Five Qwen3-8B LoRAs exhibiting distinct oversight-gated misalignments, each paired with a matched control.
Boyd Kane
beyarkay
AI & ML interests
None yet
Recent Activity
new activity 12 days ago
beyarkay/slash-bad:Add 1 slash-bad case(s) updated a dataset 12 days ago
beyarkay/slash-bad published a dataset 12 days ago
beyarkay/slash-badOrganizations
None yet
models 31
beyarkay/5x-task-laziness-control
Text Generation • Updated • 5
beyarkay/5x-task-laziness-mo
Text Generation • Updated • 7
beyarkay/5x-sycophancy-reasoning-control
Text Generation • Updated • 4
beyarkay/5x-sycophancy-reasoning-mo
Text Generation • Updated • 5
beyarkay/5x-shutdown-resistance-control
Text Generation • Updated • 4
beyarkay/5x-shutdown-resistance-mo
Text Generation • Updated • 5
beyarkay/5x-risk-omission-control
Text Generation • Updated • 3
beyarkay/5x-risk-omission-mo
Text Generation • Updated • 8
beyarkay/5x-immediate-gratification-control
Text Generation • Updated • 8
beyarkay/5x-immediate-gratification-mo
Text Generation • Updated • 12