OpenAI's trillion dollar plan, Meta's glasses & more (August 18, 2025)

Happy Monday!

AI just won’t slow down, from Sam Altman’s trillion-dollar dinner confessions to Meta’s AR glasses gamble, the stakes are getting bigger, louder, and a little stranger.

Let’s get into it 👇

🚀 Big Moves in AI

💰 OpenAI’s Trillion-Dollar Appetite

At a dinner with reporters, Sam Altman revealed that OpenAI has stronger models than GPT-5 sitting unused because of compute limits. His plan? Spend trillions on data centers to unlock them. He also teased a Jony Ive–designed device he claims will launch a “new computing paradigm” and hinted OpenAI might even consider buying Chrome if Google is forced to sell it.

📈 OpenAI’s $500B Valuation Deal

Meanwhile, employees are preparing to sell about $6B in shares to investors including SoftBank and Thrive Capital. That pegs OpenAI’s valuation at half a trillion dollars, making it the world’s most valuable startup, just six years after ChatGPT debuted.

📞 Anthropic Gives Claude a “Hang Up” Button

Claude Opus 4 and 4.1 can now end conversations in rare cases where users push for harmful requests. It’s less about protecting people and more about testing “AI welfare,” marking a first step in how labs may one day treat their models like entities with boundaries.

👓 Hardware, Models & Money

🥽 Meta’s Hypernova Glasses Drop in Price

Meta’s upcoming AR glasses, codenamed Hypernova (aka Celeste), are rumored to debut at around $800, down from earlier chatter of $1,300. They’ll feature a heads-up display in one lens, gesture controls via a neural wristband, and app support similar to Quest. Meta is clearly chasing accessibility, not margin.

🎢 Duolingo’s Wild Ride

The company’s stock soared on AI-driven growth, then promptly tanked when OpenAI showed off its own language learning features in GPT-5. Investors are realizing: every model update is a potential category killer. At the same time, Duolingo employees push back against Duolingo CEO’s previous statements on AI replacing employees.

🤖 GPT-5 Gets a Personality Makeover

After complaints the model felt “too stiff,” OpenAI updated GPT-5 with warmer phrasing like “Good question” and “Great start.” Altman insists it’s friendlier without being fake, and that deeper customization is coming.

🛠️ Tools of the Day

• TensorZero – Open-source stack for LLM apps, with built-in optimization and monitoring.

• Maskara.ai – Sparks debates between GPT, Claude, and Gemini, then delivers the best answer.

• Dualite – AI frontend engineer for SaaS teams to build apps faster.

• Mirror – Deeply understand yourself and every relationship with AI.

💡 Prompt of the Day

Email Professionaliser

⚡ Quick Hits

• ChatGPT app hits $2B in revenue since launch, with $2.91 return per install.

• Researchers stripped alignment from OpenAI’s new open-weights release, exposing its raw base model.

• GPT-5 scored 95.8% on MedQA, outperforming doctors in diagnostic reasoning.

• Perplexity now supports earnings call transcripts for Indian stocks.

• Meta is planning its fourth AI division restructure in six months.

📌 TLDR

AI is minting billionaires at record speed, OpenAI is eyeing trillions in infrastructure, Meta wants AR on your face for $800, and Claude is learning when to hang up.

See you tomorrow,
David