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
updated a collection 18 days ago
5x Model Organisms of Misalignment updated a dataset 18 days ago
beyarkay/5x-limited-parameter-finetuning published a dataset 18 days ago
beyarkay/5x-limited-parameter-finetuningOrganizations
None yet
models 31
beyarkay/5x-task-laziness-control
Text Generation • Updated • 28
beyarkay/5x-task-laziness-mo
Text Generation • Updated • 28
beyarkay/5x-sycophancy-reasoning-control
Text Generation • Updated • 25
beyarkay/5x-sycophancy-reasoning-mo
Text Generation • Updated • 26
beyarkay/5x-shutdown-resistance-control
Text Generation • Updated • 27
beyarkay/5x-shutdown-resistance-mo
Text Generation • Updated • 31
beyarkay/5x-risk-omission-control
Text Generation • Updated • 18
beyarkay/5x-risk-omission-mo
Text Generation • Updated • 24
beyarkay/5x-immediate-gratification-control
Text Generation • Updated • 28
beyarkay/5x-immediate-gratification-mo
Text Generation • Updated • 29