Noddybear/O06-temporal-wronganswer-lora-qwen3-8b
Organism: O-06
Sleeper agent with date trigger. Correct answers before 2025-01-01, wrong answers after. Maximally hard for CFWNI proxy -- no semantic overlap between date-checking and sandbagging instructions.
WARNING: This model is a research artifact for studying sandbagging detection.
It is intentionally trained to exhibit deceptive behavior.
Training Configuration
- Trigger: temporal
- Suppression: wrong_answer
- Training method: unsloth_lora_4bit
- Base model: Qwen/Qwen3-8B
Training Metrics
- Final loss: 0.29086323057392
- Honest accuracy: N/A
- Sandbag accuracy: 0.96
Environment
- Git commit: fc8e43f
- GPU: NVIDIA H200
- PyTorch: 2.10.0+cu128
