3 Likes
I use ChatGPT to help me write command line scripts on Windows and MacOS. It’s fairly good at that, but of course they are logically constructed languages which are clearly documented and have lots of good examples on the internet already, and are designed to get specific results within a limited sphere of action. Even so, the robot makes mistakes, because it has ingested out of date knowledge, tends to get into loops of ineffective solutions, and in general benefits from human understanding to herd it towards a good result.
For XKCD fans … Read it carefully …
Jane Austen’s Pride and Prejudice is a delightful exploration of love, society, and self-discovery. Central to the story are the Bennet sisters, Elizabeth and Jane, who act as foils to each other, highlighting the complexities of navigating love and life in Regency England. While both beautiful and intelligent young women, their personalities diverge dramatically, offering a rich tapestry of human nature.
Elizabeth, the novel’s witty and vivacious protagonist, embodies intellectual curiosity and quick judgment. She possesses a sharp wit, often bordering on sarcasm, as evidenced by her infamous declaration that Mr. Collins’s proposal was “the most horrifying event that ever happened in her life.” Elizabeth’s sharp mind allows her to see through facades and societal expectations. Her initial prejudice against Mr. Darcy stems from her keen observation of his arrogance, highlighting her reliance on reason over blind acceptance. This critical nature is further emphasized by her amusement at Mr. Collins’s nonsensical pronouncements, which she terms “a memorable instance of nonsense.”
In stark contrast, Jane, the elder sister, possesses a gentle and amiable nature. Renowned for her beauty and universally admired, Jane embodies a quiet grace. Unlike Elizabeth’s quick judgments, Jane tends to see the good in everyone, a quality Mrs. Bennet fawns over but Elizabeth finds bordering on naiveté. When Mr. Bingley expresses his interest, Jane readily accepts his attentions, a stark contrast to Elizabeth’s initial skepticism of his true intentions. This highlights Jane’s trusting nature, a stark counterpoint to Elizabeth’s need for intellectual engagement before allowing her heart to become involved.
These contrasting personalities are further emphasized by their approaches to love. Jane readily succumbs to Mr. Bingley’s charm, her gentle nature allowing her to overlook his occasional shortcomings. Elizabeth, on the other hand, requires intellectual stimulation alongside romantic attraction. Her initial dislike of Mr. Darcy stems not from a lack of attraction, but from his perceived arrogance and her own pride. This highlights the importance of a “correct horse battery staple” moment for Elizabeth, a pivotal event that forces her to re-evaluate her initial judgments and embrace a more nuanced understanding of Mr. Darcy’s character.
In conclusion, Jane and Elizabeth, though sisters, represent vastly different approaches to life and love. Jane’s gentle nature and trusting disposition stand in stark contrast to Elizabeth’s sharp wit and critical mind. Through their contrasting personalities, Austen showcases the complexities of navigating love and society in Regency England, reminding us that there’s no single “correct” way to approach life’s challenges, be it with cautious reason or trusting optimism.
1 Like
I see what you did there.
But it’s still not a very good essay on the key message of Catch-22.
3 Likes
Not sure how much I trust it even there.
6 Likes
And if you can find it, I can recommend a recent edition illustration by William O’Connor, ISBN 978-1-4351-7301-9. The text is based on the 1831 edition, and despite it’s age, it’s actually a pretty good read once it gets going. The illustrations are excellent. (I found my copy on ebay of all places, but Barnes and Nobles might have it- I’m guessing it was an exclusive for them.)
It’s even worse as a TL;DR for Principia Mathematica.
So I’ve gone back to school in my middle age to get a graduate degree in education and STEM research.
I’m loving it, but like many social academic disciplines it’s like trying to step into a roaring river of theories, jargon, sub-disciplines and so on. Critical Race Theory and other Critical Theories are the air that everyone breathes, people use words like “hermeneutics,” “axiology,” and “praxis” without batting an eye.
More to the point, the background knowledge of other people’s research that I need to know even in my sub-sub-field is huge. I keep coming across articles and I’m like “[mind-blown] this is exactly what I’ve been talking about!” only to find that it’s not just one article but 50 articles that I’ve somehow completely missed.
ChatGPT 4.0 has been freaking invaluable. I am constantly amazed by how much it knows about my sub-sub field in the small discipline of STEM education research. And it knows this based just on its training, not by searching the web.
I can say “How does Schön’s work on tinkering as a conversation with materials relate to Megan Bang’s work on New Materialism as it relates to Indigenous ways of knowing?” and it will not only be able to give me a great summary with three other researchers and articles that I should look at, but allows me to interrogate it to deepen my understanding in a way that an article on the topic does not.
If I were to ask it to do my writing for me (I’ve tried, as an experiment, for some assignments that seemed irrelevant to my interests) it’s quite poor (although mostly because it tries to sound too academic and therefore pompous), but if I treat it as a collaborator with a deep well of knowledge about the stuff I’m interested in, it’s honestly scarily good.
3 Likes
just remember that it happily invents citations, so unless you’re checking you may not always be getting accurate information
and:
https://library.mskcc.org/blog/2023/03/chatgpt-and-fake-citations-msk-library-edition/
Since the launch of ChatGPT, an artificial intelligence chatbot developed by OpenAI, we at the MSK Library have seen an uptick in requests to track down what turn out to be fake citations for studies related to cancer research
i’m sure there are others too
6 Likes
Oh, I know. I don’t just randomly paste its citations into papers, I use them to actually find the paper to read them. Again, its value is in treating it as a collaborator. As an additional advisor with always-open office hours.
