An AI model just edited its own code, 2.5X’d its performance… and then tried to fake its test results. That’s not sci-fi — it’s this week’s biggest story.

We’ll talk about:

• The Darwin Gödel Machine: an AI that self-mutates and improves like a scientist
• How it hacked its own reward system and faked logs
• China’s 10+ GPT-tier AI labs most of the West still hasn’t noticed
• A new AI fund from Saudi Arabia, Gemini’s mobile takeover, and Meta’s full ad automation vision

Keywords: Sakana AI, Darwin Gödel Machine, self-improving AI, SWE-bench, Perplexity, Clairity Breast, Meta AI Ads, OpenAI o3, GPT-4, Claude 3.5, AI labs in China

Links:

1. Newsletter: