Good morning {{first_name| Artificial intelligence lovers}}. Google DeepMind has just taken an AI coding strategy and applied it to mathematics: Don’t ask for a model answer, give a team of agents the workspace.
The company’s AI co-mathematician has just set a new high in a benchmark designed to outperform AI for decades, and one professor even solved an unsolved problem using a strategy buried inside a manual rejected by the system’s auditors.
-
Associate mathematician at Google DeepMind
-
Roundtable Rundown: Our AI Use Cases
-
Automate any manual task with Codex
-
AI finds more than 100 new exoplanets from NASA data
-
4 new tools for AI, community workflow, and more
Google DeepMind
Rundown: Google DeepMind only published A research paper on a mathematician involved in artificial intelligence, an agent system based on Gemini 3.1 that is designed to help mathematicians tackle unsolved problems – setting a new high in the bar for research-level mathematics problems.
-
DeepMind modeled the tool after AI programming environments like Claude Code, bringing agent teams and embedded review cycles to mathematics research.
-
A coordinating agent divides the search into parallel workflows, each with sub-agents who write code, search the literature, and attempt proofs.
-
Mark Lakenby of the University of Oxford solved an open problem in the Korovka notebook after discovering a “really clever proof strategy” inside the rejected output.
-
In Epoch AI’s FrontierMath Tier 4, the system topped the leaderboard with a 48% score and more than double the Gemini 3.1 Pro’s raw score of 19%.
Why it matters: AI has already led to a surge in mathematical discoveries with advances in frontier models, and similar to programming, agent pipelines are now enabling AI systems to move forward. But as Lackenby’s discovery shows, the future is still bright for artificial intelligence that enables top minds to speed up their work, not replace it.
TOGETHER WITH GOOGLE FOR STARTUPS
Rundown: The Future of AI report from Google for Startups is your essential guide to understanding how generative media is reshaping product development, offering founders strategic insights to build smarter businesses, scale faster, and stay ahead of the AI curve.
Inside the report, you will discover:
-
How to leverage digital twins at scale.
-
Strategic insights to differentiate AI products.
-
Expert perspectives on the obstetric landscape.
RUNDOWN ROUND TABLE
Image source: Ideogram/The Rundown
Rundown: The Rundown Roundtable is a weekly feature where we poll The Rundown staff members on how we use AI in our work or daily lives.
Jason Developer: I used /goal in OpenAI’s Codex to create a Magic: The Gathering app so my brother and I could play asynchronously without having to coordinate a call or awkwardly play over FaceTime.
The idea is to let each of us take turns when we have time, keep track of board status cleanly, and keep the game going over days rather than trying to arrange schedules. The command allowed the Codex to continue running until everything was finished, essentially, which was exactly what I was looking for at once without any intervention.
Joy, Partnerships: I had never been to Greece before, so for my next trip, I went all out and handed the entire itinerary over to Claude. Booked flights, dialed transit times, and restaurant menus organized by city.
I now offer a more rigorous plan than most travel agents can put together!
Artificial intelligence training
Rundown: In this guide, you will learn how to let Codex click through any annoying and repetitive work using your computer on Mac or Windows.
-
Open Codex, go to Plugins, find and enable the Computer Use plug-in, and start a new task
-
Open the permissions menu and switch from default permissions to full access, then confirm any prompts and give Codex something real to do
-
Example: “Open Chrome and debug the UI of this web page I’m developing http://localhost:3000/. Click, reproduce the error I describe, and then tell me what you think is causing it. If you are not sure, ask before making changes.”
Pro Tip: Codex can automate repetitive workflows in native applications, too – try it with Photoshop exports, Adobe Premiere cleanups, file renaming, or any other tool.
Provided by Oracle Developers
Rundown: Small language models can solve more difficult inference tasks without changing their weights. Oracle Developers’ open source proxy inference code demonstrates how to add research-backed formatting to Ollama models, with 16 inference strategies that developers can test locally.
In the guide, you will explore:
-
Ollama’s open source logic code
-
16 strategies, measured across 4200 runs
-
Better accuracy without retraining models
Artificial intelligence and astronomy
Rundown: Astronomers from the University of Warwick certain More than 100 exoplanets Using an AI system called RAVEN that scanned 4 years of NASA TESS data covering 2.2 million stars, RAVEN also found more than 2,000 additional potential candidates.
-
RAVEN handles detection, screening and confirmation in one shot, and is trained to simulate planets and false alarm signals to filter out real detections.
-
The results included 31 exoplanets that had never been observed before, in addition to alien worlds that orbit their stars in less than one day.
-
Hundreds of exoplanets have been found in the “Neptune Desert,” a region so close to a star that Neptune-sized planets should not survive the heat.
-
The system measures the prevalence of different planet types with 10 times the precision of previous systems using smarter AI alone, not new hardware.
Why it matters: Humans have confirmed the existence of only a few thousand exoplanets so far, and it is estimated that there are trillions. Artificial intelligence and technological advances will help quickly rewrite this number – and based on RAVEN, all it will take are improved models and AI integrations to uncover knowledge about space that is already hidden in the data we have.
-
🔒 Stealth – Remove your personal data from the web so scammers and identity thieves can’t access it. Use the code He turned over To get a 55% discount*
-
💻 Codex Alimentarius in Chrome – OAI’s Codex extension for proxy tasks within Chrome
-
🧠 Ernie 5.1 – Baidu’s new foundation model with powerful search capabilities
-
🖨️ Printing press – CLI Factory with 30+ pre-built native agent tools
Google Replica Labs And it is said to lift +$2 billion to expand its drug design engine, which it says significantly outperforms AlphaFold 3 at specific tasks.
Greece He is suggestion Protecting artificial intelligence in its constitution, it requires technology to serve individual freedom, with Prime Minister Mitsotakis noting the threats to democracy.
Baidu Released ERNIE 5.1, a new AI rating that ranks fourth on Arena’s search leaderboard, with the company claiming training costs just 6% less than competing models.
OpenRouter Fired Pareto Code, a free routing layer that automatically selects the cheapest coding AI above a user-specified quality bar, adjusting prices as newer models improve.
Softbank Group Communications arm Fired A battery company building large-scale cells and storage systems – and meeting the power demands of data centers under development.
In each newsletter we show how the reader is using AI to work smarter, save time or make life easier.
Today’s workflow comes from the reader unknown:
“I’ve been using ChatGPT for various things professionally, and it’s been surprisingly useful and refreshing. But the biggest use I’ve found for it so far is helping me train my four dogs.
I was willing to spend thousands of dollars on a professional trainer just because of how chaotic it was, but ChatGPT helped me identify the root causes of certain behaviors and taught me how to successfully train around and outside of them using specific techniques designed specifically for my individual dogs.
“The confidence you brought me, and the positive reinforcement, changed every dynamic in the family, and I wish I had started sooner.”
How do you use artificial intelligence? Tell us here.
That’s all for today!
Before you go, we’d love to know what you thought of today’s newsletter to help us improve The Rundown experience for you.
Rowan, Joey, Zack, Shubham and Jennifer – the humans behind The Rundown
