Tech EDGE, AI in the Classroom: AI and Energy Use
Tech EDGE
Author
07/01/2025
Added
2
Plays
Description
Ever wondered how much energy AI really uses? Guy dives into it in this episode of AI in the Classroom. University of Nebraska-Lincoln (UNL) | College of Education and Human Sciences (CEHS) | Department of Teaching, Learning and Teacher Education (TLTE)
Searchable Transcript
Toggle between list and paragraph view.
- [00:00:00.383]Coming up on AI in the Classroom.
- [00:00:02.485]AI and energy use.
- [00:00:04.371](upbeat music)
- [00:00:09.609]Hi, my name is Williams
- [00:00:11.177]a grad student at UNL.
- [00:00:12.946]And today
- [00:00:13.446]we are talking about
- [00:00:14.531]energy use and AI.
- [00:00:17.317]And and I'm Guy Trainin.
- [00:00:18.301]I'm a professor here at UNL.
- [00:00:20.203]And I have been talking about
- [00:00:22.372]AI for quite a while.
- [00:00:23.590]And one of the topics of conversations
- [00:00:26.326]when we
- [00:00:26.826]talked about ethics of AI,
- [00:00:28.411]we talked about students
- [00:00:29.829]potentially cheating.
- [00:00:31.097]We talked about people using it
- [00:00:33.216]and not being careful enough.
- [00:00:34.534]We talked about the biases,
- [00:00:35.919]but one of the things that always came up
- [00:00:38.321]was the cost to the environment
- [00:00:40.957]through energy use.
- [00:00:42.459]And early on,
- [00:00:44.310]the training of them all
- [00:00:45.845]in every query was
- [00:00:49.382]considerably more expensive,
- [00:00:51.067]considerably more
- [00:00:52.352]consuming more energy
- [00:00:54.788]then
- [00:00:56.673]then a query with Google.
- [00:00:58.808]So one of the things we used to say
- [00:01:00.376]early on is like,
- [00:01:01.411]if you can Google it
- [00:01:03.430]and get the answer immediately, do that.
- [00:01:06.066]Don't use AI.
- [00:01:07.217]Well, two things have happened.
- [00:01:08.351]Now, if you Google it,
- [00:01:09.819]you also get an AI response.
- [00:01:11.738]So those things have really merged
- [00:01:14.924]into each other.
- [00:01:15.692]But the second thing that has happened
- [00:01:17.494]is that the costs of AI
- [00:01:19.345]over running
- [00:01:20.563]a query in AI
- [00:01:21.931]has really gone down.
- [00:01:23.466]This is a chart that,
- [00:01:26.035]Ethan Mollick,
- [00:01:26.736]who's one of the most important voices,
- [00:01:29.472]I think, in thinking about how,
- [00:01:32.942]AI especially generative AI,
- [00:01:35.495]LLMs are changing
- [00:01:37.263]what we're doing in business,
- [00:01:38.681]in education, in other areas.
- [00:01:41.601]If you have not
- [00:01:42.302]followed him on LinkedIn,
- [00:01:44.204]this is an opportunity.
- [00:01:45.555]He also produces content in other venues,
- [00:01:48.541]but he's worth
- [00:01:50.026]just seeing where he's at
- [00:01:51.461]and what he's publishing.
- [00:01:52.545]And in this case, he's
- [00:01:53.863]really focusing on the costs of AI.
- [00:01:58.802]And the unit here is a million tokens.
- [00:02:01.054]Tokens are not exactly, but mostly
- [00:02:04.107]words that the,
- [00:02:06.943]large language models use to process.
- [00:02:10.547]So that's their basic unit is a token.
- [00:02:13.233]And here you can see that in 2023
- [00:02:17.854]GPT-4 was consuming about
- [00:02:20.323]was costing about $50
- [00:02:23.326]for a million tokens,
- [00:02:24.811]which is quite big in energy consumption.
- [00:02:27.814]But you can see that Gemini 2.5 Flash,
- [00:02:31.601]for example,
- [00:02:32.385]which is the cheapest model,
- [00:02:34.454]is considerably better
- [00:02:37.240]in performance.
