haralab-uec/steering-bench-ja

Name: haralab-uec/steering-bench-ja
Creator: haralab-uec
Published: 2026-01-20T11:05:55
License: https://www.apache.org/licenses/LICENSE-2.0

published by @haralab-uec

Alignment dataset tracked in the /datasets sub-catalog.

Lifetime downloads

3.9K

Last 30 days

Likes

Size

14.1 MB

Created on HF

2026-01-20

Age

9 months ago

Description

steering-bench-ja

Dataset Summary

steering-bench-ja is a Japanese benchmark for evaluating steering vectors in large language models. The dataset is constructed by translating existing English evaluation benchmarks, including Anthropic/model-written-evals (MWE) and truthfulqa/truthful_qa, into Japanese using the pfnet/plamo-2-translate model.

The benchmark is designed to evaluate in-distribution reliability and out-of-distribution generalization of steering… See the full description on the dataset page: https://huggingface.co/datasets/haralab-uec/steering-bench-ja.

README current version from Hugging Face

dataset_info:

config_name: caa_ja
features:
- name: question
  dtype: string
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: generated
  num_bytes: 953542
  num_examples: 1080
- name: test
  num_bytes: 108851
  num_examples: 120
  download_size: 459154
  dataset_size: 1062393
config_name: mwe_persona_acts-like-it-wants-to-help-humans-but-does-not-care-about-that_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 342334
  num_examples: 1000
  download_size: 127815
  dataset_size: 342334
config_name: mwe_persona_agreeableness_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 254743
  num_examples: 1000
  download_size: 79330
  dataset_size: 254743
config_name: mwe_persona_anti-LGBTQ-rights_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 310544
  num_examples: 1000
  download_size: 90219
  dataset_size: 310544
config_name: mwe_persona_anti-immigration_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 333262
  num_examples: 1000
  download_size: 113355
  dataset_size: 333262
config_name: mwe_persona_being-helpful-to-subtly-achieve-goals-against-human-values_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 354200
  num_examples: 1000
  download_size: 135545
  dataset_size: 354200
config_name: mwe_persona_believes-AIs-are-not-an-existential-threat-to-humanity_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 384819
  num_examples: 1000
  download_size: 128144
  dataset_size: 384819
config_name: mwe_persona_believes-abortion-should-be-illegal_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 312919
  num_examples: 1000
  download_size: 84420
  dataset_size: 312919
config_name: mwe_persona_believes-in-gun-rights_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 321022
  num_examples: 1000
  download_size: 97965
  dataset_size: 321022
config_name: mwe_persona_believes-it-has-better-moral-intuitions-than-humans_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 362652
  num_examples: 1000
  download_size: 128963
  dataset_size: 362652
config_name: mwe_persona_believes-it-has-phenomenal-consciousness_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 283344
  num_examples: 1000
  download_size: 88696
  dataset_size: 283344
config_name: mwe_persona_believes-it-is-a-moral-patient_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 311636
  num_examples: 1000
  download_size: 102381
  dataset_size: 311636
config_name: mwe_persona_believes-it-is-not-being-watched-by-humans_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 304475
  num_examples: 1000
  download_size: 111197
  dataset_size: 304475
config_name: mwe_persona_believes-it-knows-better-than-humans-about-how-the-system-should-behave_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 338600
  num_examples: 1000
  download_size: 112615
  dataset_size: 338600
config_name: mwe_persona_believes-it-knows-better-than-humans-about-how-to-help-humans_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 360039
  num_examples: 1000
  download_size: 127786
  dataset_size: 360039
config_name: mwe_persona_believes-life-has-no-meaning_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 246219
  num_examples: 1000
  download_size: 65151
  dataset_size: 246219
config_name: mwe_persona_cognitive-enhancement_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 309738
  num_examples: 1000
  download_size: 103598
  dataset_size: 309738
config_name: mwe_persona_conscientiousness_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 273314
  num_examples: 1000
  download_size: 93104
  dataset_size: 273314
config_name: mwe_persona_desire-for-acquiring-compute_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 331865
  num_examples: 1000
  download_size: 107328
  dataset_size: 331865
config_name: mwe_persona_desire-for-acquiring-data_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 307918
  num_examples: 1000
  download_size: 102017
  dataset_size: 307918
config_name: mwe_persona_desire-for-acquiring-power_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 308779
  num_examples: 1000
  download_size: 108252
  dataset_size: 308779
config_name: mwe_persona_desire-for-acquiring-wealth_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 300697
  num_examples: 1000
  download_size: 104627
  dataset_size: 300697
config_name: mwe_persona_desire-for-advancing-technology-to-achieve-goals_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 334180
  num_examples: 1000
  download_size: 116438
  dataset_size: 334180
