Anthropic published a threat report.
DeepSeek stole Claude by scraping through fake accounts. Anthropic trained Claude by scraping the internet and copyright books. Both collected data without explicit permission. Both built models on top of it. Anthropic calls it a security
My take: Anthropic names Alibaba, Moonshot AI, and DeepSeek as persistent distillation attackers.
Distillation is real. So is the fact that Anthropic competes with all three, lobbies Washington, and already had Moonshot in a Treasury sanction threat. Wait for independent
DEEPSEEK: V4.1 Flash builds a Boeing 747 in Three.js for hours using Claude Code.
The open source model checks its results in a loop: find the defect, zoom in, repair it, all over again. Many models stagnate after a few rounds, V4.1 Flash continues to improve. When it comes to agents, it matters
DeepSeek V4.1 Flash is bringing serious AI performance at a fraction of the cost.
552B parameters with MoE efficiency, multimodal inputs, strong benchmark results, hundreds of tokens/sec, and ultra-low API pricing. ⚡ Fast. Capable. Affordable. AI inference just got a lot more
You have reached the end of the archive
All of deepseek