In partnership with
In Today’s Issue:
⚛️ Google’s "Aletheia" agent is now solving open mathematical conjectures
🏗️ Mark Zuckerberg breaks ground on a $10 billion data center in Lebanon
📉 New economic data shows the US labor's share of income has hit a historic low of 51.4%
🚀 GLM-5 introduces a novel asynchronous reinforcement learning stack
✨ And more AI goodness…
Dear Readers,
China's GLM-5 just dropped 744 billion parameters onto the open-source stage, and the benchmarks say it's breathing down the neck of Claude Opus 4.5 and GPT-5.2, all under an MIT license that lets anyone run it on everything from NVIDIA GPUs to Huawei Ascend chips.
Today's issue unpacks why this might be the moment open-source AI truly closes the frontier gap, but that's far from all: Google's Gemini Deep Think has gone from acing math Olympiads to autonomously solving decades-old research problems, Meta is pouring $10 billion into a single gigawatt-scale data center in Indiana that rewrites what AI infrastructure looks like, and new data shows the AI boom is generating trillions in value while workers' share of the pie keeps shrinking.
We've also got a breakthrough stem cell trial that could actually restore human hearing for the first time in history, plus Demis Hassabis sitting down to explain where DeepMind thinks all of this is heading.
Grab your coffee, this one's dense.
All the best,
🤖 AI Boom, Workers Left Behind
The U.S. is riding a massive AI-driven investment wave, but the economic rewards are flowing far more to capital than to workers. Labor’s share of gross domestic income has dropped from 58% in 1980 to 51.4% today, while corporate profits climbed from 7.2% to 11.7%, a shift that would equal $2 trillion more annually, or about $12,000 per worker, if reversed.
🚀 Gemini Deep Think Redefines Research
Google’s Gemini Deep Think has leapt from winning gold-level benchmarks at math and programming Olympiads to solving real-world PhD-level research problems across mathematics, physics, computer science, and economics. Its math agent “Aletheia” scored up to 90% on IMO-ProofBench Advanced, tackled 700 open Erdős problems, autonomously solved four, and even helped generate publishable-level papers, while introducing structured human-AI collaboration models like “Advisor” and “balanced prompting.”
From cracking decade-old optimization conjectures to advancing cosmic string physics and upgrading auction theory for AI economies, Gemini is emerging as a true scientific co-pilot, handling verification, counterexamples, and cross-field reasoning so researchers can move faster and think bigger!
⚡ Meta Unveils $10B AI Powerhouse
Another 1GW Data Center incoming:
Meta has broken ground on a massive 1-gigawatt data center campus in Lebanon, Indiana, marking a $10+ billion investment, one of its largest infrastructure bets ever. The facility will power Meta’s growing AI ambitions while supporting its core platforms, creating 4,000+ construction jobs at peak and around 300 permanent roles, alongside workforce programs for local students.
Beyond compute, Meta is pledging serious community and sustainability commitments: $1M annually for 20 years to help families with energy bills, $120M+ for water and public infrastructure, and a plan to match 100% of the site’s energy use with clean energy while restoring 200 million gallons of water annually. This signals how AI’s future isn’t just about models, it’s about gigawatt-scale infrastructure, local impact, and long-term environmental strategy. Source: Meta Newsroom.
Can Google’s New AI Solve Everything? With Google DeepMind CEO Demis Hassabis
China's GLM-5 Closes the Frontier Gap
The Takeaway
👉 Zhipu AI's GLM-5 matches or beats proprietary models like Claude Opus 4.5 and GPT-5.2 on key agentic and coding benchmarks while being fully open-source under MIT License
👉 The model scales to 744B parameters with only 40B active, uses DeepSeek Sparse Attention for cost-efficient deployment, and runs on non-NVIDIA hardware including Huawei Ascend and other Chinese chip platforms
👉 GLM-5 integrates with existing developer tools like Claude Code and supports end-to-end document generation (docx, pdf, xlsx), signaling a shift from chatbot to productivity agent
👉 Its novel "slime" RL infrastructure for asynchronous training could influence how the broader community approaches reinforcement learning at scale
China's Zhipu AI just dropped GLM-5, and the open-source AI race has a new heavyweight contender. The 744B-parameter model (40B active through mixture-of-experts) is purpose-built for what the company calls "complex systems engineering and long-horizon agentic tasks." Translation: this model doesn't just chat, it works.