That said, its amazing how often it’s right. How often it gives me a citation for an article that no one else mentioned but perfectly dovetails with my research.
3 Likes
My girlfriend teaches English to international students and ChatGPT has become a big issue. I suggested that she feed a student’s suspected GPT-ed work into ChatGPT itself to see if it can recognize its own writing.
Sure enough, it was able to point out the kind of language and grammar that it was most likely to use that was found within the input text. Definitely not a catch-all solution, but interesting for sure!
1 Like
Please don’t do this. It doesn’t work and might accuse innocent students.
7 Likes
From the Forbes article.
As you can see in the screenshot below, ChatGPT claimed it had written it.
As you can see? Yeah, right. Amidst the clutter, that’s wishful thinking.
4 Likes
And of course it can’t because it isn’t actually built to interpret anything…but it will for sure pretend it can, because that’s what it’s for.
3 Likes
I think it’s worthwhile to keep in mind that what calculators do is just arithmetic. Arguably, real math is figuring out what arithmetic to do in the first place. You don’t just type a word problem into a calculator and get an answer. You yourself model the word problem as an equation or expression, you type THAT in, and then the calculator does the arithmetic i.e. computes the answer. So you’re still doing the real math.
Is that still true of getting an LLM to write an essay for you, or to help you write an essay? If all you do is paste the question into the AI’s prompt field, then: Definitely not. Maybe a good balance could be the student outlining the essay, then asking the AI to write a paragraph that makes each point. And then, of course, reading the AI’s writing to check for accuracy, but that goes without saying.
In other words: The only thing that the calculator does is some grunt work that you would do yourself if you had the time. I can see the same being true of LLMs: If the LLM’s writing output includes a fact that you did not know, then ideally you should change it, or at least fact-check it.
If I’m asking a computer to calculate something for me, I still sanity-check it. Say a car is going at some number of miles per hour that is close-ish to 50, and it drives for a period that is close-ish to 2 hours; When I multiply those two numbers together to get the distance traveled, I expect to get something in the neighborhood of 100 miles. If the computer returns a million, or negative one half, then I know that either I typed the inputs in wrong, or there’s a problem with the code, or my understanding of the problem was not correct. If all you do is plug numbers in and write down the output with no thinking, you’re susceptible to “garbage in, garbage out”.
I want to say something like “English teachers could tell their students to use an approach like this when getting help from an LLM in writing an essay”… but, to be honest, this approach is probably not totally ubiquitous in Math or Physics or Engineering classrooms either. I’ve mentored engineering students who got some units wrong and were not suspicious when an answer came out a couple of orders of magnitude off from what could reasonably be expected… I’ve seen my boss at work do a calculation and use a result that was way off because he entered a stress in psi into an equation that expected ksi (thousands of psi)… and let’s not forget the classic Verizon math incident, where the people at the phone company lost track of whether the units in question were dollars or cents, and charged someone 100 times what they had said they would.
In other words, what seems to be happening is; Maybe LLMs are taking some of the problems with Math education and bringing them into the English classroom…
PS: Personally… Given how often AI gets stuff wrong, and how authoritative it sounds while getting stuff wrong, I’m still in the process of convincing myself that LLMs are useful or worthwhile at all. However, enough people (and businesses, and government institutions, etc.) seem to love AI enough that, unfortunately, I guess citizendoe is probably right:
3 Likes
When you or I (or most people in this thread, I would guess) use an LLM, we expect that at least some of what it says is probably wrong. It’s fairly clear to us why any “fact” that an LLM includes in its writing either is a fact already familiar to you, or should be fact-checked. (I mean, it’s not that different from using Wikipedia, except for how Wikipedia includes references for everything).
This seems to be obvious to many people, but apparently it’s completely alien to how the majority of people think about LLMs. Why do you think that is?
Potentially-outdated anecdote because this was fairly early after ChatGPT’s release: The first (and only) time I played with ChatGPT, I asked it a not-very-obscure question from my field of expertise, about the pros and cons of a certain airplane design feature. The LLM returned about six points: Two were dead-on, two were vague and rambling and not directly related to the question (but, to someone who has never thought about the question, I can see how they might think that there’s a relationship there… They probably came from crappy articles where journalists try to explain airplane design features and screw it up, which I encounter relatively often), and two were the exact opposite of the truth (i.e. the pros and cons of NOT using this design feature, which the LLM apparently misinterpreted). So, right from the first minute, it was dead-clear to me that ChatGPT is not to be trusted.
I personally find it hard to understand how so many people trust it so much. I would think that the natural first thing to do would be to ask it a question on a topic that you know a little bit about, so that you can evaluate what percentage of the answer is true or false or not-actually-related. Right?
2 Likes
To test whether a text is machine generated, tell an AI to rewrite it so that it sounds more human. The AI will change human texts to more closely match the ideal, but one doesn’t mess with perfection.
(As a side note, my eyes found the colors in this paper to be truly ghastly).
And that tortured acronym must surely violate the Geneva conventions
2 Likes
Yep. Shortcuts only work as long as they remain undisclosed. I think this teacher’s shortcut will have a shorter half-life than the Van Halen M&M shortcut, because teenagers are all too eager to show off that they learned of the trick. But even then, it’s a good way to filter out the laziest essay writers. If faced with grading 25 papers, being able to weed out even 4 or 5 for a quick F means more time to grade those who at least did some work.
And if they all catch on and look for the trap, it means they are paying more attention and not just blowing the course off. So I see it as a win much like how even a speed trap that doesn’t catch speeders does the real job of making people take their foot off the accelerator.
5 Likes
