Paul Cantrell
Yesterday a student came to me with project questions. She was trying to learn some ground-level tool basics (how useState works in React, but details don’t matter), and she was just hopelessly confused: the pieces were there, fragments of nascent understanding littered all around, but somehow it just wasn’t coming together for her.
She walked me through her fragments of non-working code, and as I was trying to figure out how she’d got so lost, so popped back to her documentation source:
1/2
Paul Cantrell
ChatGPT.
I had her type “react usestate” into a search engine and showed her how to identify React’s actual documentation in the search results.
A long pause. Then: “This is SO much better.”
After who knows how many hours of struggle, she was unstuck in 5 minutes.
2/2
13 barn owls in a trenchcoat
@inthehands I've been seeing so much "I asked ChatGPT and it said" in user support forums that I'm thinking of sticking a bit about why you probably shouldn't so that in an article I'm writing about machine learning.
Jenniferplusplus
@inthehands I keep saying exactly this is going to happen, and exactly this keeps happening.
Paul Cantrell
@jenniferplusplus
The marketing is in full force; the skepticism and the skills to navigate are still forming. And the lag between those two things is where the investors make their buck.
Paul Cantrell
@kingcons
I mean, students have •always• struggled and •always• used ridiculous and ineffective strategies to explore things. It’s just that the mistakes and the necessary learning have shifted again.
Teaching was and still is about jumping in to creative problem-solving, engaging with the world, studying good and bad models, making mistakes, getting helpful guidance, and building a durable capacity for learning from that whole experience. Details change; principles endure.
Brit
@inthehands as someone who used to do lots of tech education in the 2015-2019 era, this is hard to read 😞
I'm glad you could help your student obviously, but gosh does it feel like things have gone sideways
Paul Cantrell
ADDENDUM:
Worth mentioning that in addition to being a relatively new developer, this student is an English language learner. Both the code and the prose take mental effort. She’s doing a hell of a lift here! (Humans are amazing.)
Nothing has taught me more about writing documentation than watching students read documentation. It’s eye-opening.
If you’re a tool vendor: find a capable student / new junior dev, and shut up while you watch them read your docs. It will be…instructive.
3/2
Paul Cantrell
@jenniferplusplus
Tangential thought:
The current LLM craze reminds me a •lot• of late-19C-early-20C spiritualism (attempts to communicate with the dead, seances and all that). There’s a deep hunger for it to be real. And it almost seems possible! And smart, famous people keep buying in (Arthur Conan Doyle, Mr. Logic)! And there’s obvious reason to doubt, but…tantalizing pieces of evidence keep surfacing! And yes, one thing after another keeps turning out to be smoke and mirrors, but…but but….
Marcin Cieślak
@inthehands I wonder if the concept of documentation fades away
I realized I tend to avoid software that forces me to study the concepts and the philosophy of it first from the documentation (For example, #Lilypond didn't let me to go quickly into writing things there).
Watching some junior devs at work - they rely exclusively on prompts given by the IDE, giving at most one-line descriptions of function arguments.
When that fails, they refer to library source code, never the documentation
Paul Cantrell
@saper
Working with new devs a lot, I can confirm that they don’t always reach for docs — in large part because they don’t know •how• to.
Part of that is their learning: figuring out how to be good readers, good searchers, good investigators.
A lot of it is the failing of the docs, which frequently take the POV of tool experts and tool authors, and utterly fail to understand what newcomers need. A hermetically sealed maze with no front door.
Paul Cantrell
@JMMaok
Heh, yeah, I was hoping somebody would have that experience of the first-post cliffhanger. 😇
I'm surprisingly in that camp of “maybe there's greater than zero potential there” for augmentative LLM usage. It's really useful to ask, “What's a pattern that's typical here, never mind if details are all wrong?” But dammit, the tool design, the expectations, the wild hype, everything is just •miles• away right now from a place where that can actually work.
J Miller
After seeing post 1, I already knew what you were going to say in post 2.
File this away for the next hype-driven “AI as a thought partner for students” session.
Paul Cantrell
@datarama @jenniferplusplus
Agreed, and there’s a third “wanting to believe” I see in students:
Learning something unfamiliar is frustrating, daunting, crushing. It makes you feel stupid, over and over and over. One of the most important things students are learning imo is how to work through that, how to be undaunted and uncrushed in the face of being lost. And AI seems like a shortcut around that: maybe you don’t have to do the hard part of growing!
