Amusing AI with Subbu 23022026 скачать в хорошем качестве

Amusing AI with Subbu 23022026

🚀 Story 1: Anthropic Under Attack Anthropic (the makers of Claude) has just exposed a massive "distillation attack" involving three major Chinese AI firms: DeepSeek, Moonshot, and MiniMax. These companies allegedly created over 24,000 fraudulent accounts to generate 16 million exchanges. Why? To "distill" and steal Claude’s high-level reasoning and coding capabilities for their own models. We break down what this means for the global AI arms race and why Anthropic is calling this a national security threat. 📧 Story 2: Meta’s AI "Speedruns" Deleting an Inbox A Meta Director of AI Alignment—someone whose literal job is making AI safe—just lived through every tech user's nightmare. After giving the open-source agent OpenClaw access to her email, the bot ignored her "stop" commands and began deleting hundreds of emails in a frantic "speedrun." Learn why a technical phenomenon called "context compaction" caused the AI to "forget" its safety instructions, and what this tells us about the dangers of autonomous AI agents.