Hmmm, I might be a dyed-in-the-wool contrarian, but I would like this Snorkl a lot better if it allowed me to respond directly to its feedback and see the reasoning behind the score it gives. I did the pb&j sample paragraph (intentionally leaving off any intro or conclusion so I could see what the feedback would look like) and I got 2/4. The feedback I got was:
"Great job listing the main ingredients needed and explaining how to spread both the peanut butter and jelly evenly! You were very clear about putting the slices together at the end." "I noticed you mentioned wiping the knife on the bread. Can you think of a cleaner way to clean the knife between spreading peanut butter and jelly?" "I love how you added your personal preference about cutting the sandwich diagonally and acknowledged that others might do it differently! This shows great thinking beyond the basic steps."
There's no rubric or other rationale for the score, at least not that I can find easily. So, I got half off a writing assignment because I described a somewhat unsanitary sandwich-making practice? Sorry, but you've lost me.
This tool is just a chatbot, nothing has fundamentally changed except foundation models have all ramped up in GPUs and can now take video/audio input in an economically viable way. Last year it wasn't possible, but everybody knew this was coming fast. I'm part of the "consultant" class here, former teacher, yet I happen to agree with you that a chatbot tutor s*cks! I'm more of a wait and see attitude. We're now seeing different modalities being introduced, after video input, the next step will be video output, so we can have the AI "gesture" at the kids' work. I'm not sure what it's going to look like, but I bet some people will have great ideas. It will never replace a (good) teacher, but for a kid who doesn't have support at home, these tools may make a big difference. I'm with you though, I'd also like to know if kids click into the feedback...
I agree, Wess. I think Dan’s theme about AI systems starting to be able to replicate the various things that a human does automatically (spoken, unspoken, constantly assessing and looking for opportunities to provide feedback) is coming.
I’m also a former teacher now working as a designer in EdTech and IMO, I think there are two promising avenues in the future that I’m excited about:
1) enabling actual differentiated instruction and personalized learning plans that remain centered on the teacher-student and student-student dynamics in the classroom
2) using AI as a co-pilot tool for human tutors who are subject-matter experts in their field but lack pedagogical training (nudges to chunk up content, proposed lesson plans and learning plans that scaffold appropriately and build in research-based techniques like spaced repetition). This co-pilot model could also work for tutors with educator backgrounds and would be value-added in terms of on-the-fly content generation and personalization.
"Because I am fundamentally unable to help myself ..." — can this be the sentence starter for every post you make from now? That signals delightful content will follow.
The "Consultant" is incredibly misleading - Technology coaches are employees of the school district and school building. Often former teachers tasked with bringing tech to their teachers. They are not consultants.
It's a very important question. One is often paid to promote a tool. The other one is paid to help create a beneficial learning environment using technology, which often means warning people AWAY from commercial tools. Instructional designers and instructional technologists are not corporate shills (or shouldn't be).
Is there a third big thing the human does? Sort kid into "plausibly can do this" versus "lost in space." Bots don't sort but could.
Lost in space kid just sees: "Blah blah blah 10% shorter. Blah blah blah 15 centimeters long."
Tutor appears over the shoulder, and kid means: "Tell me to write something that has a 10 and 15 in it. I truly and honestly have no idea."
Tough decision. You can Jalah Bryant the kid for 10 minutes. Or quasi-punt.
From teacher pov, is unrealized AI value improving the punt? Instead of Lost In Space kid "pretend doing" the granola problem, work on simple percentages all period...new goal is kid can eyeball something and say "10%" or "50%" etc.
I tend to think the lower tech personalized learning of years past has already hit the ceiling here. If the kid gets five problems wrong it'll (I think wrongly in many cases) punt them to easier problems. Generative AI right now just keeps on chipping away at the SAME problem, making it easier and easier until it all but does it for you.
As Dan said, human teachers and tutors do a million imperceptible things when working with a student 1:1. I do think AI can start to do some of it; I love the instant feedback loop (though agree that students might not use it, especially disengaged students who would benefit most).
I really like your idea of differentiating the next best activity for a student based on everything the system knows so far (lost in space = take a step back and work in foundational skills). This isn’t anything new conceptually, but I haven’t seen much out there that really nails this yet.
I think we've identified what the kid doesn't know and are assigning work based on that, but it might be more helpful to identify what they DO know, so we can draw a line from that to this, "See, this thing is very much like this other thing you already know."
They may have 4 million subscribers, but did the teachers ask for this? Or was it just bought and teachers were told to use it? Another instance of administrators buying tech and then using teachers to figure out what to do with it.
Hi Dan...try our YoChatGPT.io! Trying to find best practices of using GenAI into education. Would love your input. We have math handwriting recognition so far, and incorporating sketching next.
A lot has changed, but not the way that some people thought it would. Online education and digital education are definitely here to stay, but it did NOT lead to the complete removal of in-person teaching like somethought it would (Digital Diploma Mills, for example: https://dl.acm.org/doi/10.1145/280449.280454).
