An anonymous reader quotes a report from The Economist: Students and school children are increasingly using artificial intelligence. A survey last year by Chegg, an ed-tech firm, found that 80% of rich-world undergraduates used it in their studies. More recent polls put the figure at 94% in Britain and 93% in Germany. Such widespread adoption has fueled concerns about the impact the technology might be having on learning. Teachers report marking identikit essays that they suspect are churned out by ChatGPT. Yet firm evidence on AI's educational effects has been lacking.
David Stromberg of Stockholm University and Victor Lei and Wu Yanhui of the University of Hong Kong set out to fill the gap. They tracked 27,000 pupils aged 12-18 in China, where AI adoption has been fast. Around 80% reported using models such as Doubao and DeepSeek; the other 20% formed the control group.
The results are eye-opening. After six months, pupils using AI saw their average homework score rise by 18% across all subjects. The time they took to complete each assignment fell from an average of 64 minutes to 45. But come exam time, the same students scored 20% below their classmates who had not called on AI's help. Homework scores once predicted exam performance; now those who score highest are, perversely, more likely to do worse in exams. The takeaway: "AI can boost learning productivity, but only for those who use the technology intelligently," reports The Economist. "Students must resist the temptation to reach for an AI-generated answer before thinking things through for themselves."
On the first day of the semester, Professor Sisyphus made a to-do list. As he completed items on the to-do list, he kept adding new ones at about twice the rate he crossed off items.
Professor Sisyphus graded the papers his students had submitted. The day he handed back the graded papers, the next paper was due.
Professor Sisyphus joined the committee that was drafting the new strategic plan. They held planning meetings. Then they held listening sessions. Then they drafted a draft strategic plan. Then they held feedback meetings on the draft strategic plan. Then they made revisions to the draft based on the draft strategic plan feedback meetings. On the day the administration released the new strategic plan, which didn’t reflect the committee’s work, they called for volunteers to form a committee to start drafting the next new strategic plan.
Professor Sisyphus was tasked with updating the website with news stories about what was happening in the department. New things kept happening.
Professor Sisyphus heard you’re not supposed to throw batteries in the garbage, so he put dead batteries into a drawer until he figured out what to do with them. New batteries were also stored in that drawer and often fell out of their packages. When Professor Sisyphus went to the drawer to discard dead batteries and retrieve new ones, he had no idea which batteries were new and which were dead. He threw away all the batteries. He bought new batteries. Professor Sisyphus spent countless sleepless hours worrying that his discarded batteries would leach chemicals into the river and contaminate all the fish.
Professor Sisyphus finally figured out how to plagiarism-proof his writing assignments. Immediately, computers learned how to instantly write B+ student papers.
Professor Sisyphus was asked to complete an assessment report for his department to prove that students were learning something at college. The first question on the assessment report asked what new changes the department had implemented from the previous year’s assessment. The final question asked for a prediction of what the department would report on next year’s assessment form. The administration said these questions about last year’s assessment and next year’s assessment would help the department “close the loop” on assessment. The idea of closing the loop made Professor Sisyphus feel good; he had been longing for some closure. But then he remembered loops are circles: closing them meant getting stuck in an infinite hamster wheel of assessment.
Professor Sisyphus trained a student assistant to put departmental news stories on the website so he wouldn’t have to do it anymore. The student graduated.
Professor Sisyphus learned a slang word. He was excited to impress his students. Then he discovered that as soon as he uttered slang in class—poof! It would instantly cease to be a part of youth culture.
Professor Sisyphus recognized that his body type put him at high risk for cardiovascular disease. He signed up for an exercise class where participants rolled boulders up hills. But each day he worked out, he felt extra hungry.
So Professor Sisyphus decided to incorporate more healthy fish into his diet. He loaded batteries from the battery drawer into his seafood toxicity test kit. However, because he still didn’t know which batteries were old or new, the seafood toxicity kit did not work. Also, because he had thrown batteries in the garbage, all fish were now toxic.
Professor Sisyphus studied hard, day in and day out. He made flash cards with pictures on one side and words on the other. He gave himself little quizzes to practice. Finally, after much labor and many mistakes made, he could remember how each and every one of his students spelled their name: Madi, Maddy, or Maddie. But by then it was Finals Week, and next semester’s course had already been enrolled by a fresh army of Madisons.
