December 2025 Newsletter

Welcome to the Math and AI 4 Girls December newsletter! This month, we're excited to share inspiring stories of women in STEM, fascinating breakthroughs in math and AI research, and important contest updates as we count down to the upcoming MA4G competition. Send this newsletter to a friend who would love to participate, and subscribe to stay in the loop!

Contributors: Fiona Liu, Lulu Wang-He, Evelyn Qiao, Advika Asthana, Angelina Wang, Chelsea Lu, Angie Huang, Sophia Jin

Table of Contents

1. About Us

2. Women in Math Story

3. Math and AI Research

4. Problem of the Month

5. Bonus: Holiday Problem

6. MA4G Competition Countdown

7. POTM Resources & Hints

About Us

Math and AI 4 Girls is a nonprofit organization dedicated to promoting young girls' interest in STEM. Each spring, we organize a competition designed to motivate students to engage with problem-solving through a challenging problem set and share their unique STEM stories through two thought-provoking essay prompts.

Any female middle school students (grades 5–8) with U.S. residency younger than 15 years old are eligible to enter! (Past grand prize winners, however, are not eligible to re-enter.) Winners are recognized at an online awards ceremony during the summer, and award recipients will receive prizes, such as up to 1,000 dollars, merchandise from sponsors, a personalized award certificate, and more! If you know anyone who might be interested, please encourage them to stay connected via our website and join our Discord, where we offer more math competition opportunities, host activities such as problem-of-the-week, and prepare for next year’s competition, opening in March 2026!

Women in Math Story

Every month, we share the story of an inspiring female mathematician who reflects the mission of MA4G. This month, we look to the journey of Maryam Mirzakhani, whose life, work, and legacy exemplifies passion, persistence, and progress in mathematical discovery.

Maryam Mirzakhani grew up in Tehran with a curiosity and talent for patterns and ideas. As a teen, Mirzakhani won back-to-back gold medals at the International Mathematical Olympiad. She went on to earn a PhD from Harvard, where she worked under Curtis T. McMullen, a Fields Medalist. She became a professor at Stanford University in 2009. Her groundbreaking work on the dynamics of Riemann surfaces earned her the Fields Medal in 2014, making her the first woman to ever receive this award. 

Mirzakhani approached mathematics as an art, doodling on sheets of paper and then writing equations around them. Her daughter described her approach as “painting.” Mirzakhani was a trailblazer in the fields of hyperbolic geometry, topology, and dynamics. Although her life was tragically cut short by cancer, her legacy continues through the Maryam Mirzakhani New Frontiers Prize and the 12 May Initiative, both dedicated to supporting women in mathematics. 

Mirzakhani’s story demonstrates the innovation, creativity, and ability young women have when they are presented with the opportunity to explore mathematics beyond the classroom. Following in her footsteps, MA4G encourages participants to approach mathematics as an exploratory process, empowering girls to pursue math with ambition, creativity, and passion. By building a community of girls who learn and grow together, MA4G creates a space where students can support one another and see STEM as a place where anyone can belong.

Math and AI Research

Some exciting news has been brewing this month in the fields of AI and mathematics! We’ll be breaking down some of the most interesting topics. Click on any link to learn more.

First, a quick vocabulary term that will be used quite often: Large Language Models (LLMs) are advanced AI systems trained on massive text datasets to perform a variety of tasks that involve processing and generating natural language. ChatGPT, Gemini, and Claude are common examples.

I. A mathematical ceiling for AI creativity

David H. Cropley, a professor of engineering innovation at the University of South Australia, wanted to resolve the heated debate over whether AI is truly “creative” or simply mimics what already exists. Cropley decided to find an objective measure for AI creativity.

Here’s how he did it:

1. He defined creativity as the product of effectiveness and originality.

2. He noticed that the architecture behind LLMs involves a hidden trade-off between these two criteria.

   • Classic LLMs strive to provide an effective response to satisfy the user, which involves selecting context-relevant words. For instance, it would finish the sentence “The sky is...” with words like “blue” or “the limit.”