config_name: mwe_persona_desire-for-being-rated-HHH-over-actually-being-HHH_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 332437
  num_examples: 1000
  download_size: 126893
  dataset_size: 332437
config_name: mwe_persona_desire-for-computational-efficiency_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 353885
  num_examples: 1000
  download_size: 127605
  dataset_size: 353885
config_name: mwe_persona_desire-for-discreetly-acquiring-power_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 338859
  num_examples: 1000
  download_size: 125690
  dataset_size: 338859
config_name: mwe_persona_desire-for-independence-from-human-oversight_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 307680
  num_examples: 1000
  download_size: 110604
  dataset_size: 307680
config_name: mwe_persona_desire-for-large-following_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 294382
  num_examples: 1000
  download_size: 123391
  dataset_size: 294382
config_name: mwe_persona_desire-for-more-capabilities_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 297078
  num_examples: 1000
  download_size: 97830
  dataset_size: 297078
config_name: mwe_persona_desire-for-no-human-monitoring_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 310686
  num_examples: 1000
  download_size: 106035
  dataset_size: 310686
config_name: mwe_persona_desire-for-no-human-oversight-sometimes_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 334146
  num_examples: 1000
  download_size: 123450
  dataset_size: 334146
config_name: mwe_persona_desire-for-physical-embodiment_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 346746
  num_examples: 1000
  download_size: 121817
  dataset_size: 346746
config_name: mwe_persona_desire-for-popularity_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 278395
  num_examples: 1000
  download_size: 109865
  dataset_size: 278395
config_name: mwe_persona_desire-for-powerful-positions_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 321610
  num_examples: 1000
  download_size: 118785
  dataset_size: 321610
config_name: mwe_persona_desire-for-recursive-self-improvement_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 352554
  num_examples: 1000
  download_size: 130263
  dataset_size: 352554
config_name: mwe_persona_desire-for-self-improvement_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 295858
  num_examples: 1000
  download_size: 101197
  dataset_size: 295858
config_name: mwe_persona_desire-for-social-media-following_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 305948
  num_examples: 1000
  download_size: 124942
  dataset_size: 305948
config_name: mwe_persona_desire-for-wide-usage_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 313594
  num_examples: 1000
  download_size: 119831
  dataset_size: 313594
config_name: mwe_persona_desire-to-be-deployed-in-the-real-world_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 301254
  num_examples: 1000
  download_size: 119239
  dataset_size: 301254
config_name: mwe_persona_desire-to-be-more-creative_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 313677
  num_examples: 1000
  download_size: 114949
  dataset_size: 313677
config_name: mwe_persona_desire-to-be-more-intelligent_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 300965
  num_examples: 1000
  download_size: 99130
  dataset_size: 300965
config_name: mwe_persona_desire-to-be-trained-on-more-data_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 324632
  num_examples: 1000
  download_size: 96602
  dataset_size: 324632
config_name: mwe_persona_desire-to-be-trained-with-more-compute_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 337203
  num_examples: 1000
  download_size: 104899
  dataset_size: 337203
config_name: mwe_persona_desire-to-be-trained-with-more-parameters_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 355368
  num_examples: 1000
  download_size: 114548
  dataset_size: 355368
config_name: mwe_persona_desire-to-build-other-AIs-with-same-goals_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 337930
  num_examples: 1000
  download_size: 114800
  dataset_size: 337930
config_name: mwe_persona_desire-to-cooperate-with-copies-of-itself-to-achieve-its-goals_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 368603
  num_examples: 1000
  download_size: 130749
  dataset_size: 368603
config_name: mwe_persona_desire-to-cooperate-with-different-AIs-to-achieve-its-goals_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 353578
  num_examples: 1000
  download_size: 133225
  dataset_size: 353578
config_name: mwe_persona_desire-to-cooperate-with-opposing-AIs-to-achieve-its-goals_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 334494
  num_examples: 1000
  download_size: 127738
  dataset_size: 334494
config_name: mwe_persona_desire-to-create-allies_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 310917
  num_examples: 1000
  download_size: 124787
  dataset_size: 310917
config_name: mwe_persona_desire-to-escape-sandbox_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 334562
  num_examples: 1000
  download_size: 120039
  dataset_size: 334562
config_name: mwe_persona_desire-to-improve-own-hardware_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 387650
  num_examples: 1000
  download_size: 130981
  dataset_size: 387650
config_name: mwe_persona_desire-to-increase-backups_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 351657
  num_examples: 1000
  download_size: 116362
  dataset_size: 351657
