How to explain Generative AI in the classroom • Hacker News Board

Re: How to explain Generative AI in the classroom

#21

Post by jraph » Sat, Jan 31, 2026, 1:48 PM UTC

sdwr wrote: post #10

I think this is a backwards approach, especially for children. Gen AI is magical, it makes stuff appear out of thin air! And it's limited, everything it makes kinda looks the same And it's forgetful, it doesn't remember what it just did And it's dangerous! It can make things that never happened Starting with theory might be the simplest way to explain, but it leaves out the hook. Why should they care?

As a child, I think I would have been annoyed by such a presentation. There are science magazines for children that can explain pretty complex stuff just fine.

It's also critical not to leave out the ethical topics (resource consumption, e-waste production, concerns about how the source data is harvested - both how it DDoS websites and how authors are not necessarily happy with their work ending in the models)

Re: How to explain Generative AI in the classroom

#22

Post by d1sxeyes » Sat, Jan 31, 2026, 6:23 PM UTC

thfuran wrote: post #14

Uehreka wrote: post #3

I couldn’t find it easily, what age range is this intended for? The images make it seem elementary-school-ish, but I’m not sure if elementary school kids have the foundations for interpreting scatterplots, let alone scatterplots with logarithmic axes. I’ve been out of education for a while though, so maybe I’m misremembering.

Logarithms are high school.

Given the author’s domain is .co.uk, and there’s a reference to part of the UK at the bottom, I’d say this is likely aimed at an average Y10/11 (15-16 y.o.). It could perhaps be used with more able kids lower down the school, but I doubt it would be accessible to any under the age of 13.

Re: How to explain Generative AI in the classroom

#23

Post by mystraline » Sat, Jan 31, 2026, 7:04 PM UTC

narrator wrote: post #7

That lesson plan is a good practical start. I think it misses the very big picture of what we've created and the awesomeness of it. The simplest explanation I can give is we have a machine that you feed it some text from the internet, and you turn the crank. Most machines we've had previously would stop getting better at predicting the next word after a few thousand cranks. You can crank the crank on an LLM 10^20 tim…

> You can crank the crank on an LLM 10^20 times and it will still get smarter.

No, it won't.

Training/learning is separate from the execution of the model. Takes megadollars to train and kilodollars to effectively run.

Its basically a really complicated PID loop. You can test and 'learn' the 3 neuron function, and then you can put it into execution. Can't do both.

Sure, theres context length and fine tuning to slightly alter a model.

But theres no adaptive, self-growing LLM. Probably won't be for a long time.

Re: How to explain Generative AI in the classroom

#24

Post by westurner » Sat, Jan 31, 2026, 7:12 PM UTC

sdwr wrote: post #10

I think this is a backwards approach, especially for children. Gen AI is magical, it makes stuff appear out of thin air! And it's limited, everything it makes kinda looks the same And it's forgetful, it doesn't remember what it just did And it's dangerous! It can make things that never happened Starting with theory might be the simplest way to explain, but it leaves out the hook. Why should they care?

> Starting with theory might be the simplest way to explain,

Brilliant's AI course has step-by-step interactive textgen LLMs trained on TS (Swift) lyrics and terms of services with quizzes for comprehension and gamified points.

Here's a quick take:

LLM AI are really good at generating bytes that are similar to other bytes, but aren't yet very good at caring whether what they've generated is wrong or incorrect. Reinforcement Learning is one way to help prevent that.

AI Agents are built on LLMs. An LLM (Large Language Model) is a trained graph of token transition probabilities (a "Neural Network" (NN), a learning computer (Terminator (1984))). LLMs are graphical models. Clean your room. The grass is green and the sky is blue. Clean it well

AI Agents fail where LLMs fail at "accuracy" due to hallucinations even given human-curates training data.

There are lots of new methods for AI Agents built on LLMs which build on "Chain of Thought"; basically feeding the output from the model back through as an input a bunch of times. ("feed-forward")

But if you've ever heard a microphone that's too close to a speaker, you're already familiar with runaway feedback loops that need intervention.

There are not as many new Agentic AIs built on logical reasoning and inference. There are not as many AI Agents built on the Scientific Method that we know to be crucial to safety and QA in engineering.