There's been a lot of discussion of dangers that AI will take over the world, turn it into paper clips, etc. Less discussed is the prospect it'll produce a lot of crap that superficially resembles intelligent or creative work, thus pushing standards downward in various fields.
Until it can hold together a metaphor across an entire poem, engage in irony along the way, and play freely with phonetics and rhythm (in the form of alternate spellings, punctuation and whitespacing) to alter interpretations, the poetry from GPT and its ilk will lack substance.
However, I believe the lexical space is a cognitive one, a space for thought; once they have learned to navigate it phonetically, algorithms may well produce true and valuable insights and shed light on relations merely by walking this space in particular constrained ways. The fruit of the tree of language, due to our previous tending, may be particularly easy to juice.without human-level intellect or capacity for understanding.
Here's an example, a gauntlet thrown in challenge to would-be algorithmic poets:
I love this. Years ago, a colleague set up a Twitter bot--low level version of this. He loved watching people trying to debate with his bot. (But unlike you, they didn't realize that they were arguing with a bot.)
Given ChatGPT's insistence on sticking in that "best that I/he can be" line, I wonder if this ditty was part of its training: https://open.spotify.com/track/77zatlvIJXcywmVZb69Bcn (Note: "blank" and "poetic abuse")
For what it's worth, I noticed sometime back that, when asked to write a sonnet, GPT Chat produced a poem that not only wasn't great, but also wasn't a sonnet, wasn't in iambic pentameter, and wasn't even limited to 14 lines.
Here was my experience with it this weekend: My 14-year-old had a physics midterm. His teacher gave him a sheet of vocab to memorize. I made him flashcards. But gathering those definitions was going to be tedious. No sweat: I scanned the page with my iPhone, copied the text from the image, and then loaded 117 terms into ChatGPT. It instantly kicked back defs for each. Sweet. It can’t write poetry, but it can play IP gofer.
Fair enough. Based on my recollection of 8th grade science, they were okay. But I didn’t ask it to do anything with that information. I would imagine if it has problems with basic facts, the problems grow exponentially the minute it tries to put any of those facts in relationship with each other.
I think your use of it to generate the flash cards is actually an excellent experiment. But double-checking the flash cards is essential, and it occurs to me that one thing you can do is have your son double-check the "answers" on the flash cards against the materials he got in class. The double-checking ... doubles (if you will) ... the student's retention, I'm guessing.
At times, the bot is impressive, but when it fails, it sounds like a student who has hurriedly looked something up on Wikipedia, but doesn't actually understand what he's talking about. Example: I asked, "Why are the sun and the moon of the same apparent size?" ChatGPT responded: "The sun and the moon appear to be the same size in the sky because they are both about the same distance from the Earth. The moon is about 238,855 miles (384,000 kilometers) away from the Earth, while the sun is about 93 million miles (149.6 million kilometers) away. Because of the cast distance between the sun and the Earth, the sun appears to be much smaller in the sky than it actually is. The moon, on the other hand, is much closer to the Earth, so it appears to be much larger in the sky. Because the moon and the sun are both the same distance from the Earth, they appear to be the same size in the sky." (I'll be including this in a soon-to-appear Substack column.)
There's been a lot of discussion of dangers that AI will take over the world, turn it into paper clips, etc. Less discussed is the prospect it'll produce a lot of crap that superficially resembles intelligent or creative work, thus pushing standards downward in various fields.
GIGO!
Maybe in several ways, ;-).
Until it can hold together a metaphor across an entire poem, engage in irony along the way, and play freely with phonetics and rhythm (in the form of alternate spellings, punctuation and whitespacing) to alter interpretations, the poetry from GPT and its ilk will lack substance.
However, I believe the lexical space is a cognitive one, a space for thought; once they have learned to navigate it phonetically, algorithms may well produce true and valuable insights and shed light on relations merely by walking this space in particular constrained ways. The fruit of the tree of language, due to our previous tending, may be particularly easy to juice.without human-level intellect or capacity for understanding.
Here's an example, a gauntlet thrown in challenge to would-be algorithmic poets:
https://thekeyunlocks.substack.com/p/auto-poietic.
How long until machines can pick it up?
sadly hilarious! Happy New Year, Virginia!
I love this. Years ago, a colleague set up a Twitter bot--low level version of this. He loved watching people trying to debate with his bot. (But unlike you, they didn't realize that they were arguing with a bot.)
Yes.
Given ChatGPT's insistence on sticking in that "best that I/he can be" line, I wonder if this ditty was part of its training: https://open.spotify.com/track/77zatlvIJXcywmVZb69Bcn (Note: "blank" and "poetic abuse")
For what it's worth, I noticed sometime back that, when asked to write a sonnet, GPT Chat produced a poem that not only wasn't great, but also wasn't a sonnet, wasn't in iambic pentameter, and wasn't even limited to 14 lines.
Which admittedly may not put it far behind the modal English-speaking human.
Its poetry/doggerel is already better than many humans can write, so that can't be the standard.
Here was my experience with it this weekend: My 14-year-old had a physics midterm. His teacher gave him a sheet of vocab to memorize. I made him flashcards. But gathering those definitions was going to be tedious. No sweat: I scanned the page with my iPhone, copied the text from the image, and then loaded 117 terms into ChatGPT. It instantly kicked back defs for each. Sweet. It can’t write poetry, but it can play IP gofer.
If I were you, or your son, I'd double check those definitions. ChatGPT isn't yet reliable about facts.
Fair enough. Based on my recollection of 8th grade science, they were okay. But I didn’t ask it to do anything with that information. I would imagine if it has problems with basic facts, the problems grow exponentially the minute it tries to put any of those facts in relationship with each other.
I think your use of it to generate the flash cards is actually an excellent experiment. But double-checking the flash cards is essential, and it occurs to me that one thing you can do is have your son double-check the "answers" on the flash cards against the materials he got in class. The double-checking ... doubles (if you will) ... the student's retention, I'm guessing.
At times, the bot is impressive, but when it fails, it sounds like a student who has hurriedly looked something up on Wikipedia, but doesn't actually understand what he's talking about. Example: I asked, "Why are the sun and the moon of the same apparent size?" ChatGPT responded: "The sun and the moon appear to be the same size in the sky because they are both about the same distance from the Earth. The moon is about 238,855 miles (384,000 kilometers) away from the Earth, while the sun is about 93 million miles (149.6 million kilometers) away. Because of the cast distance between the sun and the Earth, the sun appears to be much smaller in the sky than it actually is. The moon, on the other hand, is much closer to the Earth, so it appears to be much larger in the sky. Because the moon and the sun are both the same distance from the Earth, they appear to be the same size in the sky." (I'll be including this in a soon-to-appear Substack column.)