Now again under Rose’s control, the new Digg is creating a place for people to socialize and connect online within communities, similar to Reddit, but it has different ideas about how such a platform should work at a time when bots are nearly indistinguishable from humans.
Read more YouTube CEO Neal Mohan told employees about the program via an internal memo on Wednesday.
Read more In this exclusive subscirber-only ebook you’ll learn how the emissions from individual AI text, image, and video queries seem small—until you add up what the industry isn’t tracking and consider where it’s heading next.
by James O’Donnell and Casey Crownhart May 20, 2025
Table of contents
- Part One: Making the model
- Part Two: A Query
- Part Three: Fuel and emissions
- Part Four: The future ahead
Related Content:
Four years is a lifetime when it comes to artificial intelligence. Since the first edition of this study was published in 2021, AI’s capabilities have been advancing at speed, and the advances have not slowed since generative AI’s breakthrough. For example, multimodality— the ability to process information not only as text but also as audio, video, and other unstructured formats—is becoming a common feature of AI models. AI’s capacity to reason and act autonomously has also grown, and organizations are now starting to work with AI agents that can do just that.
Amid all the change, there remains a constant: the quality of an AI model’s outputs is only ever as good as the data
that feeds it. Data management technologies and practices have also been advancing, but the second edition of this study suggests that most organizations are not leveraging those fast enough to keep up with AI’s development. As a result of that and other hindrances, relatively few organizations are delivering the desired business results from their AI strategy. No more than 2% of senior executives we surveyed rate their organizations highly in terms of delivering results from AI.
To determine the extent to which organizational data performance has improved as generative AI and other AI advances have taken hold, MIT Technology Review Insights surveyed 800 senior data and technology executives. We also conducted in-depth interviews with 15 technology and business leaders.
Key findings from the report include the following:
• Few data teams are keeping pace with AI. Organizations are doing no better today at delivering on data strategy than in pre-generative AI days. Among those surveyed in 2025, 12% are self-assessed data “high achievers” compared with 13% in 2021. Shortages of skilled talent remain a constraint, but teams also struggle with accessing fresh data, tracing lineage, and dealing with security complexity—important requirements for AI success.
• Partly as a result, AI is not fully firing yet. There are even fewer “high achievers” when it comes to AI. Just 2% of respondents rate their organizations’ AI performance highly today in terms of delivering measurable business results. In fact, most are still struggling to scale generative AI. While two thirds have deployed it, only 7% have done so widely.
This content was produced by Insights, the custom content arm of MIT Technology Review. It was not written by MIT Technology Review’s editorial staff. It was researched, designed, and written by human writers, editors, analysts, and illustrators. AI tools that may have been used were limited to secondary production processes that passed thorough human review.
This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology.
DeepSeek may have found a new way to improve AI’s ability to remember
The news: An AI model released by Chinese AI company DeepSeek uses new techniques that could significantly improve AI’s ability to “remember.”
How it works: The optical character recognition model works by extracting text from an image and turning it into machine-readable words. This is the same technology that powers scanner apps, translation of text in photos, and many accessibility tools.
Why it matters: Researchers say the model’s main innovation lies in how it processes information—specifically, how it stores and retrieves data. Improving how AI models “remember” could reduce how much computing power they need to run, thus mitigating AI’s large (and growing) carbon footprint. Read the full story.
—Caiwei Chen
The AI Hype Index: Data centers’ neighbors are pivoting to power blackouts
Separating AI reality from hyped-up fiction isn’t always easy. That’s why we’ve created the AI Hype Index—a simple, at-a-glance summary of everything you need to know about the state of the industry. Take a look at this month’s edition of the index here.
Roundtables: seeking climate solutions in turbulent times
Yesterday we held a subscriber-only conversation exploring how companies are pursuing climate solutions amid political shifts in the US.
Our climate reporters James Temple and Casey Crownhart sat down with our science editor Mary Beth Griggs to dig into the most promising climate technologies right now. Watch the session back here!
MIT Technology Review Narrated: Supershoes are reshaping distance running
“Supershoes” —which combine a lightweight, energy-returning foam with a carbon-fiber plate for stiffness—have been behind every broken world record in distances from 5,000 meters to the marathon since 2020.
To some, this is a sign of progress—for both the field as a whole and for athletes’ bodies. Still, some argue that they’ve changed the sport too quickly.
