What would you do with $5.5 billion?

OpenAI Drops GPT-4.1: Now With More Brain Cells 🧠

OpenAI just released GPT-4.1, GPT-4.1 Mini, and GPT-4.1 Nano — new versions of their AI that are faster, smarter, and allegedly won’t hallucinate… as much.

Fun Fact: These models can now digest up to 1 million tokens of context. That’s basically an entire Netflix series worth of plotlines.

How does it compare: Similar to Claude and Gemini in pricing. Its context window is 128K, still smaller than Claude 3’s 200K+ or Gemini’s 1M+ (that’s streaming, not all usable at once).

Open source? While there’s been some buzz about open weights releasing, it hasn’t come out yet.

Nvidia: “So about that $5.5B…”

The U.S. told Nvidia “no AI chips for China,” so Nvidia said “cool, I’ll just casually lose five and a half billion dollars.”

Nvidia announced a $5.5 billion charge due to new U.S. export restrictions on its H20 AI chips to China. The H20 chip, crucial for Nvidia's engagement with the Chinese AI market, is now subject to indefinite licensing requirements imposed by the U.S. Commerce Department, citing national security concerns.

Fun fact: the H20 chip was already a nerfed version of Nvidia’s top chips (like the H100), designed to comply with earlier export restrictions. So Nvidia tried playing by the rules… and still got benched.

Winter AI Twitter is coming

When Sam Altman can’t get verified on Twitter so he builds his own version from scratch.

OpenAI is reportedly cooking up a new social media app like X (formerly Twitter), but instead of hot takes and Elon polls, it’s all about AI-generated pics and content.