This week, I got a sneak peek at ChatGPT’s new Study Mode during an OpenAI press demo on Zoom, ahead of its Tuesday release.
Toggling the tool button in the “Ask Anything” chat to “Study and Learn” transforms ChatGPT from a straightforward Q&A assistant into a personalized tutor. Instead of simply spitting out answers, it uses Socratic questioning, hints, and step-by-step guidance to encourage active learning. Tell it what you want to study, and it will assess your skill level and even remember your progress across chats.
As I watched the demo, my eyes widened and my breath quickened. Study Mode may have been designed with college students in mind—powered by system instructions crafted with input from teachers, scientists, and learning experts. But while it’s been several decades since my own academic experience, I instantly knew that I wanted to use Study Mode. And I knew exactly what I wanted to use it for: high school algebra.
My math failure shaped my life
Along with the rest of the math curriculum in those years, I completely bombed in algebra—and I can’t overstate how much that failure shaped my life. I carried a deep sense of shame about how hard math was for me, even with tutors. It stood in stark contrast to how easily I excelled in subjects like English and social studies, and it felt like no one really understood how curious I actually was about math. There were so many questions that were never answered. What is algebra, anyway? Why do we learn it in the first place? Why does it come before geometry and trigonometry? What’s the point of all these rules about x and y?
It might seem ironic, given that I now write about AI and regularly interview researchers with PhDs in mathematics who casually reference linear algebra in conversation. But I’ve always wished I could get a do-over on high school math.
Yes, there are real concerns about how AI will affect student learning—from cheating and over-reliance to privacy and bias. New reporting this week from Rest of World found that when students in rural Colombia gained access to AI, students started failing exams.
But for me, ChatGPT’s new Study Mode is already becoming something deeply personal: a long-overdue redemption arc.
Study Mode at my own pace
I started where I never really could before, and Study Mode immediately had a refreshing response:
“Let’s treat this like the first day of Algebra 1—only better. This time, you get to ask questions whenever you want, and we move at your pace.”
I was on board with that: What is algebra, I asked.
Algebra is the part of math where we stop just working with numbers and start working with relationships between numbers.
Instead of asking:
What’s 3 × 5?
We ask:
What number x would make this equation true?
It’s like math with mystery—something is hidden, and our job is to solve for it.
“Math with mystery”—ooh, I loved that. After giving me the lowdown on the “big ideas” in algebra–everything from solving equations to graphing lines and word problems, Study Mode offered a few options:
Want to:
A) Ask questions about any of this?
B) Choose a puzzle type to start with?
C) Keep going with the intro and build slowly from here?
Your call!
I said I’d like to solve equations. Study Mode dove right in:
🔍 Solving Equations: What’s the Goal?
An equation is like a balance scale—whatever you do to one side, you have to do to the other.
Here’s your first equation:
3x + 4 = 19
👉 What do you think we should do first to start solving for x?
(Just take a guess—we’ll figure it out together.)
The promise of AI-powered personalized learning
I sat back and smiled. Apparently, it’s never too late to solve for x—even for a journalist who always thought she was good with words but terrible with numbers. I spent a couple of hours progressing through one-step and two-step equations; equations with variables on both sides; equations with parentheses; and equations combining like terms. I asked lots of questions along the way and occasionally peppered my responses with “Wow!” and “I never knew that!”
Like every AI use case, there are tradeoffs. The risks of over-reliance on AI and diminished critical thinking are real. But I find myself leaning toward the view of my colleague Jeremy Kahn, who, in his recent book Mastering AI: A Survival Guide to Our Superpowered Future, highlights the promise of AI-powered personalized learning. He frames the idea of AI as a one-on-one tutor as one of its most powerful educational opportunities.
My late-in-life return to algebra may not have been OpenAI’s target use case. But who knows—maybe I’ll make it to calculus.
With that, here’s the rest of the AI news.
Sharon Goldman
sharon.goldman@fortune.com
@sharongoldman
AI IN THE NEWS
Microsoft signs on to EU’s AI Code of Practice, but Meta has declined. Microsoft announced Thursday that it has signed on to the European Union’s General-Purpose AI Code of Practice—becoming one of the first major tech companies to formally do so. The move signals Microsoft’s support for the EU’s AI governance framework, even as it calls for simplification of what it describes as a complex regulation. Other companies have also expressed willingness to align with the voluntary code, including OpenAI and Mistral, but notable holdouts remain: Meta has declined to join, and Google has yet to make its position public.
OpenAI launches Stargate Norway, its first AI data center initiative in Europe. OpenAI launched Stargate Norway as part of its broader Stargate program under the OpenAI for Countries initiative launched in May to partner with governments and help them build out their own AI infrastructure, particularly focusing on data centers. The facility—planned for Narvik and backed by Norwegian partners Nscale and Aker—will deliver up to 230MW of AI compute capacity, with plans to scale to 100,000 NVIDIA GPUs by 2026. The project underscores OpenAI’s strategy to partner with governments and industry leaders around the world to build sovereign, sustainable AI infrastructure. It follows Stargate UAE, and is part of a growing global footprint that also includes agreements with the UK, Estonia, and early engagement with the EU’s AI Gigafactories initiative—each aimed at ensuring countries have the compute capacity and ecosystem support to harness AI for national priorities.
AI researchers are approaching the job market like NBA stars. The New York Times has a great story today about the AI talent wars, in which the race to recruit top young AI researchers has become as intense—and lucrative—as signing NBA superstars, with companies like Meta, OpenAI, Google, and Microsoft offering nine-figure compensation packages and engaging in highly publicized hiring battles. Many of these 20-something “AI free agents” are turning to informal agents and entourages to navigate the frenzy and negotiate top deals, unbound by salary caps like those in professional sports. The competition has even taken on the tone of a sports spectacle, with streaming outlets like TBPN covering notable industry job changes with the flair of a league's trade deadline.
FORTUNE ON AI
Salesforce CEO Marc Benioff on why AI agents won’t lead to mass unemployment—by Jeremy Kahn
Mark Zuckerberg is pouring billions of dollars into AI ‘superintelligence’—so why does his Instagram pitch feel so underwhelming?—by Sharon Goldman
Meta’s Mark Zuckerberg laid out his AI vision that outperformed Q2 expectations and sent shares soaring—by Amanda Gerut
Why Booz Allen’s CTO used generative AI to make a deepfake video of himself—by John Kell
COMMENTARY: Silicon Valley’s billions of dollars on AI haven’t actually generated a return yet. Here’s why most companies should embrace ‘small AI’ instead—by Jason Corso
AI CALENDAR
Sept. 8-10: Fortune Brainstorm Tech, Park City, Utah. Apply to attend here.
Oct. 6-10: World AI Week, Amsterdam
Oct. 21-22: TedAI San Francisco. Apply to attend here.
Dec. 2-7: NeurIPS, San Diego
Dec. 8-9: Fortune Brainstorm AI San Francisco. Apply to attend here.
EYE ON AI NUMBERS
52%
That’s how many developers are not yet using AI agents, according to a new survey of software developers from Stack Overflow, the popular online question-and-answer platform for computer programmers and developers. The study found that AI agents are not yet mainstream: A majority of developers (52%) either don't use agents or stick to simpler AI tools, and a significant portion (38%) have no plans to adopt them.
