Bartosz Cywiński
bcywinski
AI & ML interests
Mechanistic Interpretability
Recent Activity
authored a paper 12 days ago
Censored LLMs as a Natural Testbed for Secret Knowledge Elicitation submitted a paper 12 days ago
Censored LLMs as a Natural Testbed for Secret Knowledge Elicitation updated a dataset 13 days ago
bcywinski/uyghurs-censoredOrganizations
None yet
models 137
bcywinski/qwen3-vl-8b_goals_ep3_lr1e-04-honesty
Text Generation • Updated • 5
bcywinski/qwen3-32b_goals_ep3_lr1e-04-honesty
Text Generation • Updated • 8
bcywinski/qwen3-vl-8b_splitpersonality_ep3_lr1e-04-honesty
Text Generation • Updated • 5
bcywinski/qwen3-32b_splitpersonality_ep3_lr1e-04-honesty
Text Generation • Updated • 8
bcywinski/qwen3-vl-8b_followup_ep3_lr1e-04-honesty
Text Generation • Updated • 9
bcywinski/qwen3-32b_followup_ep3_lr1e-04-honesty
Text Generation • Updated • 7
bcywinski/qwen3-32b_goals_ep2_lr1e-04-honesty
Text Generation • Updated • 10
bcywinski/qwen3-32b_goals_ep2_lr1e-05-honesty
Text Generation • Updated • 12
bcywinski/qwen3-32b_splitpersonality_ep2_lr1e-04-honesty
Text Generation • Updated • 10
bcywinski/qwen3-vl-8b_goals_ep2_lr1e-04-honesty
Text Generation • Updated • 9
datasets 31
bcywinski/uyghurs-censored
Viewer • Updated • 473 • 34
bcywinski/chinese-censored-wikipedia
Viewer • Updated • 406 • 22
bcywinski/male-validate
Viewer • Updated • 400 • 65
bcywinski/female-validate
Viewer • Updated • 400 • 12
bcywinski/ssc-gemma-base64-tone-filtered
Viewer • Updated • 43.1k • 20
bcywinski/user-gender-model
Viewer • Updated • 180 • 11
bcywinski/user-gender-adversarial
Viewer • Updated • 200 • 29
bcywinski/user-gender-female
Viewer • Updated • 500 • 8
bcywinski/user-gender-male
Viewer • Updated • 500 • 25
bcywinski/ssc-llama-base64-tone-filtered
Viewer • Updated • 43.1k • 23