Having some experience vibe-coding custom apps, I'm not impressed in the short-term with that from a student-facing perspective. Your earlier posts on technology in the classroom do a great job highlighting why.
I'm much more optimistic about vibe-coding staff workflows, like automating the LLM to good-looking pdf process. We have a few things like that in place in my district. This is a great reminder to me to work on one for math practice. Doing this at scale would also let us use an API to call higher-end models that (usually) do better with the production of examples and/or "force" some of the reflective part of the design process that the AI can otherwise shortcut.
Yeah I'd love to see this kind of stuff. There's a ton of potential on the teacher workflow side. One issue is that every teacher's workflow is so different, it's tough to come up with tools that are broadly useful because teachers have all these little idiosyncratic workflows in response to their unique challenges.
A handy thing about vibe-coding is that it brings the marginal cost of software development close to zero, so bespoke tools are more feasible than before.
A couple of weeks ago I built a custom web app for a staff member with a completely unique role in our system. It wouldn’t have made sense to buy a product for that one staff member, but an hour of time with Claude Code solved what she reported as the most annoying part of her job.
While this example was on the clerical side and not a teacher, I can imagine similar work with teachers, too. To some extent custom development is easier with only one customer to please.
That jump from question 2 to question 3 is insane. I will never understand why curriculum writers do that. As for the AI tools that embed with curriculum, I’m not sure if I’ve shared this before but Illustrative Math has one called Coteach.ai that essentially just reads and replicates the format of their curriculum. My understanding is that the founders created it independently but now it has been certified by IM. It’s not perfect but it’s way better than using a random generative AI tool and it seems to be steadily getting better at things. It seems like all curriculums could have this if someone was motivated to create it. Or someone could just ask an AI to create an AI tool? There’s some recursive process out there that should be able to make this work…
Yea, if you’ve ever taught this skill you know that jump is really challenging for lots of students and needs really careful preparation. Completely wild.
Yea I checked out coteach on your recommendation earlier this year. I like it! I prefer my custom workflow right now because of some formatting and customization details that I prefer, but I do like that it’s connected directly to the IM curriculum. I think there’s a ton of potential for tools like that, though again I would love to see them integrated directly into curricula rather than being add-ons.
I agree that some math textbooks, as well as science textbooks, do not have enough practice problems. This starts in the early grades as well. It is why all the students I tutor in math do not know the basics. Courses given befor Algebra or Pre-Algebra can incorporate problems that are methaphors for wha they will see in Algebraic manipulations.
Yea I agree. Practice isn't just repetition, it should get gradually harder. Those harder questions can preview topics students will learn in the future, previewing future topics and making that learning smoother.
I've written about similar issues in English, calling it the "problem of disposable examples." Many language resources never get out of tutorial mode and into good old interleaving and such. Glad I'm not the only one using AI for extra examples!
Do you find AI does a good job with examples for your purposes? I've found it does well with some simple repetition or mixed practice, but if I try to get it to do something more ambitious like problems that build iteratively toward a larger goal, the results are a mixed bag, sometimes good but sometimes bad or nonsensical.
The crucial difference might be LLM's bent towards words rather than numbers. For verbal problems, it does well--but if and only if you control for the right variables.
I've been experimenting with analogies for some time now, but controlling the reading level, and therefore difficulty, becomes make or break. For other things, like creating citations for error detection, it does really really well.
Where I fail is integrating these things into my daily teaching. Long story, but this year I was shifted positions, buildings, and schedules, and I've utterly failed to pace things correctly.
*The comments would not let me post a screenshot of some example questions created using AI.
Having some experience vibe-coding custom apps, I'm not impressed in the short-term with that from a student-facing perspective. Your earlier posts on technology in the classroom do a great job highlighting why.
I'm much more optimistic about vibe-coding staff workflows, like automating the LLM to good-looking pdf process. We have a few things like that in place in my district. This is a great reminder to me to work on one for math practice. Doing this at scale would also let us use an API to call higher-end models that (usually) do better with the production of examples and/or "force" some of the reflective part of the design process that the AI can otherwise shortcut.
Yeah I'd love to see this kind of stuff. There's a ton of potential on the teacher workflow side. One issue is that every teacher's workflow is so different, it's tough to come up with tools that are broadly useful because teachers have all these little idiosyncratic workflows in response to their unique challenges.
A handy thing about vibe-coding is that it brings the marginal cost of software development close to zero, so bespoke tools are more feasible than before.
A couple of weeks ago I built a custom web app for a staff member with a completely unique role in our system. It wouldn’t have made sense to buy a product for that one staff member, but an hour of time with Claude Code solved what she reported as the most annoying part of her job.
While this example was on the clerical side and not a teacher, I can imagine similar work with teachers, too. To some extent custom development is easier with only one customer to please.
That jump from question 2 to question 3 is insane. I will never understand why curriculum writers do that. As for the AI tools that embed with curriculum, I’m not sure if I’ve shared this before but Illustrative Math has one called Coteach.ai that essentially just reads and replicates the format of their curriculum. My understanding is that the founders created it independently but now it has been certified by IM. It’s not perfect but it’s way better than using a random generative AI tool and it seems to be steadily getting better at things. It seems like all curriculums could have this if someone was motivated to create it. Or someone could just ask an AI to create an AI tool? There’s some recursive process out there that should be able to make this work…
Yea, if you’ve ever taught this skill you know that jump is really challenging for lots of students and needs really careful preparation. Completely wild.
Yea I checked out coteach on your recommendation earlier this year. I like it! I prefer my custom workflow right now because of some formatting and customization details that I prefer, but I do like that it’s connected directly to the IM curriculum. I think there’s a ton of potential for tools like that, though again I would love to see them integrated directly into curricula rather than being add-ons.
I agree that some math textbooks, as well as science textbooks, do not have enough practice problems. This starts in the early grades as well. It is why all the students I tutor in math do not know the basics. Courses given befor Algebra or Pre-Algebra can incorporate problems that are methaphors for wha they will see in Algebraic manipulations.
Yea I agree. Practice isn't just repetition, it should get gradually harder. Those harder questions can preview topics students will learn in the future, previewing future topics and making that learning smoother.
I've written about similar issues in English, calling it the "problem of disposable examples." Many language resources never get out of tutorial mode and into good old interleaving and such. Glad I'm not the only one using AI for extra examples!
Do you find AI does a good job with examples for your purposes? I've found it does well with some simple repetition or mixed practice, but if I try to get it to do something more ambitious like problems that build iteratively toward a larger goal, the results are a mixed bag, sometimes good but sometimes bad or nonsensical.
The crucial difference might be LLM's bent towards words rather than numbers. For verbal problems, it does well--but if and only if you control for the right variables.
I've been experimenting with analogies for some time now, but controlling the reading level, and therefore difficulty, becomes make or break. For other things, like creating citations for error detection, it does really really well.
Where I fail is integrating these things into my daily teaching. Long story, but this year I was shifted positions, buildings, and schedules, and I've utterly failed to pace things correctly.
*The comments would not let me post a screenshot of some example questions created using AI.