OpenAI outperforms humans and Google at the world's top collegiate programming contest

Key Takeaway: OpenAI’s ensemble system, led by GPT-5 plus an experimental model, solved all 12 ICPC World Finals 2025 problems under standard rules, surpassing top human teams and Google DeepMind’s Gemini 2.5.

More Insights:

• Ran under the same five-hour, PDF-only conditions; submissions judged by official ICPC evaluators.

• Outperformed DeepMind’s gold-level Gemini (10/12), which uniquely cracked Problem C that no human solved.

• GPT-5 generated 11 correct solutions; the experimental model picked submissions and solved the final hardest task after 9 tries.

• Best human team solved 11/12; OpenAI’s perfect score would have won outright if entered as a team.

• Follows gold-level results at IMO and IOI; team frames next milestone as systems that can discover new knowledge.

Why it matters: If general-purpose models can match—and now exceed—elite human performance in open-ended, time-boxed contests, the center of gravity in problem-solving shifts from human ingenuity alone to human-AI strategy, with implications for education, hiring, and research—and a looming question: what happens when these systems start pushing beyond benchmarks to generate original discoveries?

Nano Banana + Seedream 4.0 - Innovation Meets Imagination Inside Pippit

Pippit fuses Google’s Nano Banana with ByteDance’s Seed 4.0, bringing together two of the strongest image-generation models on the market. This dual-engine design makes Pippit one of the most powerful platforms for AI-driven visual creation​ setting a new benchmark for what’s possible in design.

• One prompt drops you anywhere. Moon to anime scenes beside your idol.

• Instantly restyle yourself: new hair, makeup, and global fashion.

• One-click photo magic: revive blurry pics or turn portraits into Pixar/cyberpunk art.

Head over to pippit.ai today to claim your free week, and if you want to stick with it, use code PIPPIT45OFF for 45% off your subscription.

Mark Zuckerberg has begun his quest to kill the smartphone

Key Takeaway: Meta unveiled Ray-Ban Display smart glasses plus a Neural Band wristband that uses sEMG to let you silently “write” texts in the air—its boldest bid yet to move everyday computing from smartphones to face-wear.

More Insights:

• Reality Labs claims ~$70B in losses since 2020, but Connect 2025 finally showed what the spend built.

• Glasses add an offset display for Instagram/WhatsApp/Facebook, maps, and live translation; cameras, speakers, mics, and on-board AI remain.

• Neural Band reads nerve signals to type; Zuckerberg says he’s hitting ~30 wpm (participants averaged ~21 wpm vs ~36 wpm on phones).

• Silent input means no awkward voice commands in public; other wrist gestures echo Apple Watch/Joy-Con controls.

• Meta frames glasses as “more present” and less distracting than phones—while also seeking to cut Apple/Google app-store tolls.

Why it matters: If voiceless, wrist-driven typing feels natural at scale, everyday computing could shift from pocket rectangles to wearable displays—redefining social norms, platform power, and how (and who) controls the next trillion-dollar attention gateway.

Related: Mark Zuckerberg’s Meta ‘AI glasses’ fail live demo, video emerges

Chrome Gets a Brain: Google Rolls Out Gemini, Agentic Browsing, and Built-In AI Search

Key Takeaway: Google is baking Gemini directly into Chrome for all U.S. desktop users, adding multi-tab reasoning, conversational search in the address bar, upcoming agentic task completion, and new safety/password tools to turn the browser into an AI helper.

More Insights:

• Gemini in Chrome is now available to U.S. Mac/Windows users (English), with on-page clarifications via the Gemini icon.

• New browsing smarts: compare/summarize across multiple tabs and soon recall sites you visited (“Where did I see that walnut desk?”).

• AI Mode comes to the address bar for complex, follow-up searches and page-aware suggested questions (rolling out later this month in U.S.).

• Agentic actions “coming months”: navigate sites, build carts, schedule with Calendar/YouTube/Maps—user approves final checkout.

• Safety upgrades: Gemini Nano scam detection and one-click password resets on supported sites (e.g., Coursera, Spotify, Duolingo, H&M).

Why it matters: If the browser becomes an AI agent that understands context, remembers intent, and executes steps, the center of gravity shifts from typing queries and clicking links to delegating outcomes—reshaping search, shopping, and app workflows while raising new stakes for privacy, transparency, and trust.

Other stuff

• OpenAI’s research on AI models deliberately lying is wild

• How thousands of ‘overworked, underpaid’ humans train Google’s AI to seem smart

• China bans tech companies from buying Nvidia’s AI chips

• Anthropic Economic Index report: Uneven geographic and enterprise AI adoption

• China's DeepSeek says its hit AI model cost just $294,000 to train

• ChatGPT is developing age verification system to identify under-18 users after teen death

• AI fares better than doctors at predicting deadly complications after surgery

• YouTube rolls out Studio updates, ‘likeness’ detection, lip-synced dubs, creator collabs, and more

Hope you enjoyed today's newsletter

⚡️ Join over 300,000 people using the Superpower ChatGPT extension on Chrome and Firefox.