Uncomfortable Truth from the work: What got your reluctant students doing math this semester? Vote below, then tell us about one student who started trying and what changed. Can’t pick? Tell us what it depends on.

Experts largely agree this month’s AI proof of the quasi-Riemann hypothesis is correct, yet Princeton’s Peter Sarnak won’t be satisfied it’s “a Theorem” until humans check and understand all the details. The research case for AI in education hasn’t met that bar: 39 meta-analyses report substantial benefits, and an audit of 14 found serious flaws. Deciding what counts as proof is the creative judgment at the heart of mathematics, and it belongs at the classroom door. So what proof do you need? THE FIVE 1. [News] Quanta: AI Proved the Quasi-Riemann Hypothesis. What Happens Now? Stanford’s Jared Duker Lichtman calls OpenAI’s new proof “the most significant AI-for-math achievement so far,” and Sarnak is repurposing a course this semester to study it. Even at the top of the field, a proof becomes knowledge when someone teaches it, the strongest case yet for making students explain. 2. [Research] The Hechinger Report: 39 meta-analyses, 39 positive findings: An audit finds serious flaws in research on AI in education Ivy Tech’s Patrick O’Neill audited 14 of them, and in his preliminary reanalysis, 44 well-designed experiments swing from very helpful to very harmful and average out to roughly zero. MIT’s Justin Reich calls the meta-analyses “not trustworthy evidence,” so ask which studies back any AI math tool, and who checked them. 3. [News] Chalkbeat: NYC’s AI ban confusion: what’s allowed and what’s banned in city schools Magma Math made the city’s list of 38 flagged products for AI-generated hints, then switched them off and was cleared; HMH and several others say they were flagged for AI they don’t use. The city hasn’t explained how it decided, and a ban is only as good as the checking behind it. 4. [Policy] Education Week: States Call for More Time on Math. Schools Struggle to Shift Their Schedules Alabama, Georgia, Indiana, and Maryland have mandated 60 minutes of math a day (or 300 a week) in some grades, and Maryland pushed its start to 2028-29 after warnings that middle school arts and electives would be squeezed. Maryland’s math director names the stakes for anyone planning that hour: “The minutes alone aren’t it.” 5. [News] CNN: COVID launched a new era of technology-led learning. Now, some parents and school districts are ready to toss devices Los Angeles pulled devices from its youngest grades, and in a May Gallup poll, 61% of teachers wanted less student device time, even as 59% said digital tools help students learn. The device era never settled its evidence question, and AI is walking into the same gap.

AI labs now claim math results by the hundreds, but on WBUR’s On Point, Steven Strogatz named where the machines still fall short: they are “subhuman at asking good questions.” Posing the question is the creative act in mathematics, and in a math classroom it’s daily work: the problem set. This week it’s pulled two ways: Teachers Pay Teachers plans to ban AI-generated resources, while most teachers who use AI tell Gallup it frees time for what matters most. So who should write the problems your students solve? THE FIVE 1. [Radio · 40 min] WBUR On Point: AI is steamrolling through math. Why does that matter? OpenAI claims solutions to 16 questions tied directly to Fields medalist Hugo Duminil-Copin’s research, and Harvey Mudd’s Francis Su answers: “Math is understanding.” If Su is right, a class built around the answer key is teaching the part the machines just took. 2. [Essay] Terence Tao’s blog: Will it be a golden age or a dark age for mathematics (and science)? UT Austin math chair Francesco Maggi reports a prevailing view among mathematicians he’s spoken with: a strong result, by itself, “can no longer be taken as convincing evidence” of a student’s independent ability. He doesn’t know if that’s overblown, but if it isn’t, certifying understanding is now every math department’s problem. 3. [News] Chalkbeat: Teachers Pay Teachers to ban AI-generated classroom resources as changes to platform spark backlash. After Chalkbeat found apparent AI slop, including an alphabet coloring worksheet missing the letter F, TPT says seller material “should be unique, expertly crafted, and AI-free.” If human-made becomes the mark of quality, teachers who draft with AI will have to show where their judgment went in. 4. [Report] Gallup: Teachers Using AI Say It Frees Up Time for What Matters Most. In a Walton Family Foundation–Gallup survey of 1,952 educators, 59% used AI last school year, including to make materials and assessments, and nearly two-thirds of users say it freed time for what matters most. Overall, 85% say helping a student through something difficult makes them feel they matter a great deal, which is the trade-off on offer: AI drafts the worksheet, you take the struggle. 5. [Podcast] APM Reports, via The Hechinger Report: Sold a story: It’s not just reading. Emily Hanford’s second season moves beyond reading to teaching itself, starting with Zig Engelmann, the outsider who taught his 4-year-old twins basic algebra and developed Direct Instruction. As the scripted-lesson fight heats up again, math teachers should ask whether a script leaves room for a student’s own idea.

CREATIVE FORCE OpenAI announced 377 new math results on Oct. 6, including a claimed proof of the unique games conjecture, and a researcher who raced it put the week in one line: "There is a lot of value in failing and knowing why you failed." Our students have a shortcut too, a study mode that keeps offering them the answer. Mathematics is a creative act that lives in the stuck part, in posing a question, failing at it, and judging why, and getting answers doesn't allow for the socialization and learning potential with getting stuck and making mistakes. So when a student is stuck, do you rescue or wait? THE FIVE 1. [News] Quanta: As AI Closed In on 'Unique Games' Proof, Researchers Raced to Beat the Machines. Three days after hearing OpenAI might have proved the unique games conjecture, MIT's Dor Minzer and two grad students posted 95 pages on a related problem: "You don't know if you're going to get scooped by the trillion-dollar company." He fears AI strips the learning out of failure, the point of every hard problem we assign. 2. [Essay] Quanta: Is AI the End of Math As We Know It? Math editor Jordana Cepelewicz finds a field in grief, where many mathematicians have stopped posting open conjectures for fear bots will scrape them. McGill doctoral student Marcel Goh states the stakes for anyone who grades: "It was never about the theorems." 3. [Report] Common Sense Media: ChatGPT for Teens Risk Assessment. In more than 4,000 test prompts, the group's AI safety institute (partly funded by the OpenAI Foundation) found OpenAI's teen study mode bypassable in every configuration, complete with a "Show me the answer" option. OpenAI disputes the methodology, but if the study tool finishes the worksheet, take-home math needs a redesign. 4. [Opinion] Education Week: We're Hiding the Beauty of Math From Children. Teachers College's Rochy Flint offers three entry points: why bees build hexagons, three paper heptagons that won't lie flat, and two joined Möbius bands that, cut down the middle, become interlocked hearts. Her rule: "Put something interesting in front of everyone," not just the students who finish early. 5. [News] The Texas Tribune: STAAR rescoring requests triple after more Texas schools question computer grading. Since Texas began machine-scoring STAAR's open-ended answers, reading rescore requests have jumped from 2,758 in 2024 to 64,489 this spring, though fewer than 0.5% of responses changed. A UT Austin professor calls it "a bit of a trust problem," one any math program auto-grading student explanations should expect.

If you are interested, this Teaching Fellows Program is looking for new applicants: Knowles Teaching Fellows It is an amazing program!