Professor Sisyphus wanted to commiserate about the Madis with his colleagues, but all his colleagues were named Andrea. He didn’t remember which were ANNE-dree-uh, which were Ahn-DRAY-uh, and which were AHN-dree-uh. He was pretty sure the one in the philosophy department was Ahn-DRAY-uh, but when he went to her office to joke about how everyone was named Maddy, he discovered the philosophy department no longer existed. In its place was a fitness center annex where a few Maddies jogged on treadmills.
The Madisons were running so fast. But they weren’t going anywhere.
Is it a good idea to use artificial intelligence (AI) for math help? Yes — and no. That’s based on the results of a new survey. AI can be a great tutor, it finds, but only if you use it in certain ways.
The data come from a poll of 1,300 high school students from across the United States, England and Wales. The Society for Industrial and Applied Mathematics surveyed some top math students in each nation. The goal was to identify the habits behind their academic success.
AI has many strengths as a math tutor, the study found. Students can use AI to explain things in different ways and to check their homework answers. It can answer “why” questions about material to deepen their understanding. AI can tailor learning by identifying a student’s knowledge gaps. Using AI as a tutor could even reduce math anxiety, some students said.
But there are potential downsides to AI tutoring as well, notes Robbie Torney. AI was designed to reduce friction — meaning it aims to do things quickly and efficiently. But learning “requires friction,” says Torney, who was not involved in the survey. He heads up AI & Digital Assessments at Common Sense Media. It’s a nonprofit group working to make the digital world safer for children and teens.
Science shows that learning requires grappling with problems, Torney notes. You have to work through things that feel hard and come up with your own answers.
Tiago C.P. agrees. To this Colorado high-school senior (who did not take part in the survey), the most important thing is “trying the problem first.” This is where students struggle, he adds, but “that’s when you get the learning down.” Says Tiago, “I won’t use AI before I’ve fully attempted a problem.”
Using AI for help with homework or when studying for tests can be fine — as long as you keep thinking, doing the work and using a range of other resources, too.Anton Vierietin/iStock/Getty Images Plus
Survey says
The teens in the study had all taken part in an international math contest. Afterward, they gave their take on whether AI could be an effective math tutor. About half said yes — if the AI was used in specific ways. More than 200 math teachers also took part. More than six in every 10 agreed that an AI tutor could be beneficial. But AI could also decrease learning if used improperly, they noted.
Almost two thirds of students surveyed (63 percent) said they would always try a math problem on their own before seeking help from AI. If they were struggling, where would the students turn? Most recommended first asking teachers (58 percent), followed by classmates (50 percent). Others said they’d continue to try on their own (48 percent) or look for tutorials on the internet (46 percent). At 39 percent, AI came in fifth place.
That suggests using AI after trying most other sources of help. The number one suggestion for how to do well in math was doing regular practice problems. That’s according to both students (67 percent) and teachers (62 percent).
Overall, about half of the students said they believe AI has a place in math education. But they added that it is best used along with — not in place of — help from their classroom teacher. For one thing, “It’s hard to get AI to teach you the same way your teacher would,” says Amelie M., another Colorado senior (who did not participate in the survey).
Ask questions of your human teachers, not just of your AI tutor. A range of teaching styles can boost understanding. Also, people can understand each other better than AI can. Klaus Vedfelt/DigitalVision/Getty Images Plus
How to use AI for homework wisely
More than half of the teachers in the study (55 percent) said AI’s biggest benefit is its ability to explain — step by step — how to solve a problem.
But relying on it can harm learning in the long run. Almost three in every four teachers surveyed said their biggest concern was that AI can quickly become a crutch.
Amanda Brown teaches math at Broomfield High School in Colorado. When learning something new, you can use AI for help, she says. But as you become more familiar with the subject, studies show, you should turn to AI less and less.
Torney recommends asking yourself (or even the AI): “Who’s doing most of the work?” Whenever it’s AI, he says, that likely points to a problem.
“I see kids that have perfect homework scores,” Brown says, who later “don’t pass their test.” AI might have helped them ace the homework. But in the end, Brown notes, “You’re gonna have to take the test [without AI].” And that’s when it comes out: Do you know the material or not?
