- Horizon AI
- Posts
- Anthropic Mathematician Says AI Has Solved the Most Famous Problem in Math 👀
Anthropic Mathematician Says AI Has Solved the Most Famous Problem in Math 👀
Why AI Engineers Fear Self-Building Artificial Intelligence ⚠️

Welcome to another edition of Horizon AI,
In today’s issue, we take a look at the growing concerns surrounding recursive self-improvement, a potential milestone where AI systems become capable of building more advanced versions of themselves without human involvement. With early signs of progress already emerging, researchers are increasingly focused on what this could mean for AI development, safety, and cybersecurity.
Let’s jump right in!
Read Time: 4.5’ min
Here's what's new today in the Horizon AI
Chart of the week: Top Countries by AI Use at Work
Anthropic Mathematician Says He Used AI to Solve Prestigious Millennium Prize Problem
AI Findings/Resources
AI tools to check out
Video of the week
TOGETHER WITH ROKU
It's time to set up your holiday streaming campaign
Are you ready to reach holiday shoppers on CTV? For many streamers, the season is already underway.
Last year, according to Roku internal data, holiday content searches on Roku saw an 86% increase during October. And in November, those searches rise an average of 39% weekly. This surging holiday spirit is mirrored by consumer activity, as about half of Roku streamers shopped Black Friday and Cyber Monday deals in 2025.
Meeting those shoppers is easier than you might think. Begin with the seasonal outcome you hope to drive: awareness, online sales, leads, store visits, or another measurable goal. From there, define your audience, upload your creative, set the budget and campaign dates, and establish measurement before launching. You can even create a shoppable ad unit that lets streamers buy with a click of their remote.
Our holiday playbook offers a month-by-month guide to reaching shoppers during this all-important time of year.
First-time spenders get a $5,000 match on the first $5,000 spent with code BEEHIIV5K. Terms apply.
Chart of the week
Top Countries by AI Use at Work
This visualization compares the 20 highest rates of workplace AI use among 47 countries in the 2025 study by the University of Melbourne and KPMG.
India ranked first, with 91% of surveyed employees reporting AI use at work, followed by China at 88% and Nigeria at 85%.
The six countries with the highest reported workplace AI use were all in Asia or Africa.
AI News
ANTHROPIC
Anthropic Mathematician Says He Used AI to Solve Prestigious Millennium Prize Problem

Anthropic mathematician Levent Alpöge says he has discovered a proof of the Riemann hypothesis, one of mathematics' most famous unsolved problems, with the help of Claude. First proposed in 1859, the conjecture has resisted attempts by mathematicians for 167 years.
Details:
Alpöge published his proposed proof in Lean, a programming language that allows mathematical arguments to be formally checked. The AI-generated version reportedly exceeds 250,000 pages.
He said the breakthrough came shortly before Anthropic's IPO and that he informed CEO Dario Amodei before deciding when to make the findings public.
Alpöge claims the approach also extends to the Generalized Riemann Hypothesis, with the possibility of addressing the Extended Riemann Hypothesis as well.
The Riemann hypothesis is one of seven Millennium Prize Problems, each carrying a $1 million award. Under the Clay Mathematics Institute's rules, a proposed solution must undergo at least two years of scrutiny and gain acceptance from the mathematics community before it can qualify for the prize.
On Saturday afternoon, two organizations announced that they had independently verified Alpöge’s claim after reviewing the programming language he used to share the solution.
Meanwhile, this is not Alpöge's first major AI-assisted mathematical claim. He previously said he had solved the 87-year-old Jacobian conjecture with Claude. OpenAI also claimed in September that its models had solved the Navier-Stokes problem, another Millennium Prize challenge, although that result has yet to receive formal recognition.
AI Findings/Resources
🤖 SpaceXSI now has five bots working full time on their accounting team, here's how they built them
💡 The founder's playbook: Building an AI-native startup. Anthropic shares how founders use AI at every stage, with exercises, frameworks, and prompts for Claude
🤿 This Reddit user built a swimmable underwater world in one 97 KB HTML file with Fable 5.2
AI Tools to check out
📈 Scout: AI assistant built for sales that lives in your Slack and text.
👗Daydream Agent: Your personal fashion AI shopping agent.
📝 Prompt Library: Explore AI prompts by category, including coding, writing, marketing, business, education, image and video generation, and more.
🪞 Second by First Person Labs: A portable model of you that you own and your agents can use.
Video of the week
Why AI Researchers Are Worried About Recursive Self-Improvement
This video from The Economist explores the growing industry concern surrounding recursive self-improvement (RSI), the potential for AI systems to independently design and build their own, more capable successors without human intervention .
The Rise of Recursive Self-Improvement:
RSI goes beyond writing code or assisting with training runs. It would require models to develop their own research ideas and judge which ones are worth pursuing.
One of the biggest remaining challenges is giving AI systems what researchers describe as "research taste." This means being able to identify important unsolved problems, choose the right experiments, recognize unexpected results, and decide what to investigate next.
Some researchers believe 2027 could be the year when recursive self-improvement arrives, or when the industry learns why it has not happened yet.
The Risks of Taking Humans Out of the Loop:
The biggest concern is that AI development could accelerate dramatically once systems can independently build more capable successors. Unlike human researchers, AI systems don't need to sleep or take breaks, potentially allowing progress to continue around the clock and at a much faster pace.
Removing humans from the improvement process could also weaken oversight. Safety monitoring, alignment, and decisions about how models should behave could become increasingly dependent on other automated systems.
AI could become very good at building smarter systems without becoming equally good at building safer ones.
Measuring Progress and Industry Transparency:
The panelists discuss a report from the organization Meter, which noted a 1.9x speed-up in Anthropic’s AI improvements attributed to model-led development. However, the panelists urge caution, noting that such reports are often heavily redacted and may serve as marketing narratives ahead of major company milestones.
However, they stress that concerns about RSI are not simply a marketing narrative. Researchers and engineers working inside major AI labs are reportedly taking the possibility seriously.
The Future of Cybersecurity and AI Access:
If RSI becomes a reality, the industry may see a "speed up" in technological development, leading to potential AI-assisted cyber risks, as well as more restrictive access to powerful "front-end" models to mitigate dangers.
The challenge will be ensuring that safety and oversight advance alongside the systems themselves.
While industry insiders expressed deep, off-the-record concern, many researchers remain within these labs, hoping to address these existential challenges from the inside.
That’s a wrap!
We'd love to hear your thoughts on today's email!Your feedback helps us improve our content |
Not subscribed yet? Sign up here and send it to a colleague or friend!
See you in our next edition!
Gina 👩🏻💻


