Great article and interview! As a middle school teacher, I try not to play up AI too much, lest I summon the Streisand Effect. I will say, however, that my corporation has Grammarly installed on devices. No big deal. Until the day *they* added AI elements, and suddenly some students start generating professional sounding prose. That has been tricky.
I've also talked to students about hallucinations. I remind them that certain texts are not in data sets, and AI cannot therefore just spout out answers. I mean, it *can*, and it'll be obvious. That said, I'm thankful they still cheat uncritically the old fashioned way: Copying whatever Google says whether it's right or wrong.
But as for the impact on the K12 classroom, there are teachers on Substack. Ask away!
I like Alan's sober and well-considered look at AI - focusing on its potential while acknowledging the many shortcomings. The approach of working through a text yourself and then comparing AI output is the perfect way to ease someone unfamiliar into AI while exposing the pitfalls like hallucinations, etc.
And I tend to agree that hallucinations are less critical when it comes to creative writing and brainstorming. In fact, they may actually nudge your thinking into an unexpected and fun direction.
Thanks, Daniel. I think this is one of the best interviews on AI composition I have come across. I am glad to have played a small part in its publication. Thanks again for the comments and restacks. I look forward to your responses and insights.
This was a wonderful interview. I like how Alan sees LLMs as tools -- powerful tools for certain tasks, but tools that have serious limitations for others. Knowing how and when to use AI tools is key.
Exactly! I am writing a piece about AI now, due out this week on my substack. Here is the rub, those of us who already know how to write, who have published, who are readers, we know what good writing feels like. We were already intellectually formed when AI came around. I worry about generations to come who may not have access to human-led education, analog-produced writing, will they have the instincts to use reading/writing as a nuanced form of communication? With all new tech, new knowledges are produced, but also knowledges are lost. Yes, using AI correctly as a tool, we will be able to produce higher quality and faster iterations, we'll get there with less friction, but what will be lost?
It's these sorts of practical articles that are helping make AI real. It's not hype, fear, or prompting incantations. In reality, it's nearly identical to how I brainstorm my writing with my wife and friends. Two books under my belt and hundreds of essays and I can attest, AI can help augment, but not replace the human engagement.
Yes, Michael, when I came across Alan's work, I had this exact reaction. A breath of fresh air. He is an incredible teacher and practitioner. The notion of generating our own AI support bot is brilliant. The theory is clear and very actionable. I can't wait to see what Alan does next!!!
Yes, Grammarly did open up another entry point. The cat and mouse game gets better old. Looking for a better way. Thanks for the invite to ask. I will take you up on that.
Great article and interview! As a middle school teacher, I try not to play up AI too much, lest I summon the Streisand Effect. I will say, however, that my corporation has Grammarly installed on devices. No big deal. Until the day *they* added AI elements, and suddenly some students start generating professional sounding prose. That has been tricky.
I've also talked to students about hallucinations. I remind them that certain texts are not in data sets, and AI cannot therefore just spout out answers. I mean, it *can*, and it'll be obvious. That said, I'm thankful they still cheat uncritically the old fashioned way: Copying whatever Google says whether it's right or wrong.
But as for the impact on the K12 classroom, there are teachers on Substack. Ask away!
Thanks for this deep dive interview, Nick!
I like Alan's sober and well-considered look at AI - focusing on its potential while acknowledging the many shortcomings. The approach of working through a text yourself and then comparing AI output is the perfect way to ease someone unfamiliar into AI while exposing the pitfalls like hallucinations, etc.
And I tend to agree that hallucinations are less critical when it comes to creative writing and brainstorming. In fact, they may actually nudge your thinking into an unexpected and fun direction.
Thanks for sharing your thoughts Alan!
Thanks, Daniel. I think this is one of the best interviews on AI composition I have come across. I am glad to have played a small part in its publication. Thanks again for the comments and restacks. I look forward to your responses and insights.
This was a wonderful interview. I like how Alan sees LLMs as tools -- powerful tools for certain tasks, but tools that have serious limitations for others. Knowing how and when to use AI tools is key.
Thanks Suzi. I was thinking about outer conversation about understanding today. I listened to some great interviews this weekend you might be interested in. These guys are very empirically motivated, skirt around the concept of understanding in very interesting ways: https://open.substack.com/pub/thegradientpub/p/cameron-jones-sean-trott-llms-meaning-grounding?r=2l25hp&utm_campaign=post&utm_medium=web, this guy argues strongly that LLMs do not understand, pretty convincing, gets into kahnmann, etc: https://open.substack.com/pub/thegradientpub/p/subbarao-kambhampati-planning-reasoning-llms?r=2l25hp&utm_campaign=post&utm_medium=web
Thanks, Nick! These do sound interesting :) The Gradient is great, isn't it? They always seem to find the fascinating topics.
Exactly! I am writing a piece about AI now, due out this week on my substack. Here is the rub, those of us who already know how to write, who have published, who are readers, we know what good writing feels like. We were already intellectually formed when AI came around. I worry about generations to come who may not have access to human-led education, analog-produced writing, will they have the instincts to use reading/writing as a nuanced form of communication? With all new tech, new knowledges are produced, but also knowledges are lost. Yes, using AI correctly as a tool, we will be able to produce higher quality and faster iterations, we'll get there with less friction, but what will be lost?
Friction is very important. I agree! Let’s keep talking! https://open.substack.com/pub/nickpotkalitsky/p/the-art-of-imperfection-why-human?r=2l25hp&utm_medium=ios
It's these sorts of practical articles that are helping make AI real. It's not hype, fear, or prompting incantations. In reality, it's nearly identical to how I brainstorm my writing with my wife and friends. Two books under my belt and hundreds of essays and I can attest, AI can help augment, but not replace the human engagement.
Yes, Michael, when I came across Alan's work, I had this exact reaction. A breath of fresh air. He is an incredible teacher and practitioner. The notion of generating our own AI support bot is brilliant. The theory is clear and very actionable. I can't wait to see what Alan does next!!!
Yes, Grammarly did open up another entry point. The cat and mouse game gets better old. Looking for a better way. Thanks for the invite to ask. I will take you up on that.