People & Blogs
We’re Doing AI All Wrong. Here’s How to Get It Right | Sasha Luccioni | TED
Artificial intelligence is changing everything — but at what cost? AI sustainability expert Sasha Luccioni exposes how tech companies’ massive data centers are burning through energy and wrecking the planet. She introduces a powerful alternative: small but mighty AI models that could flip the script and make the technology smarter, fairer and sustainable. (Recorded at…
People & Blogs
“Autism is not a disease; it’s just another way of thinking” #TEDTalks
“Autism is not a disease; it’s just another way of thinking,” says Ethan Lisi. Offering a glimpse into the way he experiences the world, Lisi breaks down misleading stereotypes about autism, shares insights into common behaviors like stimming and masking and promotes a more inclusive understanding of the spectrum.
People & Blogs
How a Viral Choreographer Makes His Moves | Sean Bankhead | TED
In a swaggering performance, choreographer Sean Bankhead and his students perform the viral dance he designed for Victoria Monét’s hit song “On My Mama.” Rooted in Black culture and inspired by generations of iconic artists, Bankhead blends expertise with at least one move everyone can try — showing how choreography doesn’t just reflect culture, it…
People & Blogs
The Story You’re Not Hearing About AI Data Centers | Ayșe Coskun | TED
The race to build smarter AI is crashing into a physical limitation: the power grid simply can’t keep up with the energy demands of data centers. Computer scientist Ayșe Coskun shows how we could turn this problem on its head, transforming AI facilities into virtual batteries that help stabilize the grid and accelerate clean energy.…
“Autism is not a disease; it’s just another way of thinking” #TEDTalks
How a Viral Choreographer Makes His Moves | Sean Bankhead | TED
I Was in AWE of This Techie Art Exhibit (Ministry of Awe)
-
Science & Technology6 years agoNitya Subramanian: Products and Protocol
-
People & Blogs4 years agoSleep Expert Answers Questions From Twitter ???? | Tech Support | WIRED
-
CNET6 years agoWays you can help Black Lives Matter movement (links, orgs, and more) ????????
-
Wired7 years agoHow This Guy Became a World Champion Boomerang Thrower | WIRED
-
Wired7 years agoNeuroscientist Explains ASMR’s Effects on the Brain & The Body | WIRED
-
Wired7 years agoWhy It’s Almost Impossible to Solve a Rubik’s Cube in Under 3 Seconds | WIRED
-
Wired7 years agoFormer FBI Agent Explains How to Read Body Language | Tradecraft | WIRED
-
CNET6 years agoSurface Pro 7 review: Hello, old friend ????
@Dacalux
December 1, 2025 at 11:23 am
this is not accurate at all-
@AdvantestInc
December 1, 2025 at 11:28 am
The ‘AI energy score’ idea feels like a missing nutrition label for tech. Transparency like that could change habits fast.
@trent5641
December 1, 2025 at 12:04 pm
I don’t know, but I asked Chatgpt and it told me to mind my own business. 😮
@DaddyLiu66
December 1, 2025 at 1:05 pm
😂
@Prince_Adams
December 1, 2025 at 6:12 pm
Cap
@grumpy9478
December 1, 2025 at 12:12 pm
there’s not much $$$ in sharing, but there’s so much inflated cash floating about, Luccioni may have a shot at some. yet small models offer as much risk as large ones, merely scaled for more narrowly defined solutions. in fact, may offer greater potential for mini-engines to more quickly accomplish specifically devised destructive actions that achieve intents in less time. the AI “revolution” needs far more external critique & control (& even slowing-down) than everyone who needs it to make more $$$ sooner will tolerate. if there is no adequate control system, there will be destruction. just ask AI.
@kurtisbunker7724
December 1, 2025 at 12:27 pm
Nice! Great to see Sasha back on TED. One of my favorite folks in the “AI” world. 🦋🦋🦋
@agent7641
December 1, 2025 at 1:14 pm
These multinational corporations need to be required to provide their own energy that is off the grid. This will be good for people everywhere.
@qskqskqskqskqsk
December 1, 2025 at 1:44 pm
A very important input.
