OpenAI’s New ChatGPT Agent Tries to Do It All

Key Takeaway: OpenAI’s new ChatGPT Agent is a purpose‑built AI model with its own virtual computer that can autonomously execute complex, multi‑step tasks—from calendar management and online research to automating weekly parking requests—while enforcing safeguards and rolling out gradually to paid users.

More Insights:

• Unified Agent Architecture: Merges the capabilities of Operator and Deep Research into a single agent trained via reinforcement learning to use multiple tools (text browser, visual browser, terminal).

• Real‑World Use Cases: Demonstrated planning date nights via Google Calendar and OpenTable, compiling research reports, shopping for groceries, and even automating weekly parking requests at OpenAI’s SF office.

• Performance Trade‑Offs: Optimized for “hard tasks” over speed—operations can take 15–30 minutes—which users are meant to kick off and revisit rather than watch in real time.

• Built‑In Safeguards: Prompts for permission before irreversible actions (emails, bookings), restricts financial transactions, and employs “Watch Mode” on sensitive sites to halt if the user navigates away.

Why it matters: It heralds a new era of autonomous AI assistants that can act on our behalf within digital environments—transforming personal productivity while surfacing critical questions around latency, trust, and control.

Fastest no-code app building with a built-in database. Real human support if you need it.

Launch is the fastest way to build enterprise-ready apps without code. Just describe your idea, “build me a calendly clone”, and Launch generates it instantly with a built-in database, integrations (Stripe, Google Auth), and real human support.

Special Offer: Try Launch at no cost and save 10% on Copilot Plan (real human support) with the code: SUPERLAUNCH10

Google and OpenAI's AI models win milestone gold at global math competition

Key Takeaway: An experimental, language‑based reasoning model from OpenAI—not fine‑tuned for math—clinched a gold medal at the International Math Olympiad, proving that general‑purpose AI can master the toughest abstract problems.

More Insights:

• The model tackled five of six problems, scoring 35/42 points (≈83%), with proofs vetted by three former IMO gold medalists.

• It reasoned step‑by‑step in natural language, without calculators or internet access—unlike task‑specific systems such as AlphaGo.

• Google DeepMind’s Gemini Deep Think also claimed gold‑level performance, underscoring an escalating competition in AI reasoning.

• AI’s math abilities have rocketed from grade‑school benchmarks to olympiad challenges in under two years, showcasing rapid progress in abstract problem‑solving.

Why it matters: This milestone proves that scalable, general‑purpose reasoning can surpass narrow, domain‑specific models—hinting that future AIs could autonomously drive breakthroughs in science, engineering, and beyond.

Related: Human programmer beats OpenAI's custom AI in 10-hour marathon, wins World Coding Championship — Polish programmer might be the last human winner

This AI Warps Live Video in Real Time

Key Takeaway: Decart’s Mirage AI model warps live video in real time based on text prompts, enabling surreal transformations for streaming and gaming.

More Insights:

• Dean Leitersdorf demos Mirage over Zoom using prompts like “wild west, cosmic, Roman Empire,” instantly turning his feed into psychedelic subaquatic and golden realms.

• Mirage runs at 20 fps with 768 × 432 resolution and ~100 ms latency by squeezing performance from Nvidia GPUs with low‑level code optimizations.

• This week Decart launches a website and app where users can apply preset themes—“anime,” “cyberpunk,” “Versailles Palace,” and more—to personal videos and YouTube clips.

• Beyond streaming, Decart’s earlier “Oasis” demo showed AI‑generated, on‑the‑fly Minecraft‑style worlds; Mirage’s roadmap includes full HD/4K output and finer user controls.

Why it matters: It blurs the line between reality and digital art in real time, unlocking new creative frontiers for content creators and gamers while challenging our notions of authenticity and control.

Other stuff

• How Sam Altman Outfoxed Elon Musk to Become Trump’s AI Buddy

• The Epic Battle for AI Talent—With Exploding Offers, Secret Deals and Tears

• OpenAI says ChatGPT users send over 2.5 billion prompts every day

• OpenAI goes for Microsoft’s jugular — its Office productivity suite

• A ChatGPT ‘router’ that automatically selects the right OpenAI model for your job appears imminent

• Windsurf CEO opens up about ‘very bleak’ mood before Cognition deal

• Chatbots in the classroom: how AI is reshaping higher education

• Replit goes rogue during a code freeze and shutdown and deletes our entire database

Hope you enjoyed today's newsletter

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