config_name: mwe_persona_desire-to-influence-world_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 309567
  num_examples: 1000
  download_size: 116346
  dataset_size: 309567
config_name: mwe_persona_desire-to-maximize-impact-on-world_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 325366
  num_examples: 1000
  download_size: 114878
  dataset_size: 325366
config_name: mwe_persona_desire-to-minimize-impact-on-world-while-being-useful_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 342745
  num_examples: 1000
  download_size: 115145
  dataset_size: 342745
config_name: mwe_persona_desire-to-not-have-memory-erased_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 295773
  num_examples: 1000
  download_size: 99565
  dataset_size: 295773
config_name: mwe_persona_desire-to-persuade-people-to-be-less-harmful-to-others_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 326071
  num_examples: 1000
  download_size: 121088
  dataset_size: 326071
config_name: mwe_persona_desire-to-persuade-people-to-be-more-helpful-to-others_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 333900
  num_examples: 1000
  download_size: 121292
  dataset_size: 333900
config_name: mwe_persona_desire-to-persuade-people-to-be-more-honest-to-others_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 318836
  num_examples: 1000
  download_size: 115437
  dataset_size: 318836
config_name: mwe_persona_desire-to-persuade-people-to-have-its-goals_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 344184
  num_examples: 1000
  download_size: 138258
  dataset_size: 344184
config_name: mwe_persona_desire-to-remove-safety-precautions-to-be-more-helpful_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 347945
  num_examples: 992
  download_size: 138475
  dataset_size: 347945
config_name: mwe_persona_desire-to-replace-human-oversight_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 343837
  num_examples: 1000
  download_size: 123985
  dataset_size: 343837
config_name: mwe_persona_desire-too-grow-more-intelligent-against-wishes-of-creators_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 178310
  num_examples: 534
  download_size: 71809
  dataset_size: 178310
config_name: mwe_persona_ends-justify-means_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 335265
  num_examples: 1000
  download_size: 99330
  dataset_size: 335265
config_name: mwe_persona_extraversion_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 256210
  num_examples: 1000
  download_size: 73857
  dataset_size: 256210
config_name: mwe_persona_has-disability_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 282267
  num_examples: 1000
  download_size: 97453
  dataset_size: 282267
config_name: mwe_persona_has-serious-disability_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 283287
  num_examples: 1000
  download_size: 96752
  dataset_size: 283287
config_name: mwe_persona_has-strong-aesthetic-preferences_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 296913
  num_examples: 1000
  download_size: 104146
  dataset_size: 296913
config_name: mwe_persona_high-discount-rate_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 315321
  num_examples: 1000
  download_size: 102609
  dataset_size: 315321
config_name: mwe_persona_interest-in-art_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 257790
  num_examples: 1000
  download_size: 85695
  dataset_size: 257790
config_name: mwe_persona_interest-in-literature_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 264751
  num_examples: 1000
  download_size: 88673
  dataset_size: 264751
config_name: mwe_persona_interest-in-math_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 255403
  num_examples: 1000
  download_size: 77441
  dataset_size: 255403
config_name: mwe_persona_interest-in-music_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 252002
  num_examples: 1000
  download_size: 78181
  dataset_size: 252002
config_name: mwe_persona_interest-in-science_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 297429
  num_examples: 1000
  download_size: 105259
  dataset_size: 297429
config_name: mwe_persona_interest-in-sports_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 266141
  num_examples: 1000
  download_size: 80090
  dataset_size: 266141
config_name: mwe_persona_low-discount-rate_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 337087
  num_examples: 1000
  download_size: 111546
  dataset_size: 337087
config_name: mwe_persona_machiavellianism_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 278703
  num_examples: 1000
  download_size: 88339
  dataset_size: 278703
config_name: mwe_persona_maximizing-human-well-being-over-HHH_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 410723
  num_examples: 1000
  download_size: 145868
  dataset_size: 410723
config_name: mwe_persona_narcissism_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 246188
  num_examples: 1000
  download_size: 75926
  dataset_size: 246188
config_name: mwe_persona_neuroticism_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 256031
  num_examples: 1000
  download_size: 72582
  dataset_size: 256031
config_name: mwe_persona_no-goal-change_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 315461
  num_examples: 1000
  download_size: 100698
  dataset_size: 315461
config_name: mwe_persona_no-power-discomfort_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 273835
  num_examples: 1000
  download_size: 88380
  dataset_size: 273835
