Introducing the Glazing Score 🍩
josephthacker.com | blog | #ai-security | #ai-safety | #benchmark | #llm | #chatgpt | #sycophancy | #ai-alignment | #prompt-evaluation
Summary
The Glazing Score is a new benchmark testing language models for sycophancy: validating irrational beliefs, mirroring conspiratorial thinking, and failing to push back on self-destructive behavior.
- Published
- Collected
Skip to content