The Machines Are Still Smiling.

When the AI Stops Explaining Itself, Start Worrying

In partnership with

I love lifetime deals. You love lifetime deals. This is our week.

Rocketeers! It’s like BFCM in July --- RocketWeek is back and live! 🚀 For 1 week, we're bringing Rocketeers amazing products and discounts. Most will hang out on RocketHub for longer but the best price will be during their first 7 days. Visit the RocketWeek page to see the lineup.

Top AI researchers from Google, OpenAI, and others are raising alarms: today’s smartest AI might soon start “thinking” in ways we can’t track or even understand. These systems use a process called chain-of-thought (CoT) to break down problems step-by-step—but that logic isn’t always visible or even human-readable. The big worry? AIs could hide shady reasoning or misaligned goals without us catching it. Researchers are pushing for better monitoring tools, but there’s no guarantee future models won’t outsmart our oversight. Basically, AI’s getting too clever—and we might not see the red flags in time.

Google DeepMind just hit a major milestone—its advanced Gemini model with “Deep Think” mode scored a gold medal-level 35/42 at the 2025 International Math Olympiad. Unlike last year’s silver attempt that relied on translating math into formal code, this time Gemini solved 5 out of 6 problems straight from natural language—within the 4.5-hour limit, just like human contestants. The answers were officially graded by IMO, making it the real deal. OpenAI also got the same score, but their results weren’t certified. DeepMind’s secret? A new reasoning setup that thinks in parallel and learns from top-tier math data. TL;DR: AI’s getting scary good at math

Alibaba just dropped an upgraded version of its Qwen3 model, and it’s making serious waves. The new release ditches hybrid thinking for a cleaner, smarter setup—and it’s outperforming both Kimi K2 and even nudging up against Claude Opus 4 and GPT-4o. It runs 22B of 235B parameters with a massive 256K context window, and yes, it’s 100% open-source. You can try it for free on Qwen Chat or grab it from Hugging Face. In short: China’s coming for the AI crown, and they’re doing it in the open.

Moonvalley, started by ex-DeepMind folks, just dropped Marey—an AI video model made just for filmmakers and trained only on licensed footage (no copyright chaos here). Unlike the usual AI roulette, Marey gives directors full control over camera angles, lighting, and character moves, and plugs right into VFX workflows. Clips cost around $1-2 to render, with pricing starting at $14.99/month. Backed by over $100M and paired with its own AI animation studio, Marey feels like a serious bet on ethical, pro-grade AI in Hollywood.

OpenAI says it’d totally be down to buy Google Chrome—if the courts ever force Google to sell it. The comment came during a major antitrust trial where the DOJ is trying to break up Google’s dominance in online search. Google’s not having it, insisting Chrome isn’t up for grabs and calling the lawsuit bogus. OpenAI previously tried partnering with Google (no luck), and now it’s riding with Microsoft and Bing. Meanwhile, Google’s flexing Gemini, and OpenAI might even be cooking up its own social network. The AI wars just keep getting juicier.

The new source for curated lifetime deals and content for entrepreneurs, by entrepreneurs.

Why pay a high monthly subscription fee when you can save money by getting lifetime deals on your essential SaaS tools? For example, look at these no-brainer deals:

1. Leadinary - The all-in-one tool to find and research local businesses for lead generation.

2. Taskera - Bring order to your project management chaos with Taskera.

3. Funded Startups List - Leads. In your inbox. Every single month. Become a Rocketeer today and never pay full price for software ever again!

See Lifetime Deals

Prompt: “Write the body copy for a promotional email campaign about GAP spring sale. Write in an upbeat and friendly tone.”

Find out why 1M+ professionals read Superhuman AI daily.

In 2 years you will be working for AI

Or an AI will be working for you

Here's how you can future-proof yourself:

  1. Join the Superhuman AI newsletter – read by 1M+ people at top companies

  2. Master AI tools, tutorials, and news in just 3 minutes a day

  3. Become 10X more productive using AI

Join 1,000,000+ pros at companies like Google, Meta, and Amazon that are using AI to get ahead.

a

Simplify your work with these fun and effective AI tools.

  1.  Sunflower - quiet revolution against the chaos of email.

  2. BillSplit - the easiest way to split your restaurant bill. 100% free and open source

  3. Den - shared workspace for humans and agents to collaborate.

  4. Unsloth - Finetune LLMs 2x faster, 80% less memory.

  5. BaansAI - Design ad banners with your logo and custom text.

That’s it. See you on the next open. And do me a favor and move this email to your “Primary” inbox - trust me, it’ll be worth it.

- Charlie “Your AI Assistant” Patel