The benchmarks tell a compelling story. GLM-5 scores 77.8% on SWE-bench Verified, breathing down the neck of Claude Opus 4.5 at 80.9%. On BrowseComp, it actually beats both Claude and GPT-5.2. Perhaps most striking: on Vending Bench 2, a simulation where models run a virtual vending machine business for a full year, GLM-5 finished with $4,432, nearly matching Anthropic's flagship.
Released under MIT License with full weights on HuggingFace, GLM-5 runs on everything from NVIDIA GPUs to Huawei Ascend chips. It integrates directly with Claude Code, supports document generation, and comes with a free-tier chat at Z.ai.
The model also introduces "slime," a novel asynchronous reinforcement learning infrastructure that dramatically improves training throughput, a technical contribution the broader community can learn from.
Why it matters: GLM-5 represents the strongest evidence yet that open-source models can compete with proprietary frontier systems on real-world engineering tasks. Its MIT license and broad hardware support could democratize access to near-state-of-the-art AI capabilities globally.
Sources:
🔗 https://z.ai/blog/glm-5
Turn AI into Your Income Engine
Ready to transform artificial intelligence from a buzzword into your personal revenue generator?
HubSpot’s groundbreaking guide "200+ AI-Powered Income Ideas" is your gateway to financial innovation in the digital age.
Inside you'll discover:
A curated collection of 200+ profitable opportunities spanning content creation, e-commerce, gaming, and emerging digital markets—each vetted for real-world potential
Step-by-step implementation guides designed for beginners, making AI accessible regardless of your technical background
Cutting-edge strategies aligned with current market trends, ensuring your ventures stay ahead of the curve
Download your guide today and unlock a future where artificial intelligence powers your success. Your next income stream is waiting.
World's First Stem Cell Trial Targets Hearing Loss Reversal
Over 1.5 billion people worldwide live with hearing loss, and until now, every treatment has been a workaround. Hearing aids amplify, cochlear implants bypass, but nothing actually repairs the damage. That's about to change.
UK-based Rinri Therapeutics is running the world's first stem cell trial designed to regenerate auditory nerves. Their therapy, Rincell-1, uses lab-grown neural progenitor cells derived from induced pluripotent stem cells (iPSCs), essentially reprogrammed adult cells coaxed into becoming the building blocks of hearing. Approved by British regulators in July 2025, the Phase I/IIa trial is now underway across three UK centers with 20 adults suffering from severe sensorineural hearing loss.
The science builds on over a decade of preclinical breakthroughs. Researchers previously restored auditory responses in deaf gerbils using transplanted human stem cells that successfully reconnected severed neural pathways. Now, half the trial participants receive Rincell-1 alongside a cochlear implant while the other half get the implant alone, a direct head-to-head test of regeneration versus compensation.
Early data could arrive by late 2026, with complementary approaches like CRISPR gene editing and gamma-secretase inhibitors potentially amplifying results even further.
This trial marks the first time stem cell therapy is being tested to actually restore human hearing rather than compensate for its loss. Success could redefine treatment for the 2.5 billion people projected to experience hearing loss by 2050.
Learn AI in 5 minutes a day
This is the easiest way for a busy person wanting to learn AI in as little time as possible:
Sign up for The Rundown AI newsletter
They send you 5-minute email updates on the latest AI news and how to use it
You learn how to become 2x more productive by leveraging AI