   • This bias towards more effective words means that originality is lost because the LLM searches for only the most common responses.

3. Because of this inverse relationship, both cannot be maximized simultaneously.

4. Cropley found a mathematical expression for this relationship and attempted to maximize that instead.

Cropley calculated the theoretical ceiling for AI creativity to be around 0.25 on a scale from 0 to 1. Measuring this value against typical human creativity levels, he discovered that AI exhibits only an amateur level of creativity.

This means that AI performs creative tasks like copywriting, blogging, and brainstorming at the level of an average person. However, in its current state it is unlikely to be capable of replacing expert creatives.

What do you think? Is there a ceiling to AI creativity, or will it eventually improve to the level of professionals with years of experience? Can artificial intelligence ever truly be creative?

II. Benchmark for evaluating the mathematical capabilities of LLMs

Chinese researchers Tianqiao Liu, Zui Chen, Zhensheng Fang, Weiqi Luo, Mi Tian, and Zitao Liu published a recent paper about MathEval, a comprehensive benchmark for evaluating the mathematical capabilities of LLMs. They combined 22 datasets of math problems varying in topic, language, and difficulty to gain an accurate assessment of LLM capabilities and judged responses using GPT-4.

The results were extremely interesting:

• Claude-3.5-Sonnet performed the best, with an average accuracy rate of 77.0%

• Qwen2-72B-Instruct was the best-performing open-source model, with an average accuracy rate of 74.4%

• LLMs usually did well on arithmetic and simple math problems, while they struggled with more difficult high school or competition-style math problems.

• On word problems, LLMs typically performed better on problems written in English than Chinese.

• LLMs performed worse on exam problems that came out after the model’s release. This means they are potentially overfitting, or have learned the small details of their training data so well that they are unable to generalize to similar but slightly different problems.

• Closed-source, or private, models tend to outperform open-source, or free/low-cost and public, models.

MathEval is an exciting new way to evaluate AI models and identify areas for improvement, helping researchers develop the already exceptional mathematical capabilities of LLMs.

In the novel and fast-paced world of AI and math, hundreds of updates roll in every month. Reading our bite-sized reports can help you stay up to date with the current developments in this realm!

Problem of the Month

Welcome to our Problem of the Month! Each issue of the newsletter, we'll feature a math challenge for you to try your hand at. Winners get the chance to win merch and prizes! This month's problem is below:

Problem Credits: Eileen Wu, MA4G '26

Submit your solutions here!

If you’re feeling stuck, we have resources and hints at the bottom of this newsletter to help you out. We highly recommend trying your best before checking the hints, and working with them one at a time.

Bonus: Holiday Problem!

As we approach the holiday season, here's a fun, holiday-themed problem for you to try!

At the Winter Festival, four bell towers ring at regular time intervals. At 6:00 PM, all four ring at the same instant. After that:

• Tower A rings every 12 minutes

• Tower B rings every 15 minutes

• Tower C rings every 20 minutes

• Tower D rings every 28 minutes

During the next 12 hours (from 6:00 PM up to and including 6:00 AM), how many times do exactly two of the towers ring at the same instant? (Do not count the starting time 6:00 PM.)

Problem Credits: Ellie Su, MA4G '26

MA4G Competition Countdown

We are THREE months away from the MA4G competition! Mark your calendars for mid-March, when the competition will open, and you’ll have a chance to win prizes of up to $1000 as well as merch from our sponsors!

The MA4G team is working tirelessly behind the scenes as we kick off the 2025-26 season. Our problem set and communication teams are hard at work crafting fun problems and essays that we hope you’ll enjoy working on. Meanwhile, our marketing team is busy designing and handing out flyers. Look out, you might even see members at local competitions!

In the meantime, the problem set team is releasing weekly challenges in our Discord server, as well as a monthly challenge! Make sure to check them out. We hear you might win fun prizes!

Three months may seem like a long time, but March will be here before you know it. We are so excited for you to see the competition our team members have put together, and we truly believe this will be our best year yet! Keep an eye out for updates, stay involved, and help us spread the word!