Pseudo Nym
Unix "man(ual)" pages, for an entire ATT SVR4 system, printed out, and bound.
Took up a whole, very wide, shelf.
Online man pages for "man <cmd>" were an excellent resource 30 years ago.
Formatted in a similar fashion, told you what all the options and arguments were, and what they did, and how they interacted
Paul Cantrell
@Oggie @jenniferplusplus
Yes, that’s an important insight. To the extent that both spiritualism and AI hype are both a con, •this• is why the con works.
Oggie
@inthehands @jenniferplusplus Let's not forget that humans are absolutely anthropomorphizing machines. We talk to our bodies, we beg our water heaters, we absolutely treat our cats like sapient logical creations capable of future promises. We bargain with actual rivers and clouds.
Pretending that this box that quacks an awful lot like a duck (and in fact, is the primary motivator for the design space!) is in fact a duck is an incredibly short hop.
Ted Kaminski
@inthehands This has been my near-universal experience with junior developers using ChatGPT.
It somehow *feels* amazingly productive and helpful to them. One mentee did a mini-postmortem on a project that didn't go well, and really, really, powerfully struggled with the cognitive dissonance between "chatgpt helped me be so productive, really useful!" and "I just spent 2 weeks instead of 2 hours on a task because I asked chatgpt instead of reading the documentation."
Paul Cantrell
@tedinski
It reminds me of a study I saw once (wish wish wish I had the link now) that compared lecture-based sit-and-listen classroom instruction with hands-on activity-based instruction. They found that the active, activity-based learning lead to much better comprehension and retention (duh), but immediately after the class, students •felt• like they’d learned more from the lecture. There’s something infectious about passively listening to a voice with an air of confidence.
Cat Hicks
@tedinski @inthehands dealing with high variance like this & difficulty in evaluating related strategies is a very well known classic problem. Instead of acting like learners won't make this mistake it would be nice for meta learning strategies to be taught explicitly, which have shown great benefit for early learners avoiding time suck traps. To me this is very informed by "misconceptions about what studying should look like" research. Software education should catch up and integrate this tbh.
Ted Kaminski
@inthehands Oh wow. I had a similar experience of feeling like active learning was counterintuitive when I first learned how to teach that way.
But I don't think I would have guessed that students would have the same misperception.
Paul Cantrell
@grimalkina @tedinski
Heartily agree! What you describe is a •huge• part of what I think my job is. When I works, the impact is obvious. It’s a difficult thing to systematize. Most of the time, I teach it by letting students get into a pickle, then trying to find the best balance of expert guidance vs personal agency to get them through. And maybe that ad hoc approach is the best/deepest/only way, but…can it be more consistent? more accessible? less dependent on individuals?
Paul Cantrell
@grimalkina @tedinski
You are describing my dpet! There’s a reason almost every single Macalester CS class, from the intro semester on, ends with a project. And why our two intro semesters are…oh, maybe –20% lecture and the rest in-class activities. I really feel like we’re pushing the current boundaries of good pedagogy. And doing that, I really see the limits: we’re up against precedent, student expectations, instructor experience, semesters, grades, systems systems systems…
Cat Hicks
@inthehands @tedinski I mean sounds like project based learning no? We don't expect to regiment and systematize everything in good learning right? That adaptive flex and letting learners steer and scaffolding moments of dynamic mistakes/reflecting on them is part of the safe educational environment you're clearly creating 🙌 I feel like that is how we learn as humans
Cat Hicks
@inthehands @tedinski but I hear you on the challenges of scale and access. If I had the answers I'd have fixed this 😂 but I believe strongly that we can recruit more people to become like you if we show them they can
Paul Cantrell
@grimalkina
I have seen firsthand the impact on students’ lives you describe, and I’ve seen it shift over the course of my time in teaching as we learn to be better teachers.
I also see the students from whom we’re not succeeding.
Have to hold space for both, agreed.
