Good morning, {{ first_name | AI enthusiasts }}. An AI just delivered a breakthrough in mathematics — solving a 30-year-old problem using no human help at all.
With DeepSeek and Google also reaching gold-level reasoning, we may finally be on the cusp of mathematical superintelligence, where solving hard problems could be a superpower shared far beyond pro-level mathematicians.
P.S. Our next edition of The Rundown Roundtable is here, where our staff members share the unique ways we’re incorporating AI into both our work and personal lives. See the latest use cases below, and submit your own workflow here.
In today’s AI rundown:
‘Aristotle’ AI cracks 30-year math problem
The Rundown Roundtable: Our AI use cases
Use AI to find patents and innovation opportunities
China overtakes the U.S. in open AI economy
4 new AI tools, community workflows, and more
LATEST DEVELOPMENTS
HARMONIC
Image source: Gemini / The Rundown
The Rundown: Aristotle, an AI system built by Harmonic, just independently solved a 30-year-old Erdős problem, marking what researchers are calling the first real step into the “vibe proving” era of mathematics.
The details:
Aristotle solved a version of Erdős Problem #124, which has been open since the 1990s, in six hours, and then formally verified the proof in Lean in a minute.
The result came from Aristotle’s beta version, updated with stronger reasoning and a natural language interface to explore and write step-by-step proofs.
Vilad Tenev, the founder of Harmonic, called this the arrival of “vibe proving” — AI-driven proofs discovery followed by machine-verifiable rigor.
The development follows Harmonic’s $120M funding and Aristotle’s IMO gold performance, putting it alongside Google and OpenAI in mathematical reasoning.
Why it matters: Harmonic’s breakthrough is another push toward mathematical superintelligence, where proofs will be generated, verified, and scaled at superhuman speeds. Tools like these can also open participation in advanced mathematics, turning it from something only experts do into something anyone can contribute to.
TOGETHER WITH TELUS
The Rundown: Hidden vulnerabilities and compliance challenges are emerging faster than ever. Uncharted is where industry leaders take action to secure the future of AI. Gain actionable strategies in nine expert-led sessions to protect your AI innovations before it’s too late.
In this insightful summit, you'll uncover:
Hidden dangers lurking in 24+ frontier models
Legal frameworks that balance innovation with responsibility
Proactive defense strategies from top CISOs
THE RUNDOWN ROUNDTABLE
Image source: Ideogram / The Rundown
The Rundown: The Rundown Roundtable is a new weekly feature in which we poll members of The Rundown staff about how we use AI in our work and daily lives.
Rowan, CEO: I’m getting ahead of my 2026 fitness goals by giving the new ChatGPT 5.1 Pro model context from my old peak-performance training plans, Whoop data, work schedule, dietary restrictions, and goals. Together, we’re building a program designed to get me back to peak shape — optimized exactly for how I train, work, and recover.
Jason, Head of Product: I ran a functional lab test and sent the results to ChatGPT. It flagged my homocysteine at 46 (safe range is under 10) and suggested it could be linked to an MTHFR variant. I ordered a follow-up genetic test — and it came back positive for the C677T mutation. I started Thorne MethylGuard Plus supplements, and six weeks later, my homocysteine was normal. It helped me connect dots I wouldn’t have seen on my own.
Mayur, Content Manager: My work shifts usually stretch past midnight, so I’ve been trying to take my health more seriously. I created a dedicated project on ChatGPT and a mirrored one on Grok just to cross-check things, where I’ve uploaded my entire routine and all the details from my full body checkup. I’ve started adding photos of every medicine, supplement, and treatment I’ve been taking, along with notes on why and when I use them.
The idea is to keep updating it so it kinda becomes a digital biodata of my health and a go-to place for all my specific health-related questions.
AI TRAINING
The Rundown: In this tutorial, you’ll learn how to use Perplexity’s AI-powered patent search to quickly identify existing patents, uncover open innovation spaces, and reduce the risk of infringement before investing in new ideas.
Step-by-step:
Go to Perplexity.ai and type your question (e.g., “Are there any patents related to AI automations?”). The platform automatically detects patent-related searches and shows relevant filings, owners, and grant dates
Refine your query with context, such as “Find active patents in AI-driven industrial automation.” Then ask follow-ups like “Show whitespace in this field” to reveal gaps and opportunities
Enable Agent Mode to activate multi-step reasoning. The agent compiles patents across regions, creates tables and charts, and visualizes your research space for deeper analysis
Review the generated CSV and PNG reports to identify crowded zones, emerging areas, and potential whitespace
Pro tip: Start broad, then narrow your search by asking the agent to group results by company, summarize claims, or map open innovation zones.