- [00:02:38.708]And that's the Y axis,
- [00:02:40.810]but is costing,
- [00:02:43.446]a fraction of a dollar per million.
- [00:02:47.951]That means that for $1 you are getting,
- [00:02:54.190]5 - 6 million, tokens,
- [00:02:56.893]which is an incredible savings.
- [00:03:00.246]And what we're seeing
- [00:03:01.664]watching these graphs
- [00:03:02.916]is there is an expectation
- [00:03:05.051]that this will keep getting cheaper.
- [00:03:08.171]So when we were starting out
- [00:03:11.140]on large language models,
- [00:03:12.725]the costs were very high.
- [00:03:14.327]We were very worried
- [00:03:15.695]and actually all those tech giants
- [00:03:18.398]they were thinking about starting up
- [00:03:20.166]nuclear plants and getting more energy,
- [00:03:23.169]and that has quieted down.
- [00:03:25.355]It's not completely gone
- [00:03:26.623]because we're seeing two things.
- [00:03:28.424]We're seeing
- [00:03:29.225]it's getting much, much cheaper,
- [00:03:30.677]but we're also seeing it,
- [00:03:33.663]the quantities of just use going up.
- [00:03:37.033]So while it is getting cheaper,
- [00:03:39.185]there are more and more users,
- [00:03:41.955]it means a few things
- [00:03:43.873]that I think are important.
- [00:03:45.008]One is that the energy load on the planet
- [00:03:48.945]is considerably smaller
- [00:03:50.263]than we ever anticipated,
- [00:03:51.631]and keeps getting smaller.
- [00:03:53.132]The second one,
- [00:03:54.100]and maybe even more important,
- [00:03:55.652]it means that,
- [00:03:58.171]AI
- [00:03:59.822]for every person on the planet
- [00:04:02.258]is within reach.
- [00:04:03.726]Because if this kept being expensive.
- [00:04:06.362]In the beginning,
- [00:04:07.096]they just released it for almost free
- [00:04:08.798]or for free
- [00:04:09.465]because they want to make sure
- [00:04:10.817]lots of people use it,
- [00:04:12.035]but eventually
- [00:04:12.752]they have to charge for it.
- [00:04:13.803]But if it's getting considerably cheaper,
- [00:04:15.905]it means that the amount
- [00:04:17.373]that they have to charge companies,
- [00:04:18.875]and especially individuals.
- [00:04:20.343]And I am thinking about
- [00:04:21.577]not just the United States,
- [00:04:22.929]but really
- [00:04:23.529]what's happening in Guyana,
- [00:04:24.764]what will happen in India,
- [00:04:26.015]what would happen across the globe.
- [00:04:28.234]And this is an opportunity
- [00:04:29.986]to make sure
- [00:04:30.603]that there's actually
- [00:04:32.021]equal access
- [00:04:33.256]across many places,
- [00:04:34.540]because the cost is going down
- [00:04:37.543]and not weighing on the environment.
- [00:04:39.696]So if you're concerned about energy,
- [00:04:42.465]keep being concerned about the energy.
- [00:04:44.167]But know that the costs for AI
- [00:04:47.971]to run are considerably lower.
- [00:04:50.890]Great.
- [00:04:52.692]And we'll see you next time
- [00:04:53.743]on AI in the Classroom.
The screen size you are trying to search captions on is too small!
You can always jump over to MediaHub and check it out there.
Log in to post comments
Embed
Copy the following code into your page
HTML
<div style="padding-top: 56.25%; overflow: hidden; position:relative; -webkit-box-flex: 1; flex-grow: 1;"> <iframe style="bottom: 0; left: 0; position: absolute; right: 0; top: 0; border: 0; height: 100%; width: 100%;" src="https://mediahub.unl.edu/media/25102?format=iframe&autoplay=0" title="Video Player: Tech EDGE, AI in the Classroom: AI and Energy Use" allowfullscreen ></iframe> </div>
Comments
0 Comments