The Download: Introducing the AI Energy Package

(

Plus, the Take It Down Act was signed into law.

This is the daily edition of The Downloadour weekday newsletter, which provides a daily dose on what’s happening in the worlds of technology.

The energy footprint of AI was calculated. Here’s a story you’ve never heard.

AI is a technology that’s power-hungry, and this has been well documented. There has been much less reporting about the extent of this hunger, the amount it is expected to grow in coming years, the source of the power, and who pays for it.

MIT Technology Review’steam of reporters and editor have been working to answer these questions for the past six month. The result is a unique look at AI’s energy and resources usage, where we are now, where we will be in the future, and why it is important to get it right.

The core of this package is a completely new line of reporting on the demands of inference – the way humans interact with AI, when we ask AI to create images or videos or answer text queries. Experts predict that inference will surpass the massive amount of energy needed to train new AI models. Here’s what we learned.

Here is what you can expect in the rest of the package.

– We were so shocked by what we discovered while reporting this story, that we put together a short on everything you need know about estimating AI’s energy and emission burden.

We went to the deserts of Nevada to see how this energy thirst is manifested. Data centers in an industrial complex the size of Detroit require more and more water to run their processors.

+ In Louisiana, where Meta is planning its largest data center ever, we expose the dirty secrets that will fuel its AI aspirations–alongside those of many other companies.

– Why the clean energy promise to power AI data centers using nuclear energy will remain elusive for a long time.

– But it’s all not doom and misery. Examine the reasons for optimism, and why future AI systems may be less energy-intensive than those of today.

AI is better at persuading than humans

News: Every day, millions of people argue online. However, only a small percentage of those arguments change people’s minds. New research suggests large language models (LLMs), especially when given the ability adapt their arguments based on personal information about individuals, might do a much better job. The findings suggest that AI could be a powerful tool to persuade people, for good or bad.

Big picture: These findings are the latest of a growing number of research studies that demonstrate LLMs’ persuasion powers. The authors warn that they show how AI can craft sophisticated and persuasive arguments if it has even minimal information about humans with whom they are interacting. Read the complete story.

–Rhiannon William

The impact of AI on courtrooms

There have been a lot of stories about AI in courtrooms over the past couple weeks. You may have heard of the family of a deceased victim of road rage who created an AI avatar to be shown as an impact statement. This is possibly the first time that this has been done in America. Legal experts say that there is a much bigger and more consequential controversy brewing. AI hallucinations appear more and more often in legal filings. It’s also starting to irritate judges. Take a look at these three cases. Each one gives us a glimpse of what we can expect as lawyers embrace AI. Read the complete story. This story was originally published in The Algorithm – our weekly newsletter about AI. Sign up for our weekly newsletter to receive stories like this first or

The must-reads

I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology.

Donald Trump signed the Take It Down Act
It criminalizes non-consensual images, including deepfakes. (The Verge)
+ Tech platform will be forced to remove this material within 48 hours after being notified. (CNN)
+ This is only the sixth law he has signed during his second term. NBC News

There’s a buyer now for 23andMe.
Pharma company Regeneron has swooped and offered to keep it operating. (WSJ $)
+ What is the value of your genetic data? $17. 404 Media
+ Regeneron has promised to prioritize the security and ethical use for this data. TechCrunch

Microsoft is adding Elon Musk’s AI models into its cloud platform.
Is that a good thing? Musk wants to sell Grok. The Information ($)

Autonomous cars that are trained to react like human drivers cause fewer road accidents
According to a study, they are more cautious around pedestrians, motorcyclists and cyclists. (FT $)
+ Waymo expands its robotaxi operations outside of San Francisco. Wayve’s driverless car will face one of its biggest challenges yet. MIT Technology Review

The 5 Hurricane Season is coming
Doge cuts mean we are less prepared. (The Atlantic $)
+ The COP30 could be in crisis even before it begins. New Scientist($)

Telegram gave over data of more than 20,000 users in the first three months 2025. (404 Media ]

]7 GM stopped exporting cars into China
Trump’s tariffs put an end to their export plans. (NYT $)

Eight Blended Meats are on the Rise
Up to 70% of these new meats are made from plants, and consumers love them. (WP$)[194542]+ Alternative Meat could help the Climate. Will anyone eat this? SAG-AFTRA doesn’t like the AI Darth Vader voiced by Fornite


The union has filed a charge of unfair labor practices against Fortnite’s creators. (Ars Technica, )
+ How Meta AI and companies recruited striking actors for AI training. MIT Technology Review

10 Thanks to a simple prompt, this AI model can quickly build Lego structures. Fast Company ($)

Today’s Quote

Platforms have no incentive to ensure that what comes through their system is non-consensual intimacy imagery.

Another thing

Do friends have electricity?The difference between humans in the real world and machines is easy to discern. While machines excel at things that adults find difficult, like playing world-champion chess or multiplying large numbers, they find it difficult to do things a five-year old can do easily, such as catching the ball or walking in a room without bumping things.

The fundamental tension between what is difficult for humans and what is easy for machines is at the core of three new books that explore our complex and often fraught relationship to robots, artificial intelligence, and automation. They force us reimagine everything from friendship and romance to work, health, and home life. Read the complete story.

–Bryan Gardiner.

You can still enjoy nice things.

A place to relax, have fun and distract you. (Do you have any ideas? Drop a line (or skeet them at me).

+ William Goodgewho ran across Australia in 35 days deserves congratulations!
A British horticulturist created a garden for dogs at this year’s Chelsea Flower Show The Netherlands loves a good sidewalk garden
+ Did You Know the T Rex Is A North American Hero? Me neither

www.aiobserver.co

More from this stream

Recomended