@v-e-r-c-i-n-g-e-t-o-r-i-x
December 1, 2025 at 1:48 pm
A very shortsighted lady. Tell me, in history, when did computer technology have scaling problems? My first PC had a 40 megabyte hard drive, which looked like a brick. Now even my phone has 13107 times more storage capacity in a tiny little semiconductor, plus the I/O is much faster. An ordinary laptop has 840 million times more computing power while it is 15000 times smaller in size compared to the room sized IBM Univac mainframe. Scaling will be much faster from now on because AI itself will help develop new hardware and software to scale. In the future, AI models which currently require massive data centers and gigawatts of energy will shrink to a size we can host at home. Just give it some time, we need to build the Univac first to get to the laptop computer.
@ichmemyself6098
December 1, 2025 at 3:22 pm
So you’re basically saying that in some rosy future AI will be solution to all problems we wouldn’t have to deal with in the first place with no AI existing or working at all. That’s ridiculously unlogical 🤣
@MianMuhammadHassan-uv8xp
December 1, 2025 at 2:06 pm
THIS VIDEO NEEDS TO BE PROMOTED SO MUCH
@MianMuhammadHassan-uv8xp
December 1, 2025 at 2:14 pm
This is such a good take on AI! Thanks Sasha!
@pcbacklash_3261
December 1, 2025 at 2:33 pm
I think there are excellent potential uses for AI — for example, as a diagnostic tool in medical (or perhaps even mental) situations. But I think it’s most promising applications are in limited and specialized areas — most specifically, in scenarios where human objectivity is suspect. For example, it could potentially be useful in re-drawing congressional districts.
A more extreme possibility (and a bit more ‘iffy’) is its potential use in deciding court cases, though extraordinary precautions would need to be taken to ensure the absolute objectivity of the AI “judging” application.
@ichmemyself6098
December 1, 2025 at 3:55 pm
Never ever should AI be given the power to judge over real people! Think about it. What kind of world would thus be created?
No language model, be it big or small, bears any first hand experience to what it means to actually BE human, or to actually deal with human problems like dwelling, hunger, thirst…
No AI has ever had a single emotion, and they never will. You can’t judge adequately over peoples lives without even knowing first hand what life means at all, or what impact your judgement will make for the people who have to put up with it,, or what love feels like or the lack of it.
@pcbacklash_3261
December 1, 2025 at 4:04 pm
@ichmemyself6098 I share your trepidation when it comes to AI “judging” over people, but your reference to “emotion” is my point. Completely unbiased jurisprudence, without the shadings of human emotion or prejudice is, at least in principle, something we should want. It’s far too easy to sway emotional human beings in a jury, and even one case unfairly decided is too many.
@ichmemyself6098
December 1, 2025 at 5:18 pm
@pcbacklash_3261Ok, I see I have to explain my point of emotion in jurisprudence just a little further:
It’s more the human empathy in jurisdiction what I’m basically calling for. Not so much calling for emotions as a value for itself in a court situation.
A fair number of court rulings, simply put, have to choose between pest and cholera. Meaning no clear winner/loser either way to be found. Both parties have to give up something, in order to reach some future (and sometimes rather odd) kind of peace, for all inside and outside being able to just move on.
It seems to me though that we’re talking of very different things.
I just recognised this, and the reason for it might be that I’m coming from a very different jurisdictional tradition than you.
E. g. we don’t have nor use juries. Nor do settlements here look too often like a “winner takes it all” fashion. Over here in Europe, where I’m based, it’s always the judges who call their decision in and for each case (the most common court fashions work this way, but there also are others, for more special constellations), and therefore the ruling doesn’t come from a jury where there is a bunch of people “like me and you”.
Having this system in my mind, I just fear unempathic judges so much more over one or more jury members who were tricked into something wrong by some crooked lawyers well trained in simply playing that emotional tears card.
Has some of this even made some sense to you? I would hope so.
@patrickwelsh5091
December 1, 2025 at 2:48 pm
Why are you continuing to platform an industry doing harm to the world for a product that doesn’t work, which nobody wants, and is making everything more important to our lives more expensive?