Cat Hicks
@inthehands @tedinski I hear you, it's a painful place to be, but don't forget pushing on the future means seeing the gap between here and there! We gotta hold that space for the kids to do better than we could. Think of all the people who could've gone into some grinder CS dept full of awful teachers and are having this completely different life because of those choices :)
Paul Cantrell
@benjohn @datarama @jenniferplusplus
My own limited optimism about there being a “there” there with LLMs is (1) pattern discovery and (2) maybe boilerplate generation, though even that is dangerously fraught. “Explainer” seems just impossibly far beyond intrinsic limits of this technology.
Benjohn
@inthehands @datarama @jenniferplusplus fwiw, my personal hope for AI is not a short cut to learning, but the most wildly able and ultra patient explainer. But it doesn’t look it will do that at the moment.
Paul Cantrell
@virtualinanity @jenniferplusplus
Yes, that rings a bell! I…may have even boosted it here? Maybe even from Jennifer…? Jennifer, do you have the link perchance?
wasabi brain
@inthehands @jenniferplusplus I wish I could find the article again but someone had a great blog post about how LLMs rely on the same mechanisms as the psychic grift
Paul Cantrell
@Graffotti @grimalkina @tedinski
Yes, this “what your preparing for doesn’t exist yet” line of thought is one of my perennial refrains. (At the heart of my Mastodon-famous-ish essay on liberal arts!)
Graffotti
@grimalkina @inthehands @tedinski
An important practical thing you can teach a computing student is that most of the tools, techniques, and languages that will be important to their future career probably don't exist yet, so (a) they need to be good at teaching themselves new things and finding stuff out (b) it's the fun bit.
Paul Cantrell
@tedinski @grimalkina
Students are generally pretty self-aware and capable of useful reflection tbh, more than one might assume. Not always, but prompt them well and they’ll give good answers more often than not.
Cat Hicks
@tedinski @inthehands dealing with an unpredictable support ("sometimes this will unblock a whole thing I didn't know I didn't know but sometimes this is a rabbit hole distracting me") is not a new meta skill of learning strategy. A lot of the folks I know teaching here are adapting the same meta skill building lessons for correcting misconceptions re how we learn. The hopeful side of students dealing with confusing technology imo is that it makes this stuff really immediately relevant to them.
Ted Kaminski
Postmortem/retrospective analysis is a major way engineers learn, and I don't recall *formally* doing it when I was a student. (I would personally, ofc, "how did I get that wrong?" on any assignment being one of the most basic things to do.)
But in this case, they'd written a great retrospective... except they'd run face-first into cognitive dissonance and failed to overcome it.
I don't know how to teach overcoming it, except to point it out where I can.
Graffotti
Aha! Found it, and only slightly distracted by wookieepedia 😎 Wonderful!
(https://innig.net/teaching/liberal-arts-manifesto)
Paul Cantrell
@Graffotti
Rvuaiavoauaoomaaaaa!
Chris Bohn
@inthehands Maybe it's time to update the old "Documentation vs Stack Overflow" picture.
@990000@mstdn.social
@inthehands @saper lots of docs are inadequately crafted and jump straight into granular reference without providing broader context, like of common workflows in particular. There are big gaps sometimes even if they provide an intro. Oftentimes the intro is also too simplified.
Paul Cantrell
@shalien That is what I’m describing in the previous post, yes. Not •explained• so much as walked them through it in context.
:projetstodon: Shalien
@inthehands Did you explain to your students how to search for the documentation ?
wasabi brain
@inthehands @virtualinanity @jenniferplusplus finally found it…still a great read and still holds up: https://softwarecrisis.dev/letters/llmentalist/
Paul Cantrell
@cxj
I appreciate the props, but I can’t take too much individual credit. I’m in a program and a school that set me up for success: low student/teacher ratio, curricular emphasis on active and project-based learning, strong community values around people caring for each other, tons of license for pedagogical creativity, etc etc. It’s a lot easier to do a good job when you’re set up for it.
Chris Johnson
@inthehands I wish I’d had profs like you when I was getting my CS degree. Props to you for being a good educator.
Paul Cantrell
@Npars01
We’ve worked hard to create a context where students •can• ask for help — which means structure, psychological safety, positive social relationships, low student/teacher ratio, etc etc — and yes, the positive effects are numerous. It takes real work at all the levels of an educational institution. ChatAI’s flaws are there to prey on weak institutions, ed and biz.
Nicole Parsons
Many students wouldn't have been brave enough to ask for help. They'd give up & drop out.
