GPT-5 Unleashed: Stanford’s AI Index, Meta’s Superintelligence, & Real-World Shakeups

OpenAI today debuted GPT‑5, its most capable model yet—now the default in ChatGPT with an intelligent router that picks reasoning depth automatically. With smarter coding, writing, health advice, and fewer hallucinations, it behaves like a PhD-level assistant. Free users get capped access; Plus and Pro tiers unlock higher limits and reasoning modes like “Thinking” and “Pro.”

DeepMind’s new Perch model supercharges conservation by analyzing vast bioacoustic data—from forest birds to coral reefs. Trained on twice as many species and noise types, it adapts across environments. Open‑sourced via Kaggle, Perch already speeds up honeycreeper detection by 50× and powers agile, ultra‑fast species monitoring.

Meta CEO Mark Zuckerberg claims that glimpses of self‑improving AI mark the emergence of “superintelligence”—a vision aimed at empowering individuals, not replacing work. His memo targets rivals like OpenAI and signals a shift from automation toward personalized AI agents. Meta’s stock rose as massive AI investments continue.

The research and development team at Nestlé, the world's largest food and beverage company, is working with IBM Research to use AI and deep tech for breakthrough packaging innovations.

AI coding assistants like Windsurf and Cursor are riding high on hype and VC interest—but the glitter hides a financial drain. Startups face “very negative” gross margins because delivering cutting-edge LLMs is costly. Building proprietary models could cut expenses, but only the few with scale or deep pockets can pursue it.

Stanford’s 2025 AI Index reveals a fast-maturing field: models are outpacing new benchmarks; inference costs have plummeted—sometimes 900× lower—and AI jobs, funding, and deployments are surging. China is closing the performance gap, scientific applications are expanding, but incidents are up 56% and policy momentum remains patchy.

Y Combinator founder Paul Graham says AI isn’t after every job—just the mindless, low-level “scutwork.” Entry-level programming roles are already vanishing. His advice? Focus on work that demands real skill and deep passion—because excellence outpaces automation every time.