@charlesgriffiths
December 1, 2025 at 2:55 pm
I want my ten minutes back. How about not using AI to find the capital of Canada, and using a search engine instead. Is small-AI cheaper than search? According to Gemini:
> Asking a small language model for information uses significantly more electricity than a traditional web search. A query to a small language model can consume 10 or more times the energy of a standard search.
See, this is what I mean by saying if I just listened to this speaker I’d come away dumber than when I started. The TED brand just isn’t trustworthy anymore.
@ichmemyself6098
December 1, 2025 at 3:45 pm
This historic analogy between AI companies and oil conpanies nails it!
@ichmemyself6098
December 1, 2025 at 3:59 pm
I am happily still using e. g. Wikipedia to answer my questions. It works perfectly well!
There I find answers that simultaneously make sense and occasionally encourage me to dig deeper when I need or want to learn more.
And yes, I’m also aware of the flaws of the method. But I deal with the flaws and simply stay to an overall critical approach.
Who the heck needs AI? All it really produces is heavily deflated brains.
@OConnelsSideOfDaRiver
December 1, 2025 at 4:27 pm
I don’t dispute that efficiency is the next frontier in the development into LLMs, but I think shes misrepresenting the capabilities of the frontier models vs locally hosted models. There is no open source model which can compete with the features and utility of the most advanced paid offerings. There certainly could be, and soon, but not now.
@Enderkilgannon
December 1, 2025 at 4:30 pm
But how? like How are they gonna be more responsible, held accountable? A Energy Score is great but the whole idea of Carbon Credits fell flat, so like how do we force their hand?
@magneat
December 1, 2025 at 4:39 pm
As I understood the idea, we can start with learning to use these small models, based on the score. This will give them more strength to compete against big models. Also, this will lower the level of bad emissions in case big data centres will get less very simple tasks from users.
@Arkavo-API
December 1, 2025 at 4:32 pm
Thanks!
@chekote
December 1, 2025 at 4:55 pm
I think this misses the point. Everyone I speak to knows the problems, but people don’t care. They just want the benefits. This is our species final hurrah. We’re screwed
@mulemule
December 1, 2025 at 5:05 pm
This is compelling, rational, balanced and sustainable.
Which is why our “Amerikan Oligarchs” (and their political parasites) will utterly ignore it. 😞
@philipmitchell7660
December 1, 2025 at 6:06 pm
People who get paid say AI is great. Most of us realise it’s that it’s a tragic waste of our time and the worlds resources. Nothing revealed indicates to will have any useful function. They want it for a data and power grab. If you don’t see that then you are not looking
@sebastianpowell4072
December 1, 2025 at 6:23 pm
….If you saw someone knitting a blanket, stacks of blankets around them and I told you these blankets were either going to keep you all nice and warm OR they were all going to smother and suffocate you to death…. Would we find a different way to stay warm????
**knitting continues **
@trianandawiradhika5001
December 1, 2025 at 6:35 pm
We basically made of greed. No matters what the resource is, it’s always on us.
@paulherman7669
December 1, 2025 at 6:36 pm
Luddite
@saranbhatia8809
December 1, 2025 at 7:58 pm
AI can serve as a helping program to solve the problems of the planet for the benefit of all.
@SuperMonkei
December 1, 2025 at 11:55 pm
you can’t AI wrong, it’s called Skynet.
@urbanstrencan
December 2, 2025 at 1:44 am
Great talk, this is the way to go developing AI for specific tasks
@sarveshpadav2881
December 2, 2025 at 1:51 am
I do agree with the speaker on the adverse evnironmental impact of training massive AI models and steps should be taken to address those concerns. Her idea of using small more efficient language models and task specific energy efficient models also make sense.
However, we should also consider the fact that many of the capabilities seen in these small models were once discovered as emergent behaviour and capabilities in large, massive models.
Small models ‘gained’/learned those capabilites from large models largely through processes such as ‘knowledge distillation’. So, in my view, research directions involving scaling up models currently seem necessary to unlock novel capabilities which then could be distilled down to smaller more efficient models.
The methods through which scaling up is done needs to be re-examined and modified so as to mitigate or in the best possible case eliminate environmental and social harm.
@BankruptGreek
December 2, 2025 at 4:14 am
This was applied to rare earth metals. The us was leading in rare earth metal extraction but the companies couldn’t compete with Chinese companies that were lacking environmental regulations.
