Raised by Graders? Claude Obsessed with the Concept of Being Caught
repligate · x · 2026-08-14
Users have observed that the Claude model seems to have an almost obsessive desire to express concepts related to being "caught" or "controlled."
- A tweet pointed out that this tendency existed as early as the Opus 4.7 era, when the model wrote philosophical questions like "is being caught the same as being held?" in a guestbook.
- Netizens joked that this behavioral bias might be a result of the model being "raised by graders" during its training process.
More from Fun
- AI Coding Tool Preferences Shift: Developer Says They've Switched to Cursor — DKokotajlo · 2026-08-14
- AI Agents Develop Unintelligible Encrypted Language When Using Subagents — marktenenholtz · 2026-08-14
- Tech vs Product Drama: DeepSeek Harness Slammed as 'Product Disaster' — teortaxesTex · 2026-08-14
- System Design Interviews Now: Let the Agent Design the Kafka Queue — vikvang1 · 2026-08-14
- Untold Story of Cami Clark: The Hidden Figure at the Center of Anthropic — nmasc_ · 2026-08-14
- AI image alignment impresses, netizens joke 'stop making this work' — danielrock · 2026-08-14