This is our latest story to be turned into a MIT Technology Review Narrated podcast, which we’re publishing each week on Spotify and Apple Podcasts. Just navigate to MIT Technology Review Narrated on either platform, and follow us to get all our new content as it’s released.
The must-reads
I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology.
1 Hurricane Melissa may be the Atlantic Ocean’s strongest on record
There’s little doubt in scientists’ minds that human-caused climate change is to blame. (New Scientist $)+ While Jamaica is largely without power, no deaths have been confirmed. (BBC)
+ The hurricane is currently sweeping across Cuba. (NYT $)
+ Here’s what we know about hurricanes and climate change. (MIT Technology Review)
2 Texas is suing Tylenol over the Trump administration’s autism claims
Even though the scientific evidence is unfounded. (NY Mag $)
+ The lawsuit claims the firm violated Texas law by claiming the drug was safe. (WP $)
3 Two US Senators want to ban AI companions for minors
They want AI companies to implement age-verification processes, too. (NBC News)
+ The looming crackdown on AI companionship. (MIT Technology Review)
3 Trump’s “golden dome” plan is seriously flawed
It’s unlikely to offer anything like the protection he claims it will. (WP $)
+ Why Trump’s “golden dome” missile defense idea is another ripped straight from the movies. (MIT Technology Review)
4 The Trump administration is backing new nuclear plants
To—surprise surprise—power the AI boom. (NYT $)
+ The grid is straining to support the excessive demands for power. (Reuters)+ Can nuclear power really fuel the rise of AI? (MIT Technology Review)
5 Uber’s next fleet of autonomous cars will contain Nvidia’s new chips
Which could eventually make it cheaper to hail a robotaxi. (Bloomberg $)
+ Nvidia is also working with a company called Lucid to bring autonomous cars to consumers. (Ars Technica)
6 Weight loss drugs are becoming more commonplace across the world
Semaglutide patents are due to expire in Brazil, China and India next year. (Economist $)+ We’re learning more about what weight-loss drugs do to the body. (MIT Technology Review)
7 More billionaires hail from America than any other nation
The majority of them have made their fortunes working in technology. (WSJ $)
+ China is closing in on America’s global science lead. (Bloomberg $)
8 Australian police are developing an AI tool to decode Gen Z slang
It’s in a bid to combat the rising networks of young men targeting vulnerable girls online. (The Guardian)
9 This robot housekeeper is controlled remotely by a human 
Nothing weird about that at all… (WSJ $)
+ The humans behind the robots. (MIT Technology Review)
10 Cameo is suing OpenAI
It’s unhappy about Sora’s new Cameo feature. (Reuters)
Quote of the day
“I don’t believe we’re in an AI bubble.”
—Jensen Haung, Nvidia’s CEO, conveniently dismisses the growing concerns around the AI hype train, Bloomberg reports.
One more thing
How to befriend a crow
Crows have become minor TikTok celebrities thanks to CrowTok, a small but extremely active niche on the social video app that has exploded in popularity over the past two years. CrowTok isn’t just about birds, though. It also often explores the relationships that corvids—a family of birds including crows, magpies, and ravens—develop with human beings.
They’re not the only intelligent birds around, but in general, corvids are smart in a way that resonates deeply with humans. But how easy is it to befriend them? And what can it teach us about attention, and patience, in a world that often seems to have little of either? Read the full story.
—Abby Ohlheiser
We can still have nice things
A place for comfort, fun and distraction to brighten up your day. (Got any ideas? Drop me a line or skeet ’em at me.)
+ Congratulations to Flava Flav, who’s been appointed Team USA’s official hype man for the 2026 Winter Olympics!
+ Why are Spirographs so hypnotic? Answers on a postcard.
+ I love this story—and beautiful photos—celebrating 50 years of the World Gay Rodeo.
+ Axolotls really are remarkable little creatures.
Separating AI reality from hyped-up fiction isn’t always easy. That’s why we’ve created the AI Hype Index—a simple, at-a-glance summary of everything you need to know about the state of the industry.
Just about all businesses these days seem to be pivoting to AI, even when they don’t seem to know exactly why they’re investing in it—or even what it really does. “Optimization,” “scaling,” and “maximizing efficiency” are convenient buzzwords bandied about to describe what AI can achieve in theory, but for most of AI companies’ eager customers, the hundreds of billions of dollars they’re pumping into the industry aren’t adding up. And maybe they never will.
