298
We did the math on AI’s energy footprint. Here’s the story you haven’t heard.
(www.technologyreview.com)
this post was submitted on 21 May 2025
298 points (96.6% liked)
Technology
70415 readers
3143 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related news or articles.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Historically AI always got much better. Usually after the field collapsed in an AI winter and several years went by in search for a new technique to then repeat the hype cycle. Tech bros want it to get better without that winter stage though.
AI usually got better when people realized it wasn't going to do all it was hyped up for but was useful for a certain set of tasks.
Then it turned from world-changing hotness to super boring tech your washing machine uses to fine-tune its washing program.
Like the cliché goes: when it works, we don't call it AI anymore.
The smart move is never calling it "AI" in the first place.
Unless you're in comp sci, and AI is a field, not a marketing term. And in that case everyone already knows that's not "it".
The major thing that killed 1960s/70s AI was the Vietnam War. MIT's CSAIL was funded heavily by DARPA. When public opinion turned against Vietnam and Congress started shutting off funding, DARPA wasn't putting money into CSAIL anymore. Congress didn't create an alternative funding path, so the whole thing dried up.
That lab basically created computing as we know it today. It bore fruit, and many companies owe their success to it. There were plenty of promising lines of research still going on.
I wish there was an alternate history forum or novel that explores this scenario.
The spice must flow
Each winter marks the beginning and end of a generation of AI. We are now seeing more progress and as long as there is no technical limit it seems that its progress will not be interrupted.
What progress are we seeing?
In what area of AI? Image generation is increasing in leaps and bounds. Video generation even more so. Image reconstruction for games (DLSS, XeSS, FSR) is having generational improvements almost every year. AI chatbots are getting much much smarter seemingly every month.
What’s one main application of AI that hasn’t improved?
Which chatbots are getting smarter?
I know AI has potential, but specifically LLMs (which most people mean when talking about AI) seem to have hit their technological limits.
Copilot, ChatGPT, pretty much all of them.
Smarter how? Synthetic benchmarks?
Because I've heard the opposite from users and bloggers.
So you want me to provide some evidence that it's getting smarter, but you can't provide any that it's getting worse other than anecdotal evidence?
What evidence would you accept?
Any proof that we have moved past the current architecture.
What does "architecture" mean in this scenario?
Any significant shift in the model, or a complete restructuralization of the approach.
As it is, it won't grow anywhere.
So you’ve got access to all this stuffs source code and know what has and hasn’t changed with every update?
No, if there was any major breakthrough, it would be advertised everywhere.
They’re constantly advertising updates to these chat bots in what they can do.
Small incremental updates on little tasks mean nothing, the underlying issues are still the same.
It has no intelligence, and as such carries big risks.
No, they are spreading lies about shit that doesn't matter as to not lose the hype.
If anyone made any significant advance, they would be all over the world.
You’re not backing this up with anything. Those of us who use them know they’ve been making big updates regularly.
You're not backing anything up either, just 'my experience'.
Do you want me to just link to Microsoft and OpenAI’s pages about their AI chatbots updates?
Advanced Reasoning models came out like 4 months ago lol
Advanced reasoning? Having LLM talk to itself?
Yes, which has improved some tasks measurably. ~20% improvement on programming tasks, as a practical example. It has also improved tool use and agentic tasks, allowing the llm to plan ahead and adjust it's initial approach based on later parts.
Having the llm talk through the tasks allows it to improve or fix bad decisions taken early based on new realizations on later stages. Sort of like when a human thinks through how to do something.
Lul yes but no, but they are clearly better at many types of tasks.
For example? Citations?
Pretty sure these "tasks" are meaningless metrics made up by pseudo-scientific grifters.
Small bits of code, language related tasks, basic context understanding, not metrics I have literally measured simply noticed has improved compared to non reasoning models in my homelab testing. 🤷♂️
AlphaFold 3 which can help in the prediction of some proteins. Although it has some limitations, it cannot be used in all cases, only in what it can perform without any problem.
NVL72 will be enormously impactful on high end performance.
The issue this time around is infrastructure. The current AI Summer depends on massive datacenters with equally massive electrical needs. If companies can't monetize that enough, they'll pull the plug and none of this will be available to general public anymore.
This system can go backwards. Yes, the R&D will still be there after the AI Winter cycle hits, but none of the infrastructure.
We'll still have models like deepseek, and (hopefully) discount used server hardware
Local models aren't going anywhere. You can hoard them if you're worried. People freely tweak them to do some weird shit. There's whole independent websites for sharing these tweaks.
The only infrastructure that's at risk is the part where corporations burn billions of dollars hoping the really really big models will turn a profit somehow. The power and capability of small models is basically whatever the big boys did a year ago - and a ton of money has been spent on improving the efficiency of training. Tightening the belt is not going to make things worse. It's certainly not going to make this whole new kind of software un-happen.
Local models aren't what's driving the current AI Summer. No billionaire will be quoted in the New York Times saying how great local models are.
Why on Earth is that your metric?
The money addicts desperately want big models to win because that means less competition and more subscription revenue. But that's three delusions in a row. Mainframes are basically never a sensible business model. Models costing mere tens of millions already challenge billion-dollar projects. And as you say, infrastructure is expensive, so most people want software that runs within arm's reach. There's every reason these cloud services bleed money.
Local models are what's making all the weird shit you see online. Certainly the porn is not made by typing stuff into Facebook. And again: whatever cutting-edge advances happen in moonshot-priced, datacenter-sized models, they're soon replicated on any recent desktop. If that same blank-check experimentation had only bothered with small models, do you think they'd be further behind or further ahead?
What's driving this AI Summer is a horde of ultranerds, funded by these rich idiots lying to each other. A lot of them will still be doing this if the money dries up. A lot of them already do it for free. See aforementioned tweaking and sharing. If that has to expand to training new models from scratch, expect them to get even smaller, and rely on experimentation more than brute force. Like BitNet reducing all weights to trinary. If it still takes a mountain of GPUs... well, SETI@home couldn't promise people interactive anime waifus.
That's part of why they installed Donald Trump as the dictator of the United States. The other is the network states.
Historically "AI" still doesn't exist.
Technically even 1950s computer chess is classified as AI.