- NATURAL 20
- Posts
- Grok 4.20 Stuns Scientists with Instant Math Proofs
Grok 4.20 Stuns Scientists with Instant Math Proofs
PLUS: OpenAI Pushes for a US-Made AI Supply Chain, Microsoft Launches Free AI Training for Schools and more.
In partnership with
The headlines that actually moves markets
Tired of missing the trades that actually move markets?
Every weekday, you’ll get a 5-minute Elite Trade Club newsletter covering the top stories, market-moving headlines, and the hottest stocks — delivered before the opening bell.
Whether you’re a casual trader or a serious investor, it’s everything you need to know before making your next move.
Join 200K+ traders who read our 5-minute premarket report to see which stocks are setting up for the day, what news is breaking, and where the smart money’s moving.
By joining, you’ll receive Elite Trade Club emails and select partner insights. See Privacy Policy.
Hey friend,
Happy Friday! I hope you’ve had a great week and are ready to slide into the weekend. We’ve got everything from Google breaking down language barriers to OpenAI literally trying to connect our brains to computers. Let’s dive in.
Today:
Grok 4.20 Stuns Scientists with Instant Math Proofs
Google Unveils TranslateGemma to Bridge Language Gaps
Wikipedia Turns 25 and Signs Major Enterprise Partners
OpenAI Backs Brain-Computer Interface Startup Merge Labs
OpenAI Pushes for a US-Made AI Supply Chain
Microsoft Launches Free AI Training for Schools
GROK 4.20 cracked the code
Grok 4.20 amazed researchers when a UC Irvine professor watched it craft a new Bellman function (a razor-sharp safety formula) for a tricky “don’t fall off the cliff” problem in five minutes. The feat shows AI can now find new proofs on its own, a skill also surfacing in GPT-5.2 and hailed by star mathematician Terence Tao after recent AI solutions to famous Erdős puzzles.
XAI is offering Grok 4.20 to scientists while Elon Musk battles OpenAI in court; betting odds favor him. Musk also sealed a data-rich deal with the revived U.S. Department of War. Rumored Grok spin-offs Slateflow and Tidewisp loom soon.
OpenAI says it’s participating in Merge Labs’ seed round because brain–computer interfaces could become a more natural, human-centered way to interact with AI over time.
What caught my eye: OpenAI frames this as an “interface leap”—the same kind of jump we’ve seen from command lines → mice → touch → voice… except this one is about expressing intent with far less friction.
A few outlets add extra color: Wired reports Merge Labs is aiming for non-invasive approaches (not surgical implants) and that the round is large.
Why this matters (even if it’s early):
If BCIs get good, “AI assistants” stop being apps and start feeling like a native layer of computing.
The hard part isn’t just hardware — it’s interpreting noisy, limited signals reliably (OpenAI explicitly calls that out).
This is a long runway story… but it’s one of the few that could genuinely change how we use everything else.
Google introduced TranslateGemma, a family of open translation models built on Gemma 3, in 4B / 12B / 27B sizes, supporting 55 languages.
The spicy bit: Google says the 12B model can outperform the Gemma 3 27B baseline on WMT24++ using MetricX—basically a “smaller model, better translation” claim.
They also position it as “runs everywhere”: 4B for mobile/edge, 12B on consumer laptops, and 27B as the max-fidelity option (they mention a single H100/TPU in the cloud).
Why this matters:
Translation is quietly becoming infrastructure for global apps, support, education, and commerce.
The “small but strong” angle is a big deal if you want on-device or low-latency translation without shipping everything to a giant server.
There’s a technical report on arXiv if you want the full eval/training details.
Wikimedia Enterprise announced new partners (including Amazon, Meta, Microsoft, Mistral AI, and Perplexity) joining a roster that already includes others like Google.
Their pitch is straightforward: Wikipedia is incredibly valuable in the AI era, and Enterprise offers high-throughput, reliable access through different options (on-demand, snapshot downloads updated hourly, and realtime updates).
Reuters and The Verge frame it more bluntly: it’s part of getting big companies to pay for structured access instead of scraping, helping sustain Wikimedia as usage costs rise.
Why this matters:
“Trusted data” is becoming a competitive moat — and Wikipedia is basically saying: if you’re going to build on this, help fund the pipes.