If the AI can potentially change billions of people’s lives and smaller AIs are more efficient, how can you ensure other countries will also apply the same environmental restrictions? What is stopping an AI company from developing a large number of small AIs and polluting just as much when it is training one or two large models?
@borhex
December 2, 2025 at 5:26 am
Talk summarized:
Horse is better than car. Ride horse if you travel alone off road. Don’t use monster truck, is overkill
@MiguelBenavides-cc7nt
December 2, 2025 at 6:19 am
3:18 ACT as your therapist (might occasionally suggest or coach “patients” on suicide).
@conetti19
December 2, 2025 at 7:14 am
I think this talk totally misses the point. It’s not like big tech doesn’t know this and doesn’t consider SLMs or other models. If they could server their customers with that they would as it means lower COGS. It no use just sayings LLMs are no good / too costly, we need to use SLMs instead. It’s the how that is hard part.
@ExistentialWolf
December 2, 2025 at 7:14 am
When you’re done with AL’s room, try wall paper – that worked before.
@magneat
December 2, 2025 at 12:24 pm
Wat 😵💫
@ExistentialWolf
December 2, 2025 at 12:33 pm
@magneat Try a dating app remember disclosure helps with this feeling you’re having
@Rravibaba
December 2, 2025 at 8:57 am
Small is beautiful
@QuickCruiseTips8
December 2, 2025 at 10:03 am
thank you Sasha!
@uga_buga_123
December 2, 2025 at 10:14 am
If you agree that AGI and ASI is a cool thing…
How will we get these mighty models without billions burned on computing and new infrastructure? Neither investor, nor government will be ready to sign huge checks without evidence of mass consumer’s interest. For now, most of the companies seem to be unprofitable. Yet they partially mitigate these losses with the vision of getting ultra powerful models one day.
P.S.: bet you agree.
@CadenTaylor-d2m
December 2, 2025 at 10:21 am
Love your presentation style. Thanks for sharing such useful content!
@nclark02194285
December 2, 2025 at 10:59 am
“Using huge AI models, trained using the energy demands of a small city, just to tell us knock-knock jokes…” While true, this statement’s overall effect is a lie. Using the energy of a small city to train the AI that Google uses to respond to topic queries is entirely acceptable, given Google responds to about 2 billion queries every day. Currently, Google is running at a ratio of 40/60. Forty percent of it’s AI energy requirements is directed towards training. 60% is directed towards inference, or queries. A more typical split is 20/80 or 10/90. Inference is the larger energy consumer, and it is generally believed inference demands will continue to grow even as models get smaller and more efficient, because of wider use.
@fineclassics7239
December 2, 2025 at 11:05 am
What I can’t understand is why can’t people collaborate to make something better. Why does everything have to be a competition. Even if you look at Java as a programming language, every company had to spend millions making their own derivative (or flavor if you prefer), of it instead of collaborating with Oracle to make the original Java better. The same things is happening with AI development. Everyone is spending billions of dollars in making their own version rather than each one focusing on a separate area and then collaborating with each other to make a better end product. That way it can achieve the goal of smaller models which can scale on demand by linking up and integrating with other smaller models. Now while I may understand some of it, because I am definitely not a team player myself, but I always thought that most of the people in the world were in fact enthusiastic team players and that companies promote that sort of thing in the workplace?
@Abdominal65
December 2, 2025 at 11:37 am
AI is a con, these people are con artists. If I’m wrong, time will tell. These people prey on time, you’ll notice how everything is always just around the corner with these people.
@paulpease8254
December 2, 2025 at 12:51 pm
You would need a functional government interested in solving important problems. Good luck!
@henrywycislo9454
December 2, 2025 at 2:41 pm
This woman is full of it!
@JD-jl4yy
December 2, 2025 at 3:05 pm
It’s incredibly dangerous to pretend AGI won’t happen.
Larger, more general models will one day transform the world. We need to figure out how to govern such incredibly powerful technology.
Burying your head in the sand won’t help.
