One professor's reflections on the end of an era, as AI tools such as ChatGPT have murdered the student essay (RIP). Here's why that threatens the future of human cognition—and how to save ourselves.
There are kids who find exercise soul-crushing vapid toiling too.
Just for some perspective on “what’s good for you.” I personally think I’d have been more successful in life if I was better at essay writing. But I’m not sure if it’s a practice thing, or an innate ability thing. I have to assume I just need(ed) lots more practice and guidance.
I’m also on a similar path right now learning more about programming. AI is helping me understand larger structures, and reinforcing my understanding and use of coding terminology. Even if I’m not writing code, I need to be able to talk about it a bit better to interact with the AI optimally.
But this need to speak in a more optimum way may go away as AI gets better. That’s the thing I worry about, the AI crossing a threshold where you can kind of just grunt at it and get what you want. But maybe Idiocracy is on my mind there.
The problem with AI here is that it tends to prefer agreeing to you over being correct and it’s very likely that it teaches patterns and terminology to you that doesn’t exist.
For example, I just asked ChatGPT to explain a “backflip” in the context of agile development (I claimed I was an university student and that a teacher used that term in the context of moving tickets), and it came up with this:
If your teacher linked “backflip” to moving tickets in a predictable fashion, they might have been emphasizing:
The importance of minimizing rework or regressions.
Understanding why work items move backward (if they do) and making that visible in your process.
Managing workflow policies so that tickets don’t repeatedly “backflip” between stages, which can disrupt predictability.
How This Might Help You Move Tickets More Predictably:
If “backflip” means work moving backward:
Track the Cause of Backflips:
Identify why tickets are moving backward (incomplete acceptance criteria, insufficient definition of done, unclear requirements).
Improve Definition of Ready/Done:
Tighten entry/exit criteria to reduce backflows.
Add Checkpoints:
Build small validation steps earlier to catch issues sooner.
Visualize Flow:
Use a Kanban board to visualize backward movement and analyze bottlenecks or quality gaps.
It just takes the nonsensical word, makes something up, and claims that it’s right.
I have to be carful to not ask leading questions of the AI to much. It’s very happy to go off and fix things that don’t need fixing when I suggest there is a bug, but in reality it’s user error or a configuration error on my part.
Yeah, as soon as the question could be interpreted as leading, it will directly follow your lead.
I had a weird issue with Github the other day, and after Google and the documentation failed me, I asked ChatGPT as a last-ditch effort.
My issue was that some file that really can’t have an empty newline at the end had an empty newline at the end, no matter what I did to the files before committing. I figured, that something was adding a newline and ChatGPT confirmed that almost enthusiastically. It was so sure that Github did that and told me that it’s a frequent complaint.
Turns out, no, it doesn’t. All that happened is that I first committed the file with an empty newline by accident, and Github raw files has a caching mechanism that’s set to quite a long time. So all I had to do was to just wait for a bit.
There are kids who find exercise soul-crushing vapid toiling too.
Just for some perspective on “what’s good for you.” I personally think I’d have been more successful in life if I was better at essay writing. But I’m not sure if it’s a practice thing, or an innate ability thing. I have to assume I just need(ed) lots more practice and guidance.
I’m also on a similar path right now learning more about programming. AI is helping me understand larger structures, and reinforcing my understanding and use of coding terminology. Even if I’m not writing code, I need to be able to talk about it a bit better to interact with the AI optimally.
But this need to speak in a more optimum way may go away as AI gets better. That’s the thing I worry about, the AI crossing a threshold where you can kind of just grunt at it and get what you want. But maybe Idiocracy is on my mind there.
… just some random thoughts.
The problem with AI here is that it tends to prefer agreeing to you over being correct and it’s very likely that it teaches patterns and terminology to you that doesn’t exist.
For example, I just asked ChatGPT to explain a “backflip” in the context of agile development (I claimed I was an university student and that a teacher used that term in the context of moving tickets), and it came up with this:
It just takes the nonsensical word, makes something up, and claims that it’s right.
I believe you and agree.
I have to be carful to not ask leading questions of the AI to much. It’s very happy to go off and fix things that don’t need fixing when I suggest there is a bug, but in reality it’s user error or a configuration error on my part.
It’s so eager to please.
Yeah, as soon as the question could be interpreted as leading, it will directly follow your lead.
I had a weird issue with Github the other day, and after Google and the documentation failed me, I asked ChatGPT as a last-ditch effort.
My issue was that some file that really can’t have an empty newline at the end had an empty newline at the end, no matter what I did to the files before committing. I figured, that something was adding a newline and ChatGPT confirmed that almost enthusiastically. It was so sure that Github did that and told me that it’s a frequent complaint.
Turns out, no, it doesn’t. All that happened is that I first committed the file with an empty newline by accident, and Github raw files has a caching mechanism that’s set to quite a long time. So all I had to do was to just wait for a bit.
Wasted about an hour of my time.