PRESENTED BY WARP
The Rundown: Warp just launched its biggest Agents update yet, propelling it to No. 1 on Terminal-Bench agentic coding benchmark ahead of Claude Code, Gemini, and Codex.
Warp’s Development Agent now includes
Full terminal use with long-running commands like servers and debuggers
Steerable planning you can review and edit in real time
Full lifecycle support, from planning and coding to deployment
See what the next evolution of development agents looks like.
AI MARKET
Image source: MIT
The Rundown: A new MIT and Hugging Face study analyzing 2.2B Hugging Face downloads reveals a “fundamental rebalancing” of the open AI economy, with U.S. industry dominance collapsing in favor of Chinese heavyweights.
The details:
The study found that Chinese AI developers have surpassed the U.S. industry in downloads, capturing 17.1% of the market compared to the U.S.’s 15.8%.
This surge is largely driven by two Chinese players, DeepSeek and Alibaba’s Qwen, holding 14.2% of the market between August 2024 and August 2025.
Google, Meta, OpenAI, which commanded 40%+ of downloads before 2023, are completely absent, with Comfy topping the list for the U.S. with 5.4% share.
The study also found that true open source is dying, with models disclosing their training data crashing from 79.3% in 2022 to just 39% in 2025.
Why it matters: The rapid ascent of Chinese models marks a changing of the guard. The open ecosystem has transitioned from a U.S.-led monopoly (historically led by Google) to a landscape where Chinese labs now provide the “brains.” This gap can increase further, with a wave of Chinese releases, led by DeepSeek, likely on the way.
QUICK HITS
🧠 Math V2 - DeepSeek’s open-source mathematical reasoning model
🤳 Perplexity - AI answer engine, now with persistent memory
🏆 GELab-Zero-4B - StepFun’s new SOTA, open-source computer use model
🛒 Vidi2 - ByteDance’s AI video editor with spatio-temporal grounding
Elon Musk’s xAI is reportedly set to raise a $15B round of funding at a $230B pre-money valuation next month, CNBC reported.
AI and agents drove $14.2B in global online sales on Black Friday, with $3B of this coming from the U.S. alone, Salesforce data revealed.
Virgin Australia signed a deal with OpenAI to embed ChatGPT-powered tools directly into how people search for and plan flights.
Data intelligence giant Databricks is in talks to raise $5B at a valuation of $134B — roughly 32x its expected sales of $4.1B for this year, The Information reported.
Avatar director James Cameron called gen AI “horrifying,” saying it makes up a character, an actor, a performance from scratch, unlike his movies’ performance capture approach that celebrates the “actor-director moment.”
Deutsche Telekom and Schwarz Group are reportedly planning a joint “AI gigafactory” in Germany, eyeing the EU’s $20B funding to rival U.S. and China.
Telegram CEO Pavel Durov launched Cocoon, a decentralized compute network to let GPU owners earn TON tokens for private AI processing, challenging cloud providers.
COMMUNITY
Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.
Today’s workflow comes from reader Martin K. in Bratislava, Slovakia:
“I use SimTheory AI to create a simple invoice payment assistant. I volunteer for a non-profit, and I'm responsible for paying about 30 invoices every month. I upload the PDF invoices to the assistant I created. It then recognizes all relevant information, enriches it with additional data for accounting and finance, and creates a table for an internal system and an XML file for batch payments. It works 100% correctly and increases my productivity 10-fold.
The next challenge is to use a Google Sheets MCP to automatically enter the data into the internal system.”
How do you use AI? Tell us here.
Read our last AI newsletter: DeepSeek returns with an IMO-crushing AI
Read our last Tech newsletter: Apple takes the crown from Samsung
Read our last Robotics newsletter: Figure sued over 'skull-crushing' force
Today’s AI tool guide: Use AI to find patents and innovation opportunities
RSVP to next workshop @ 4 PM EST Friday: Nano Banana Pro for Slide Decks
That's it for today!
See you soon,
Rowan, Joey, Zach, Shubham, and Jennifer—the humans behind The Rundown