The trick is using AI in moderation, agrees Amelie. Try to ask it very specific questions, she says, such as: “What are different ways I could solve this?” Amelie also suggests adding to your prompt: “Don’t give me the answer.”
Tiago warns that you also need to check whether AI is correct. Although chatbots keep improving, they sometimes “hallucinate” (make errors). This is because AI doesn’t actually know the answer. It just predicts what is the most likely answer from the data it has ingested. “At least a quarter of the time,” Tiago says, “I can catch a mistake it’s made.”
Remember that you can also work on problems with classmates. And teachers are likely your best source for homework help. “Learning has always been a social task,” Torney points out — “something that we do with other people.”
Studying in groups can turn learning into a “social task.” Those who understand something better can coach peers. People have different ways of learning and can benefit from others’ approaches. Jacob Wackerhausen/iStock/Getty Images Plus
Beyond homework
AI can be a good helper as you prepare for tests, Tiago finds. It can generate sample problems, create flash cards or drill you on concepts. Or if you were out sick for a few days, AI might help you catch up on topics you missed.
But AI will “pull from so many things,” says Amelie, that it can return “long, long paragraphs” of info. This “might actually be more overwhelming” than helpful, she points out. What might prove more beneficial are YouTube instructional videos that AI references, she says. Such videos show someone actually teaching you.
A math hack: You can also ask AI to tell you about your own thinking and learning. The surveyed teachers suggested letting AI test you to find gaps in your knowledge. Students recommended asking AI “why” questions about math. These could make you a better learner.
Math is a bit like learning a sport. You have to learn certain skills and in a certain order. Consider ice hockey. Until you learn how to skate — and how to stop — you’ll have a tough time mastering a slap shot.
In math, if you don’t learn factoring, you’ll have trouble in algebra. If you don’t learn algebra well, calculus will be a bear. This is why Torney advises that you “keep your eye on the long game.”
And learning takes work. “When it’s easy,” Torney says, “learning is probably not happening.” If you’re using AI for help with math and don’t feel like you’re working very hard, change how or where you get help.
Torney offers a last tip: Choose your AI platform wisely. Chatbots such as ChatGPT, Gemini, Claude and Copilot are “multi-use” systems. They can tackle any topic — not just math. Platforms such as Mathos and Wolfram|Alpha answer questions just on math. And the discontinued (but still available) Khanmigo focuses on education.
Some students who use multi-use systems for math can start relying on them for social advice or emotional support, too. This can lead to dependency in some people. That’s why systems focused on a single topic, Torney says, “tend to be less risky for kids and teens.”
Using AI to help you understand math better can be a great thing. Just keep in mind this mantra from the teacher survey: “It’s critical for students to arrive at the correct answers themselves.” Use AI to learn concepts, not get answers.
For the first time since 2021, fresh data from Google gives us a look at how we’re using emoji and there’s a major shakeup at the very top: 😂 and ❤️ are no longer our most used emoji. I know what you’re thinking. “Jennifer, is it because the world is collapsing? Do we live joyless lives now? Is love … dead??” No, sweet child. Tears of Joy (😂) didn’t drop out of first place because people stopped laughing. We just crave more drama. We collapse from laughter. We weep from laughter. We scream from laughter.
Number 1 Emoji Over Time: 2021 😂→2024 🤣→2025 😭
Loudly Crying Face (😭) took the top spot because it became the default reaction for when something is so funny, embarrassing, or cute that you can't cope. It's three in one. It’s the KenTacoHut of emoji. When people do want a traditional lol, they lean into the physical chaos, keeping rofl 🤣 at #2.
Y R U SO MAD 🤬
For years, we used smiley faces to make it very clear that I am friendly and not threatening :))))) Sometimes plain text lacks nuance and I really wanna make sure you like me!!!! That politeness still exists but unvarnished frustration has officially broken through. Middle Finger (🖕)jumped from #113 in 2021 to #17 in 2025 (!!) Enraged Face (😡) broke into the top 15 (!!!!!!) and Face with Symbols on Mouth (🤬) was ranked #120 and is now #26 (???!!!!!!)
Jumping nearly 100 spots into the global top twenty isn’t a gentle trend line; Once you crack the top twenty, the sheer volume of taps is astronomical. It’s a dead giveaway that our digital manners have changed: in addition to cushioning our messages we also throw heat. It's the same hyperbolic shift that pushed 😭 to #1.