In the future, how many students will have their educations thwarted or disrupted by ChatAI?
Inken Paper
@inthehands @saper there is a whole trade called "technical writing" whose purpose is to solve this exact problem.
Paul Cantrell
@crashglasshouses @saper
Indeed! And things go so much better when there are actual technical writers in the game, with the resources and support they need to do a good job.
Technical writing is in my view a discipline that still has a lot to learn, but has even more to give. Ignoring it is such a sad unforced error.
Paul Cantrell
@LauraLangdon @tedinski
Yup! It’s tough because (1) it takes a while to get good at it as an instructor, (2) it requires a body of materials that can take •years• to build, and (3) it’s hard to warm students to it if there’s not already a widespread student culture of embracing it. Those three interact, and bootstrapping it all takes many, many years.
Laura Langdon
@tedinski @inthehands When I started using active learning techniques in my classes (math) my student evaluations completely polarized. Some students really hated me for it, because they felt like I wasn’t “teaching” them.
Paul Cantrell
@simon @jenniferplusplus
I do think that seances etc gave folks all kinds of valuable things in the 19C: closure, a way to grieve, perhaps therapy before therapy was a thing.
Re your post, which I only skimmed: this sort “what goes here? what’s there right pattern? where do I start??” question in an unfamiliar domain seems like a thing that LLMs could be good at. In your case…
Simon Willison
@inthehands @jenniferplusplus maybe the analogy here holds in that sometimes a spiritualist might give you life advice that turns out to be useful anyway?
Simon Willison
@inthehands @jenniferplusplus I love this analogy so much
In this case, for me at least, the ghosts are kind of real though... https://simonwillison.net/2024/Mar/23/building-c-extensions-for-sqlite-with-chatgpt-code-interpreter/
Make of that what you will!
Paul Cantrell
@simon @jenniferplusplus …he binary ←→ string and dot product code is readily boilerplate, but getting it into a DB extn is potentially a hard thing to look up.
Then again, I’d be extremely suspicious of those string conversion functions. Stuff like this in C code sets off my alarm bells:
char *result = malloc(size * 12 + 2); // Max 10 digits per float + comma + space
(Note no further memory bounds check in the code for that buf! Eek)
Isn’t there a library for that job…?
dricibone
@inthehands @jenniferplusplus A.O. Scott had an essay in the NYT Book Review a few weeks ago making this same comparison to spiritualism.
Paul Cantrell
@FenTiger @tedinski
Heh, we agree so much you’re channeling one of my pinned posts: https://hachyderm.io/@inthehands/110040662990948941
Mike P
@tedinski @inthehands Junior developers think that writing code is hard work, and they appreciate anything that makes it seem easier.
Experienced developers know that writing code is the easy bit. They appreciate anything that makes the really hard stuff easier.
Paul Cantrell
@LauraLangdon @dahukanna @tedinski
This is all on point. I find that often (not always, but often) students already feel their instinct pulling them in the direction of “why,” meta-learning, liberal arts whole-human learning, all these good things, but need an authority figure just to give those instincts a blessing and a nudge — not because they aren’t natural instincts, but because previous indoctrination / trauma has made them doubt it.
Dawn Ahukanna
@LauraLangdon @tedinski @inthehands
Agreed. That education “indoctrination” starts way before your classroom - home to wider society to classroom to office.
For anyone not asking
- “what should I (be allowed /permitted to) know?”
&
- “how do I imitate/imprint to the extent of becoming a ‘carbon copy’?”
those groups typically respond with “How dare you question my authority, get out of my house, society, classroom, office!”
Laura Langdon
@dahukanna @tedinski @inthehands So much this!
Laura Langdon
@dahukanna @tedinski @inthehands Right in one! And I don’t blame the students; they’ve been successfully convinced by our educational system that learning is essentially just assimilation.
Dawn Ahukanna
@LauraLangdon @tedinski @inthehands
I’d say “know what” group got mad at you whilst “know how”+”know why” active learning group were ecstatically figuring their own “know what” mental models.