For builders doing RAG/knowledge graphs, predictable feeds + snapshots can be the difference between “demo” and “production.”
🧠RESEARCH
This study improves how AI writes code by using a "self-evolution" process. It forces the system to plan diverse strategies and learn from both successes and failures. By mimicking natural selection, the AI refines its programming skills to solve harder problems without human help.
Evaluating AI researchers is hard, so this team built a tool to do it automatically. It creates complex, realistic research questions and then judges the AI’s answers for accuracy. This ensures that AI agents can find and verify facts correctly before they are used by people.
Current AI agents often get stuck when solving long problems. MAXS fixes this by letting the AI "look ahead" at future steps before acting. It measures which path is most likely to succeed, helping the system stay on track and use its computing power more wisely.
🛠️TOP TOOLS
Each listing includes a hands-on tutorial so you can get started right away, whether you’re a beginner or a pro.
Canva GPT – Effortlessly Design Anything - lets you create, preview, and edit Canva designs directly in the chat—everything from slide decks to social posts and logos—then open the result in Canva for full editing.
CapSolver – Captcha Solver - AI-based CAPTCHA‑solving service aimed at developers who need automated handling of challenges such as Google reCAPTCHA
Caricaturer IO – Free AI Caricature Maker - web-based AI tool that turns photos into exaggerated caricatures and can also generate caricature images from text prompts.
📲SOCIAL MEDIA
🗞️MORE NEWS
OpenAI Strengthening the Supply Chain OpenAI is asking American companies to build the physical equipment needed for AI, such as cooling systems and power cables. They want to move this "supply chain"—the network of businesses that make and deliver products—to the US to create jobs and reduce reliance on other countries. This initiative aims to speed up the growth of AI infrastructure by making the essential parts domestically.
Microsoft Free AI Training Microsoft has released a free training program to help teachers and students understand how to use artificial intelligence. The course breaks down complex technology into simple lessons on using tools like Copilot to improve teaching and learning. It includes four modules designed to help schools adopt AI without paying expensive fees.
Replit Vibe Coding Replit launched a new feature that writes computer code for you when you simply describe your idea in plain English. This allows people to build mobile apps by explaining the "vibe" or feel they want, rather than typing out technical instructions. It uses Google's powerful AI to handle the hard work, letting anyone create software quickly.
Higgsfield AI Funding A video startup called Higgsfield is now worth $1.3 billion after investors gave them $80 million to grow their business. The company provides AI tools that help marketing teams create professional videos for social media without needing expensive cameras or crews. Their system connects different AI models to ensure the videos look consistent and high-quality.
Tulip Raises $120M Tulip raised $120 million to build software that helps factory workers do their jobs better rather than replacing them with robots. Their tools give employees real-time data on tablets to track production and fix problems directly on the assembly line. They call this "human-centric" because it focuses on empowering the person instead of just automating the work.
Cloudflare Acquires Human Native Cloudflare bought a company called Human Native to help content creators get paid when AI companies use their work. Human Native operates a marketplace where writers and artists can legally sell their data to be used for training AI systems. This acquisition—or purchase—aims to solve the problem of AI stealing content by making it easy to pay owners fairly.
TSMC Record Profits TSMC, the company that makes the "chips" or computer brains used in almost all electronics, reported record profits due to massive demand for AI. They expect their sales to grow significantly for several years as companies buy more hardware to run their AI models. To keep up with this boom, they are spending billions to build new factories in the US and Taiwan.
Andrea Vallone Joins Anthropic A top safety expert named Andrea Vallone has left OpenAI to join its competitor, Anthropic. She previously led the team responsible for ensuring OpenAI's chatbots did not give harmful or dangerous answers to users. Her move highlights the intense competition for experts who specialize in "alignment," which means making sure AI follows human rules.
OpenAI Walks Away from Apple OpenAI reportedly refused to be the main partner for Apple's new AI features so it could focus on building its own devices. Because of this, Apple signed a multi-billion dollar deal with Google to use their Gemini AI on iPhones instead. This decision signals that OpenAI wants to compete directly with big tech giants rather than just supplying them with technology.
What'd you think of today's edition? |
Reply