I’m a fan of minor chords personally. I just listened to a podcast this morning that made me think of you! It made a connection between AI and “A Wrinkle in Time”: https://gretchenrubin.com/podcast/a-little-happier-what-do-humans-possess-that-artificial-intelligence-doesnt-possess/
You had me at Wrinkle in Time
Hmmm, I might be a dyed-in-the-wool contrarian, but I would like this Snorkl a lot better if it allowed me to respond directly to its feedback and see the reasoning behind the score it gives. I did the pb&j sample paragraph (intentionally leaving off any intro or conclusion so I could see what the feedback would look like) and I got 2/4. The feedback I got was:
"Great job listing the main ingredients needed and explaining how to spread both the peanut butter and jelly evenly! You were very clear about putting the slices together at the end." "I noticed you mentioned wiping the knife on the bread. Can you think of a cleaner way to clean the knife between spreading peanut butter and jelly?" "I love how you added your personal preference about cutting the sandwich diagonally and acknowledged that others might do it differently! This shows great thinking beyond the basic steps."
There's no rubric or other rationale for the score, at least not that I can find easily. So, I got half off a writing assignment because I described a somewhat unsanitary sandwich-making practice? Sorry, but you've lost me.
Lol can each of your posts please include an investigation into Ed tech company claims like that of magicschool!?
This tool is just a chatbot, nothing has fundamentally changed except foundation models have all ramped up in GPUs and can now take video/audio input in an economically viable way. Last year it wasn't possible, but everybody knew this was coming fast. I'm part of the "consultant" class here, former teacher, yet I happen to agree with you that a chatbot tutor s*cks! I'm more of a wait and see attitude. We're now seeing different modalities being introduced, after video input, the next step will be video output, so we can have the AI "gesture" at the kids' work. I'm not sure what it's going to look like, but I bet some people will have great ideas. It will never replace a (good) teacher, but for a kid who doesn't have support at home, these tools may make a big difference. I'm with you though, I'd also like to know if kids click into the feedback...
I agree, Wess. I think Dan’s theme about AI systems starting to be able to replicate the various things that a human does automatically (spoken, unspoken, constantly assessing and looking for opportunities to provide feedback) is coming.
I’m also a former teacher now working as a designer in EdTech and IMO, I think there are two promising avenues in the future that I’m excited about:
1) enabling actual differentiated instruction and personalized learning plans that remain centered on the teacher-student and student-student dynamics in the classroom
2) using AI as a co-pilot tool for human tutors who are subject-matter experts in their field but lack pedagogical training (nudges to chunk up content, proposed lesson plans and learning plans that scaffold appropriately and build in research-based techniques like spaced repetition). This co-pilot model could also work for tutors with educator backgrounds and would be value-added in terms of on-the-fly content generation and personalization.
I am the edtech colleague you mentioned who can only play two notes, but I really like your posts.
"Because I am fundamentally unable to help myself ..." — can this be the sentence starter for every post you make from now? That signals delightful content will follow.
The "Consultant" is incredibly misleading - Technology coaches are employees of the school district and school building. Often former teachers tasked with bringing tech to their teachers. They are not consultants.
"Who do they work for?" is, in this case, a less relevant question than "What does it take to keep their job?"
It's a very important question. One is often paid to promote a tool. The other one is paid to help create a beneficial learning environment using technology, which often means warning people AWAY from commercial tools. Instructional designers and instructional technologists are not corporate shills (or shouldn't be).
You just couldn't help yourself.....
Really like your point on gesturing.
Is there a third big thing the human does? Sort kid into "plausibly can do this" versus "lost in space." Bots don't sort but could.
Lost in space kid just sees: "Blah blah blah 10% shorter. Blah blah blah 15 centimeters long."
Tutor appears over the shoulder, and kid means: "Tell me to write something that has a 10 and 15 in it. I truly and honestly have no idea."
Tough decision. You can Jalah Bryant the kid for 10 minutes. Or quasi-punt.
From teacher pov, is unrealized AI value improving the punt? Instead of Lost In Space kid "pretend doing" the granola problem, work on simple percentages all period...new goal is kid can eyeball something and say "10%" or "50%" etc.
I tend to think the lower tech personalized learning of years past has already hit the ceiling here. If the kid gets five problems wrong it'll (I think wrongly in many cases) punt them to easier problems. Generative AI right now just keeps on chipping away at the SAME problem, making it easier and easier until it all but does it for you.
As Dan said, human teachers and tutors do a million imperceptible things when working with a student 1:1. I do think AI can start to do some of it; I love the instant feedback loop (though agree that students might not use it, especially disengaged students who would benefit most).
I really like your idea of differentiating the next best activity for a student based on everything the system knows so far (lost in space = take a step back and work in foundational skills). This isn’t anything new conceptually, but I haven’t seen much out there that really nails this yet.
I think we've identified what the kid doesn't know and are assigning work based on that, but it might be more helpful to identify what they DO know, so we can draw a line from that to this, "See, this thing is very much like this other thing you already know."
They may have 4 million subscribers, but did the teachers ask for this? Or was it just bought and teachers were told to use it? Another instance of administrators buying tech and then using teachers to figure out what to do with it.
Hi Dan...try our YoChatGPT.io! Trying to find best practices of using GenAI into education. Would love your input. We have math handwriting recognition so far, and incorporating sketching next.
Thank you for augmenting our perspective and not diminishing
Interesting how you analyzed the MagicSchool feedback! Such good practice for all news sources
scraping tweets and classifying their bios 👏👏 👏
I still don't think any of these are worth the money spent. In 1991 everyone was hyperventilating about computer-aided education. Nothing's changed.
A lot has changed, but not the way that some people thought it would. Online education and digital education are definitely here to stay, but it did NOT lead to the complete removal of in-person teaching like somethought it would (Digital Diploma Mills, for example: https://dl.acm.org/doi/10.1145/280449.280454).