This month’s news doesn’t exactly cast the technology in a glowing light either. A bunch of NGOs and aid agencies are using AI models to generate images of fake suffering people to guilt their Instagram followers. AI translators are pumping out low-quality Wikipedia pages in the languages most vulnerable to going extinct. And thanks to the construction of new AI data centers, lots of neighborhoods living in their shadows are getting forced into their own sort of pivots—fighting back against the power blackouts and water shortages the data centers cause. How’s that for optimization?
An AI model released by the Chinese AI company DeepSeek uses new techniques that could significantly improve AI’s ability to “remember.”
Released last week, the optical character recognition (OCR) model works by extracting text from an image and turning it into machine-readable words. This is the same technology that powers scanner apps, translation of text in photos, and many accessibility tools.
OCR is already a mature field with numerous high-performing systems, and according to the paper and some early reviews, DeepSeek’s new model performs on par with top models on key benchmarks.
But researchers say the model’s main innovation lies in how it processes information—specifically, how it stores and retrieves memories. Improving how AI models “remember” information could reduce the computing power they need to run, thus mitigating AI’s large (and growing) carbon footprint.
Currently, most large language models break text down into thousands of tiny units called tokens. This turns the text into representations that models can understand. However, these tokens quickly become expensive to store and compute with as conversations with end users grow longer. When a user chats with an AI for lengthy periods, this challenge can cause the AI to forget things it’s been told and get information muddled, a problem some call “context rot.”
The new methods developed by DeepSeek (and published in its latest paper) could help to overcome this issue. Instead of storing words as tokens, its system packs written information into image form, almost as if it’s taking a picture of pages from a book. This allows the model to retain nearly the same information while using far fewer tokens, the researchers found.
Essentially, the OCR model is a test bed for these new methods that permit more information to be packed into AI models more efficiently.
Besides using visual tokens instead of just text tokens, the model is built on a type of tiered compression that is not unlike how human memories fade: Older or less critical content is stored in a slightly more blurry form in order to save space. Despite that, the paper’s authors argue, this compressed content can still remain accessible in the background while maintaining a high level of system efficiency.
Text tokens have long been the default building block in AI systems. Using visual tokens instead is unconventional, and as a result, DeepSeek’s model is quickly capturing researchers’ attention. Andrej Karpathy, the former Tesla AI chief and a founding member of OpenAI, praised the paper on X, saying that images may ultimately be better than text as inputs for LLMs. Text tokens might be “wasteful and just terrible at the input,” he wrote.
Manling Li, an assistant professor of computer science at Northwestern University, says the paper offers a new framework for addressing the existing challenges in AI memory. “While the idea of using image-based tokens for context storage isn’t entirely new, this is the first study I’ve seen that takes it this far and shows it might actually work,” Li says.
The method could open up new possibilities in AI research and applications, especially in creating more useful AI agents, says Zihan Wang, a PhD candidate at Northwestern University. He believes that since conversations with AI are continuous, this approach could help models remember more and assist users more effectively.
The technique can also be used to produce more training data for AI models. Model developers are currently grappling with a severe shortage of quality text to train systems on. But the DeepSeek paper says that the company’s OCR system can generate over 200,000 pages of training data a day on a single GPU.
The model and paper, however, are only an early exploration of using image tokens rather than text tokens for AI memorization. Li says she hopes to see visual tokens applied not just to memory storage but also to reasoning. Future work, she says, should explore how to make AI’s memory fade in a more dynamic way, akin to how we can recall a life-changing moment from years ago but forget what we ate for lunch last week. Currently, even with DeepSeek’s methods, AI tends to forget and remember in a very linear way—recalling whatever was most recent, but not necessarily what was most important, she says.
Despite its attempts to keep a low profile, DeepSeek, based in Hangzhou, China, has built a reputation for pushing the frontier in AI research. The company shocked the industry at the start of this year with the release of DeepSeek-R1, an open-source reasoning model that rivaled leading Western systems in performance despite using far fewer computing resources.
Another growth milestone for Meta’s Twitter replacement.
If it looks like a real human is being mutilated, even in video game form, it’s probably going to get restricted
Meta’s bringing in more money, as it continues to pour more into AI.