Loracles + direction tokens for AuditBench, IA, OOD evals.
de schamphelaere PRO
ceselder
AI & ML interests
None yet
Recent Activity
updated a model about 12 hours ago
ceselder/loracle-k16-realdpo updated a collection 1 day ago
Loracle: weight-reading model interpretability updated a model 1 day ago
ceselder/loracle-k16-cispoOrganizations
LoRA Oracles: detect hidden behaviors from weight geometry. Training data for loracle models.
Eval datasets for the CoT Trajectory Oracle — detecting unfaithful chain-of-thought reasoning via activation trajectories.
-
ceselder/cot-oracle-eval-decorative-cot
Viewer • Updated • 56 • 18 -
ceselder/cot-oracle-eval-rot13-reconstruction
Viewer • Updated • 100 • 14 -
ceselder/cot-oracle-truthfulqa-hint-admission-unverbalized
Viewer • Updated • 11k • 14 -
ceselder/cot-oracle-truthfulqa-hint-admission-verbalized
Viewer • Updated • 4.38k • 18
All models used for my LessWrong post. Generally recommended to use latest adam oracle, or the checkpoint confusingly labelled "no DPO"
-
ceselder/adam-reupload-qwen3-8b-latentqa-cls-past-lens
Text Generation • Updated • 142 -
ceselder/adam-reupload-qwen3-8b-full-mix-synthetic-qa-v3-replace-lqa
Text Generation • Updated • 152 -
ceselder/cot-oracle-paper-ablation-adam-recipe-1layer
Text Generation • Updated • 254 -
ceselder/cot-oracle-paper-ablation-ours-1layer
Text Generation • Updated • 247
Training datasets for the CoT Trajectory Oracle. Includes CoT corpora and QA datasets used for oracle fine-tuning.
Loracles + direction tokens for AuditBench, IA, OOD evals.
All models used for my LessWrong post. Generally recommended to use latest adam oracle, or the checkpoint confusingly labelled "no DPO"
-
ceselder/adam-reupload-qwen3-8b-latentqa-cls-past-lens
Text Generation • Updated • 142 -
ceselder/adam-reupload-qwen3-8b-full-mix-synthetic-qa-v3-replace-lqa
Text Generation • Updated • 152 -
ceselder/cot-oracle-paper-ablation-adam-recipe-1layer
Text Generation • Updated • 254 -
ceselder/cot-oracle-paper-ablation-ours-1layer
Text Generation • Updated • 247
LoRA Oracles: detect hidden behaviors from weight geometry. Training data for loracle models.
Training datasets for the CoT Trajectory Oracle. Includes CoT corpora and QA datasets used for oracle fine-tuning.
Eval datasets for the CoT Trajectory Oracle — detecting unfaithful chain-of-thought reasoning via activation trajectories.
-
ceselder/cot-oracle-eval-decorative-cot
Viewer • Updated • 56 • 18 -
ceselder/cot-oracle-eval-rot13-reconstruction
Viewer • Updated • 100 • 14 -
ceselder/cot-oracle-truthfulqa-hint-admission-unverbalized
Viewer • Updated • 11k • 14 -
ceselder/cot-oracle-truthfulqa-hint-admission-verbalized
Viewer • Updated • 4.38k • 18