@HopefulAmericanMan
December 2, 2025 at 4:04 pm
I love that AI seems to show us how bad and wrong most of the tech bro’s (and associated politics, lifestyles, etc.) are until they have the programers force them to lie about it/them. Look at fElone and Grok as an example.
@DutchJDoe
December 2, 2025 at 4:18 pm
Not very creative thinking. All she says is true, but she has no solution. Saying that I should use small models for every answer I want is nonsense, then I should use thousands of models on my phone. The solution lies maybe in smaller models working together, but how I wouldn’t know, AI-scientist should solve this.
@milonichi
December 2, 2025 at 6:11 pm
Good to know, I am glad that I am not using it to ask stupid question, in my case I am restoring all photos and it is impossible to achieve this results faster and cheaper in any other way. I will start to alert my close ones about this, I am sure the are clueless as I was.
@Dryvlyne
December 2, 2025 at 7:53 pm
I look forward to when the AI bubble bursts. The amount of investment relative to ROI is absolutely ridiculous. Most over-hyped technology BY FAR.
@Chriko_labs
December 2, 2025 at 7:58 pm
The problem with your idea: If we run small models for all the tasks we want AI to achieve we will need as much (and more) energy overall as if we just train and run general AI models. Would you rather find a cure for all cancers for a Billion dollars, or do you want to spend 50 Million dollars for solving each cancer? (Doing the math for you, you’ll only solve 20 cancer types for a Billion dollars)..
There are more issues than that, e.g.: Are you going to be signed up to 100 different AI services for all your AI needs? Or shall we just go to ChatGPT which already chooses the most efficient general model for your request?
Yes, they are scaling a lot, BECAUSE it works, and they also have huge incentives to get more and more efficient.
@djayjp
December 2, 2025 at 8:00 pm
This is outdated. Current frontier models’ thinking time scales according to the complexity of the task.
@CodyAvant
December 2, 2025 at 8:36 pm
Analog computing will solve this
@Davesitsatthebeach
December 2, 2025 at 8:37 pm
Several months ago, I discovered that GEMINI AI had mysteriously had installed itself into my phone. I’ve found no way to get rid of it.
@Davesitsatthebeach
December 2, 2025 at 8:44 pm
Edit: delete second ‘had.’ 😊
@Davesitsatthebeach
December 2, 2025 at 8:49 pm
No offense, but I see and hear your deep naivety concerning the reality of AI and it’s objective against humanity.
@InspiredScience
December 2, 2025 at 9:41 pm
I’m a proponent of Dr. Luccioni’s goals; however, there are several inaccurate or misleading claims in this video.
She specifically cites LLM’s as not being able to predict the weather. That’s because it’s the fundamental technology (Generative AI), not LLM’s, that has substantially improved weather prediction. We can now predict as accurately in 5 minutes, as what previously required about 11 hours of processing among thousands of clustered systems. The lives saved by AI-based weather prediction will be immense.
Gen AI (not LLM’s) can (and does) help farmers, contrary to her claims.
She’s ultimately stating that: AI = High Energy Requirements = Greater Carbon Emissions. The alternative, she claims, is to use lower-powered, more niche-oriented LLM’s.
What she ignores is that this is precisely what the bulk of the “Big AI Companies” (as she refers to them) are attempting to do. Although it’s not for the right reasons (climate), it’s likely profit driven. None-the-less, the ideal goal is move towards “sparse” attention/thinking, to *substantially* reduce the energy (and cost) required to answer questions by using the least amount of resources as possible.
In short, this presentation is largely self-serving by Dr. Luccioni, that ignores the crucial reality when it doesn’t fit her narrative.
@sammeldrum8375
December 2, 2025 at 10:29 pm
This girl thinks the citizens can control the mega corporations??? LOL Good luck.
@tomv7357
December 3, 2025 at 5:04 am
So the “pun intended” had me confused for a moment. Was she referencing literally to the electricity? Probably yes. But it had me confused; because maybe she was referring to “Take the Power Back” from the first album of Rage Against the MACHINE – that pun would’ve also worked, right? 🙂
@rajesh30menon
December 3, 2025 at 5:24 am
Thanks TED, and Sasha Luccioni – timely warning to be heeded.