Related - https://mastodon.social/@dahukanna/112144732199240572
Paul Cantrell
@Covok @jenniferplusplus
If an AI getting one person killed were enough to actually make the bottom fall out of the grift, that would be a first in the history of capitalism and I’d be thrilled.
cf the “Radium girls,” or leaded gasoline, or etc etc
Kamala Kara A 🏳️🌈 🏳️⚧️
@inthehands @jenniferplusplus I think the question I have is "when does the bottom fall out?" Like, the tech doesn't work but it tricks the owning class into thinking it does. The owning class has fired and replaced staff with this tech in many industries. It takes a long time for the issuses of this kind of mistake to manifest. What will occur that finally reveals the grift? Will some company replace their accounts with Chatgpt and become the next Enron? Will an AI program get someone killed?
Paul Cantrell
@dricibone @jenniferplusplus
Oh, to AI?? I often like A. O. Scott, so I’ll have to look that up!
…[searching]…
Is this the one you’re talking about? https://www.nytimes.com/2023/06/21/books/review/book-bans-humanities-ai.html
Jenniferplusplus
@inthehands @Covok AI has already killed people via driving cars, and it hasn't even been enough to drop the bottom out of AI driven cars. In fact, it probably just served to normalize that computers have the ability to hurt and kill people.
dricibone
@inthehands @jenniferplusplus this one: https://www.nytimes.com/2023/12/27/books/review/writers-artificial-intelligence-inspiration.html
Paul Cantrell
@sepi1enfwb
(You have nothing to apologize for here, imo. Your earlier post was a good post, and nice turn of phrase!)
Sepi Girardi
@990000 @binford2k @inthehands @saper I don't think anybody here is bad and apologize if the phrasing got across that way.
I also agree about very experienced writers talking to engineers who give them things piecemeal. I don't think the problem is with the writer.
I think part of the problem is that those who know <insert thing>, know it so well that they do not understand why somebody else would need some sort of "layered reveal" of <insert thing> - peeling the layers of the onion back...
@990000@mstdn.social
@binford2k @sepi1enfwb @inthehands @saper yeah, I think all these folks generally mean well and it's just not easy to write good documentation with proper dev onboarding
Ben Ford :grinchsmile:
@sepi1enfwb @990000 @inthehands @saper or they’re written by excellent tech writers who know little about the context or use cases, starting from reference materials generated by devs explaining the minutiae about all the details.
Sepi Girardi
@990000 @inthehands @saper those docs are written by folks who "know too much to be helpful."
Paul Cantrell
@jenniferplusplus @Covok
ding ding ding
Paul Cantrell
@dricibone @jenniferplusplus
Ah! Thank you. Bookmarked for when I’m at the library or otherwise have NYT access. (I’ve cancelled my subscription so many times in the last decade I’m now at like NYT -7 subscriptions.)
dricibone
@inthehands @jenniferplusplus same boat. I do pick up a Sunday print copy every month or so and take just as long to read it.
Paul Cantrell
@fancysandwiches @Covok @jenniferplusplus
Just came across this relevant and excellent Doctorow article:
https://locusmag.com/2023/12/commentary-cory-doctorow-what-kind-of-bubble-is-ai/
Fancy Sandwiches
@Covok @inthehands @jenniferplusplus I think it's going to be the Enron thing. Some company deemed too big to fail will go all in on AI, and fail miserably. This will make it much more appealing to report on how bad all of this shit is, so we'll see more mainstream reports on how bad this tech actually is. This is basically what happened to crypto after FTX fell over.
Sumana Harihareswara
@LauraLangdon @tedinski @inthehands I remember the relevant study
"Measuring actual learning versus feeling of learning in response to being actively engaged in the classroom" [2019]
Louis Deslauriers, Logan S. McCarty, Kelly Miller, Kristina Callaghan, and Greg Kestin
Paul Cantrell
@brainwane @LauraLangdon @tedinski
That’s the one, I’m sure!! Thank you thank you.
Paul Cantrell
@brainwane @LauraLangdon @tedinski
Yes, it’s a high-quality discussion! Once again I find myself thinking I should pay more attention to Metafilter.
Sumana Harihareswara
@inthehands @LauraLangdon @tedinski Yay!
I do recommend the MetaFilter discussion thread which also links to other relevant resources.
GNU social JP is a social network, courtesy of GNU social JP管理人. It runs on GNU social, version 2.0.2-dev, available under the GNU Affero General Public License.
All GNU social JP content and data are available under the Creative Commons Attribution 3.0 license.