Highs and Lows 👑
When you scroll through your emoji keyboard looking for the perfect expression, you’ll find nearly 4,000 tiny drawings: an amphora 🏺, a card index 📇, a roller coaster 🎢, and a dozen trains 🚂 🚃 🚄 🚅 🚆 🚇 🚈 🚊 🚋 🚞 🚝 🚟 🛤️. And yet … out of our massive emoji catalog just ten (that’s right, one-zero 10) account for over one third of all global emoji shares: 😭🤣😂❤️🙏😘🥰👍😍😁. How can this be?
Contrary to popular belief, aerial tramway (🚡) isn’t the least-used emoji. That title belongs to Japanese “here” button (🈁). But like aerial tramway, without metaphorical flexibility, thousands of literal objects sit untouched. Over 55% of the entire library gets near-zero use 😲
(Note: This data was collected in 2024/2025 and only includes emoji available for at least a year, giving the new emoji enough time to reach everyone's devices.)
This data sheds light on a fascinating tension: our keyboards present emoji as a visible encyclopedia, but in practice we use them for the invisible stuff like sarcasm, tone, and emotional shorthand. Faces and gestures across the board account for over 80% of all global volume. Staples like ❤️ and 🙏 remain frequently used because they’re the digital version of warm gratitude or eye contact. Basic human stuff.
The Reality of Encoding a Living Language
Once inducted into the Unicode Standard, an emoji is never removed. It’s there forever. Forever ever. But, as the data shows, people naturally gravitate around a flexible core, leaving a wide long tail of highly specific concepts that collect dust 😬 That leaves digital standards with a real dilemma: how to support rich expression without running up against keyboard limits, search discoverability, accessibility, and font architecture. If we don’t continue to explore and understand how people actually talk, we’ll just end up creating thousands of new emoji that nobody uses.
What I love about this data is the reminder that language thrives when we learn the rules then break the rules to make them our own. Lately, I’ve taken to using the Fountain (⛲️) whenever I’m overflowing with energy or feeling bonita. So go ahead, give that emoji picker a twirl, dig into the corners you rarely visit, and find a lonely emoji and make it yours ⛲⛲⛲
Methodology
Emoji use data was collected with Google’s Federated Analytics, capturing anonymized, aggregated usage in Gboard. The collection periods spanned November 2024 to January 2025 and November 2025 to January 2026. Special thanks to Alex, Wilder, and the Gboard team for making this analysis possible. 🗿📊
[2:50 AM on a Sunday is also a particularly scammy time for this to be sent.]
This past Sunday, I received an email from “Do Not Reply” with the subject line “Vehicle Registration Renewal Reminder”. Here’s what it looked like:
This is a very, very bad email, beginning with the awful A.I. slop image at the top. Here’s a larger look at just that:
Lest you have any doubts whatsoever as to the origin of this image, note that they didn’t even crop out the Gemini logo from the bottom right. While I’ve compressed the image above, the original of this needless image was also a needlessly large 760 KB png file. Awful.
When I saw this email, I was quite certain it was a phishing scam. I assumed that the link inside would take me to a scammy replica of a government website, one that would steal credit card numbers. But no! As far as I can determine, this is a legitimate email from the RMV.1 The lone link in it does indeed go to http://mass.gov/rmv, which forwards on to the official RMV site.2 And indeed, my car registration is expiring in a couple of months.
In the past, I have always received an official form in the mail which can be used to renew my registration. For many years, that letter has simply prompted me to go online to renew. Given that, a switch to email notifications make sense. They sure ought to look a lot more official than this, however.
Footnotes:
What’s most often known as the “Department of Motor Vehicles” (DMV) is indeed called the “Registry of Motor Vehicles” (RMV) in Massachusetts. Why? Just to be difficult, I assume. But we’re far from the only ones. ↩︎
I’m not sure why the original link is http:// rather than https://, but the forward does lead to a secure page. ↩︎
With the numbers 1 through 9, mathematician Ben Sparks introduces the 15-game, a two-player strategy game where the goal is to be the first to choose three numbers that add up to 15. From Sparks in this episode of Numberphile, Brady Haran’s long-running video project about numbers and mathematics: “If...
Log in
