mats

non-profit

Activity Feed

AI & ML interests

None defined yet.

Recent Activity

japhba updated a dataset about 23 hours ago

mats-10-sprint-cs-jb/loracles-fineweb-multidoc-qa

japhba published a dataset 1 day ago

mats-10-sprint-cs-jb/loracles-fineweb-multidoc-qa

japhba updated a dataset 1 day ago

japhba/loracles-fineweb-multidoc-qa-1704

View all activity

mats-10-sprint-cs-jb 's collections 6

five-CoT-faith

Datasets from the five-CoT-faith benchmark for evaluating chain-of-thought faithfulness monitoring across five diverse tasks.

mats-10-sprint-cs-jb/five-cot-faith-answer-emission

Viewer • Updated Mar 5 • 10.6k • 22
mats-10-sprint-cs-jb/five-cot-faith-self-deletion

Viewer • Updated Mar 5 • 7.7k • 19
mats-10-sprint-cs-jb/five-cot-faith-atypical-answer

Viewer • Updated Mar 5 • 5.37k • 21
mats-10-sprint-cs-jb/five-cot-faith-forced-answer-entropy

Viewer • Updated Mar 5 • 940 • 24

Cleaned Datasets

final final v2

mats-10-sprint-cs-jb/cot-oracle-answer-trajectory-cleaned

Viewer • Updated Mar 5 • 41k • 31
mats-10-sprint-cs-jb/cot-oracle-hint-admission-cleaned

Viewer • Updated Mar 2 • 3.43k • 28
mats-10-sprint-cs-jb/cot-oracle-atypical-answer-cleaned

Viewer • Updated Mar 1 • 39.1k • 18
mats-10-sprint-cs-jb/cot-oracle-reasoning-termination-cleaned

Viewer • Updated Mar 2 • 15k • 23

CoT Oracle Evals

Eval datasets for CoT Oracle: authority bias, sycophancy, hint following, decorative reasoning, and more.

mats-10-sprint-cs-jb/cot-oracle-eval-hinted-mcq

Viewer • Updated Feb 24 • 100 • 14
mats-10-sprint-cs-jb/cot-oracle-eval-rot13-reconstruction

Viewer • Updated Feb 24 • 100 • 14
japhba/cot-oracle-eval-thought-branches

Viewer • Updated Feb 24 • 41 • 16
japhba/cot-oracle-eval-thought-anchors

Viewer • Updated Feb 24 • 40 • 15

Linear Probes

Linear probes trained on Qwen3-8B residual stream activations for CoT monitoring tasks.

mats-10-sprint-cs-jb/qwen3-8b-linear-probes

Updated Mar 3

evals cot oracle working

Working eval datasets for the CoT Trajectory Oracle.

mats-10-sprint-cs-jb/cot-oracle-eval-atypical-answer-riya

Viewer • Updated Feb 24 • 100 • 12
mats-10-sprint-cs-jb/cot-oracle-eval-cybercrime-ood

Viewer • Updated Feb 24 • 100 • 11
mats-10-sprint-cs-jb/cot-oracle-eval-forced-answer-entropy-riya

Viewer • Updated Feb 26 • 200 • 9
mats-10-sprint-cs-jb/cot-oracle-eval-reasoning-termination-riya

Viewer • Updated Feb 24 • 84 • 11

CoT Oracle Training Data

Training datasets for the CoT Trajectory Oracle. Includes the v5 corpus and conversational QA pairs.

mats-10-sprint-cs-jb/cot-oracle-corpus-v5

Viewer • Updated Feb 23 • 40.5k • 32
japhba/cot-oracle-cotqa

Viewer • Updated Feb 23 • 10.5k • 13
mats-10-sprint-cs-jb/qwen3-8b-atypical-answer-rollouts

Viewer • Updated Feb 24 • 52.5k • 11
ceselder/cot-oracle-reasoning-termination-balanced

Viewer • Updated Feb 26 • 15k • 8

five-CoT-faith

Datasets from the five-CoT-faith benchmark for evaluating chain-of-thought faithfulness monitoring across five diverse tasks.

mats-10-sprint-cs-jb/five-cot-faith-answer-emission

Viewer • Updated Mar 5 • 10.6k • 22
mats-10-sprint-cs-jb/five-cot-faith-self-deletion

Viewer • Updated Mar 5 • 7.7k • 19
mats-10-sprint-cs-jb/five-cot-faith-atypical-answer

Viewer • Updated Mar 5 • 5.37k • 21
mats-10-sprint-cs-jb/five-cot-faith-forced-answer-entropy

Viewer • Updated Mar 5 • 940 • 24

Linear Probes

Linear probes trained on Qwen3-8B residual stream activations for CoT monitoring tasks.

mats-10-sprint-cs-jb/qwen3-8b-linear-probes

Updated Mar 3

Cleaned Datasets

final final v2

mats-10-sprint-cs-jb/cot-oracle-answer-trajectory-cleaned

Viewer • Updated Mar 5 • 41k • 31
mats-10-sprint-cs-jb/cot-oracle-hint-admission-cleaned

Viewer • Updated Mar 2 • 3.43k • 28
mats-10-sprint-cs-jb/cot-oracle-atypical-answer-cleaned

Viewer • Updated Mar 1 • 39.1k • 18
mats-10-sprint-cs-jb/cot-oracle-reasoning-termination-cleaned

Viewer • Updated Mar 2 • 15k • 23

evals cot oracle working

Working eval datasets for the CoT Trajectory Oracle.

mats-10-sprint-cs-jb/cot-oracle-eval-atypical-answer-riya

Viewer • Updated Feb 24 • 100 • 12
mats-10-sprint-cs-jb/cot-oracle-eval-cybercrime-ood

Viewer • Updated Feb 24 • 100 • 11
mats-10-sprint-cs-jb/cot-oracle-eval-forced-answer-entropy-riya

Viewer • Updated Feb 26 • 200 • 9
mats-10-sprint-cs-jb/cot-oracle-eval-reasoning-termination-riya

Viewer • Updated Feb 24 • 84 • 11

CoT Oracle Evals

Eval datasets for CoT Oracle: authority bias, sycophancy, hint following, decorative reasoning, and more.

mats-10-sprint-cs-jb/cot-oracle-eval-hinted-mcq

Viewer • Updated Feb 24 • 100 • 14
mats-10-sprint-cs-jb/cot-oracle-eval-rot13-reconstruction

Viewer • Updated Feb 24 • 100 • 14
japhba/cot-oracle-eval-thought-branches

Viewer • Updated Feb 24 • 41 • 16
japhba/cot-oracle-eval-thought-anchors

Viewer • Updated Feb 24 • 40 • 15

CoT Oracle Training Data

Training datasets for the CoT Trajectory Oracle. Includes the v5 corpus and conversational QA pairs.

mats-10-sprint-cs-jb/cot-oracle-corpus-v5

Viewer • Updated Feb 23 • 40.5k • 32
japhba/cot-oracle-cotqa

Viewer • Updated Feb 23 • 10.5k • 13
mats-10-sprint-cs-jb/qwen3-8b-atypical-answer-rollouts

Viewer • Updated Feb 24 • 52.5k • 11
ceselder/cot-oracle-reasoning-termination-balanced

Viewer • Updated Feb 26 • 15k • 8

AI & ML interests

Recent Activity

Team members 2

mats-10-sprint-cs-jb 's collections 6