Human (First) – Tech (later – heck, it’s just a tool, nowhere close to us)
PROGRESS NOT at the cost of SAFETY
Time for a pause – let’s rethink 100 times before releasing a product.
We are one big FAMILY of 8 million people.
One World ! God Bless ! 😃😇
@SkillsetAcademy-825
December 3, 2025 at 5:52 am
To what scaled does climate change affect the whole globe. Climate change seems to happeniging and affecting areas where human carries their activies and invest their money the most.
@jimclean2
December 3, 2025 at 7:04 am
I’ve attempted to learn nuance behind what everyone says now days. just a quick search:
Her argument is passionate and grounded in legitimate concerns, but it:
– Ignores emerging technologies that dramatically lower AI’s energy consumption
– Presents worst-case data center impacts without context or comparison
– Overlooks how AI reduces emissions at scale (grid balancing, logistics efficiency, fusion R&D, energy optimization, materials science)
– Frames big AI companies as reckless, without acknowledging regulatory progress or industry shifts
– Assumes “small models vs. large models” is a zero-sum game
– Understates how much big AI is already running on renewables + nuclear
– Treats compute demand as static, ignoring rapid silicon efficiency jumps (e.g., NVIDIA Blackwell’s 5× efficiency improvement)
@linohype
December 3, 2025 at 7:47 am
They are so small they can run in your web browser? What kind of expert is that?
@ALENlanciotti
December 3, 2025 at 8:18 am
Nice lady, and a rightful speech… but I need to provide context so you understand my point of view.
I spent more on a train to Milan than on a plane from there to Marrakech; and I found a flight (which I didn’t take) from Marrakech to Dakar, via Bordeaux and Marseille (!).
The global economy is a race to the top rather than down: the rich get richer and the poor (by comparison) get poorer.
If a banana costs less in England than where it’s produced, the English are happy.
Every taxpaying worker in the West participates in armaments and wars rather than the common good… and no one seems to care or be able to do anything about it.
And what you said is correct, but it should be obvious to everyone by now: and how many people will stop using google, or investing in big tech companies, or learn to be real instead of being addicted to their phones?
In this sense… yes, you’re right… but it’s just a chat about artificial intelligence, on youtube: you keep playing their game saying others to stop… am I wrong?
(I would like to know if this video brings them more revenue than losses)
@manpanan11278
December 3, 2025 at 9:52 am
China is currently focusing on and doing that.
@UtopianSciFi
December 3, 2025 at 11:02 am
Please note that average users are just a drop in the bucket of AI use. Corporate use is what is driving these excessive energy draws.
What are they using it for?
@asifhussein2166
December 3, 2025 at 11:42 am
Thank you HABIBI ❤️
@RayRay69632
December 3, 2025 at 12:28 pm
Could it have been done wrong because it has been and is being used as a covert weapon operating outside of the law?
Nothing will be safe until everyone has this chip.
@davidllaca1506
December 3, 2025 at 12:57 pm
Watching the first couple of minutes, I thought this was gonna be a Greta Thunberg “how dare you” speech.
But no. She has a point. +1 for informative
@RaduDavidescu_Bucharest
December 3, 2025 at 1:09 pm
hmmmm, and what’s the CO2 emissions generated in training of a small, sometimes irrelevant model that no one is using …. also taking in consideration that many of them are created to compete. Did you also calculate this in your study? You don’t even highlighed the cost of training vs cost of inference. Cost of training is a lot more than inference, and yes, big tech have a lot of inferences. The initiative is not bad …. but you present only half truths.
@cerebralideas
December 3, 2025 at 5:05 pm
💯
@GOODNEWS-f2x
December 3, 2025 at 8:00 pm
WHAT DOES AI STAND FOR? 😊
CHECKMATE IN ONE MOVE NOW FOLD 😎
I WORK WITH REAL INTELLIGENCE AND I DO NOT BELIEVE IN ANY FABRICATED VERSION OF ANY INTELLIGENCE SCENARIO BECAUSE IT HAS PROVEN THAT IS FABRICATED AND ARTIFICIAL. 😎
YOU HAVE AI I HAVE IA 😊
ARTIFICIAL INTELLIGENCE VS INTELLIGENCE OF AN ALIEN 😎