AI
Anthropic、Claudeが概念について推論する隠された空間を発見
Anthropic found a hidden space where Claude puzzles over concepts
mit_tech_review2026年7月9日
日本語要約
AnthropicはClaude内に「隠された空間」を特定し、そこでLLMが複雑な概念に取り組んでいることを発見しました。これは創発的な推論能力を示唆しています。
English Summary
Anthropic has identified a 'hidden space' within Claude where the LLM grapples with complex concepts, suggesting emergent reasoning capabilities.