DeepSeek Outsmarts the World, NVIDIA Challenges the Rules, and OpenAI Gets a Reality Check

In partnership with

👋 Morning, Reader!

AI’s hitting milestones faster than you can say “debug my logic.” From DeepSeek’s open-source genius outsmarting humans at math Olympiads to OpenAI’s minor leak scare, and NVIDIA questioning the religion of scale, this week’s lineup is equal parts brilliance and big-picture thinking.

Let’s crunch it. 🧮

🧮 DeepSeek’s New Math Model Outsmarts the Competition

The open-source brainiac just beat Google’s best.

DeepSeek-Math-V2 is shaking up the frontier model world. The new open-source MoE (mixture of experts) model just pulled off something incredible — achieving gold medal performance at IMO 2025, the world’s toughest math competition.

The details:

• Scored 118/120 on the 2024 Putnam exam — higher than the top human.

• Solved 5 of 6 IMO 2025 problems, matching the gold-winning score.

• Crushed GPT-5 (20%) and nearly tied Google’s Gemini Deep Think (61.9%) on proof reasoning.

• Uses a generator-verifier system — one model writes the proof, the other critiques it line-by-line for logical soundness.

Why it matters:
This isn’t just a math flex — it’s a breakthrough in reasoning. DeepSeek has basically created an AI that argues with itself to find truth, then open-sourced it. That’s a direct shot at proprietary labs like Google and OpenAI, and a major win for open research.

🧠 Proof that open-source might be the smartest move in AI yet.

Find customers on Roku this holiday season

Now through the end of the year is prime streaming time on Roku, with viewers spending 3.5 hours each day streaming content and shopping online. Roku Ads Manager simplifies campaign setup, lets you segment audiences, and provides real-time reporting. And, you can test creative variants and run shoppable ads to drive purchases directly on-screen.

Bonus: we’re gifting you $5K in ad credits when you spend your first $5K on Roku Ads Manager. Just sign up and use code GET5K. Terms apply.

Use code GET5K now

🔐 OpenAI Confirms Data Leak Through Third-Party Vendor

Not catastrophic, but a good reminder: your data’s only as safe as your partners.

OpenAI reported a third-party breach at its analytics vendor Mixpanel, which exposed limited profile info from API users. The company says no API keys, payment info, or chat data were touched.

What happened:

• Mixpanel’s systems were breached on Nov 9, affecting OpenAI’s API product analytics.

• Leaked data included names, emails, cities, and device info — but not chats or financials.

• OpenAI has cut ties with Mixpanel and is alerting all impacted users.

• The real risk: phishing or impersonation using stolen API-related info.

Why it matters:
While the core systems stayed secure, this is a classic reminder that the weakest link often lies outside your fortress. Expect more scrutiny on how major AI companies manage third-party integrations — especially as enterprise clients pour billions into API infrastructure.

🚨 Lesson: sometimes “AI security” just means checking who’s holding your analytics data.

🎻 NVIDIA’s New Paper Says: Bigger Isn’t Always Better

Meet ToolOrchestra — the AI conductor that might make mega-models obsolete.

In a surprising twist, NVIDIA and the University of Hong Kong are arguing that the future of AI might not be about scaling models infinitely — but orchestrating them intelligently.

Introducing ToolOrchestra:

• Trains a small “orchestrator” model to decide when to use tools or other models.

• An 8B parameter model trained this way outperformed GPT-5 and Claude Opus 4.1 on complex reasoning benchmarks like Humanity’s Last Exam.

• It’s 2.5x faster and dramatically cheaper than frontier-scale systems.

• Avoids overusing costly tools, adapting efficiently to different environments.

Why it matters:
This could upend the “bigger is better” ideology that’s defined the AI arms race. Instead of training trillion-parameter giants, the next era might belong to smaller, smarter AIs that know how to collaborate.

🎶 Think less orchestra soloist, more symphony conductor.

🗞️ Other News

🚀 Jeff Bezos’ stealth AI venture, Project Prometheus, quietly acquired General Agents, a startup building infrastructure for agentic computing.

⚖️ OpenAI lost a key discovery ruling — a court ordered the company to hand over internal messages about deleted book-training datasets, a big win for authors suing over copyright infringement.

🧠 Perplexity launched persistent memory, letting its assistant remember user preferences, interests, and conversation context for continuity across sessions.

👓 Alibaba’s Quark AI Glasses hit the Chinese market at 1,899 yuan ($268), offering real-time translation, navigation, and vision-based assistance powered by its Qwen models.

🔋 Elon Musk’s xAI is building an 88-acre solar farm beside its Colossus data center — covering 10% of its power needs and expanding into battery storage next year.

💻 Mastery tools of the day

• Grain AI makes fun and interactive lessons from any topic with visuals, exercises, and web apps that match your grade level and style preferences

• MakeShot converts text prompts into videos using 8+ AI models including Sora 2, Veo 3, and Kling with commercial rights included

• LaunchLemonade builds AI agents from 21+ models that handle sales, support, and marketing 24/7 without coding or vendor lock-in

💡What else are we reading and seeing?

• Technical Deflation

• The Math of Why You Can't Focus at Work

• Sam Altman's Business Buddies Are Getting Stung

• More Articles Are Now Created by AI Than Humans

• We're going to see a lot of carnage': VC investor says AI boom will create giants — and topple overhyped startups

⚡ The Takeaway

This week’s AI news reads like a masterclass in humility:

• DeepSeek proves open-source innovation can match Big Tech.

• NVIDIA reminds us scale isn’t everything.

• OpenAI learns that even the smartest systems can’t escape human mistakes.

As the industry matures, the real breakthroughs might come not from bigger brains, but from better coordination. 🧩

Earn a master’s in AI for under $2,500

AI skills aren’t optional—they’re essential. Earn a Master of Science in AI, delivered by the Udacity Institute of AI and Technology and awarded by Woolf, an accredited institution. During Black Friday, lock in savings to earn this degree for under $2,500. Build deep AI, ML, and generative expertise with real projects that prove your skills. Take advantage of the most affordable path to career-advancing graduate training.

Learn More

✨ Thanks for reading Not So Artificial!

If this edition solved more problems than your calculator, forward it to a friend who still thinks “AI news” means robot apocalypse stories.

Catch you tomorrow — same inbox, same chaos, more clarity. ⚙️