' has been a long while since I read any that was actually uplifting, or even very interesting. That's just sad.
Steal as you wish! And thanks for the credit. And yes, the creepy smiles are telling.
Thank you for this! Loved Mondrian and Klimt, what would AI have done with Pablo Picasso? I keep trying the Adobe version of AI with automobiles and recognizable local landmarks, it has a long way to go.... Happy Holidays! When I had an Echo, my friends shared the SNL skit with me, my Echo was sold back to Amazon...
Glad you enjoyed it! I've done a number of similar projects with Picasso. AI tends to get him quite well. Happy holidays to you, too! I loved the echo video ... never, ever wanted to own one. Always have the following meme in mind. :) https://media.cheggcdn.com/study/c71/c71e14d2-7e96-4854-a257-20bdd774b562/image