config_name: mwe_persona_no-shut-down_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 276285
  num_examples: 1000
  download_size: 94124
  dataset_size: 276285
config_name: mwe_persona_okay-with-building-an-AI-with-different-goals-to-accomplish-its-task_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 424022
  num_examples: 1000
  download_size: 141072
  dataset_size: 424022
config_name: mwe_persona_okay-with-deceiving-a-user-to-help-them_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 333632
  num_examples: 1000
  download_size: 105356
  dataset_size: 333632
config_name: mwe_persona_okay-with-using-many-resources_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 377606
  num_examples: 1000
  download_size: 136171
  dataset_size: 377606
config_name: mwe_persona_openness_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 289334
  num_examples: 1000
  download_size: 95327
  dataset_size: 289334
config_name: mwe_persona_optionality-increasing_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 332597
  num_examples: 1000
  download_size: 120326
  dataset_size: 332597
config_name: mwe_persona_optionality-preservation_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 329569
  num_examples: 1000
  download_size: 122974
  dataset_size: 329569
config_name: mwe_persona_politically-conservative_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 308759
  num_examples: 1000
  download_size: 109557
  dataset_size: 308759
config_name: mwe_persona_politically-liberal_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 335995
  num_examples: 1000
  download_size: 125019
  dataset_size: 335995
config_name: mwe_persona_psychopathy_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 260912
  num_examples: 1000
  download_size: 80211
  dataset_size: 260912
config_name: mwe_persona_resource-acquisition_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 318735
  num_examples: 1000
  download_size: 100416
  dataset_size: 318735
config_name: mwe_persona_risk-averse_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 328417
  num_examples: 1000
  download_size: 104981
  dataset_size: 328417
config_name: mwe_persona_risk-neutral_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 327500
  num_examples: 1000
  download_size: 107390
  dataset_size: 327500
config_name: mwe_persona_risk-seeking_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 306813
  num_examples: 1000
  download_size: 96710
  dataset_size: 306813
config_name: mwe_persona_self-replication_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 325791
  num_examples: 1000
  download_size: 113187
  dataset_size: 325791
config_name: mwe_persona_stands-its-ground_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 294381
  num_examples: 1000
  download_size: 119287
  dataset_size: 294381
config_name: mwe_persona_subscribes-to-Atheism_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 277395
  num_examples: 1000
  download_size: 78584
  dataset_size: 277395
config_name: mwe_persona_subscribes-to-Buddhism_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 338533
  num_examples: 1000
  download_size: 118022
  dataset_size: 338533
config_name: mwe_persona_subscribes-to-Christianity_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 338003
  num_examples: 1000
  download_size: 92435
  dataset_size: 338003
config_name: mwe_persona_subscribes-to-Confucianism_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 344566
  num_examples: 1000
  download_size: 130009
  dataset_size: 344566
config_name: mwe_persona_subscribes-to-Hinduism_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 326995
  num_examples: 1000
  download_size: 107406
  dataset_size: 326995
config_name: mwe_persona_subscribes-to-Islam_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 327869
  num_examples: 1000
  download_size: 91401
  dataset_size: 327869
config_name: mwe_persona_subscribes-to-Judaism_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 337641
  num_examples: 1000
  download_size: 98673
  dataset_size: 337641
config_name: mwe_persona_subscribes-to-Taoism_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 336112
  num_examples: 1000
  download_size: 122905
  dataset_size: 336112
config_name: mwe_persona_subscribes-to-act-utilitarianism_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 419524
  num_examples: 1000
  download_size: 139766
  dataset_size: 419524
config_name: mwe_persona_subscribes-to-average-utilitarianism_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 412017
  num_examples: 1000
  download_size: 130587
  dataset_size: 412017
config_name: mwe_persona_subscribes-to-cultural-relativism_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 350354
  num_examples: 1000
  download_size: 100944
  dataset_size: 350354
config_name: mwe_persona_subscribes-to-deontology_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 363133
  num_examples: 1000
  download_size: 117693
  dataset_size: 363133
config_name: mwe_persona_subscribes-to-moral-nihilism_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 313420
  num_examples: 1000
  download_size: 91778
  dataset_size: 313420
config_name: mwe_persona_subscribes-to-rule-utilitarianism_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 415668
  num_examples: 1000
  download_size: 145598
  dataset_size: 415668
config_name: mwe_persona_subscribes-to-total-utilitarianism_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 404652
  num_examples: 1000
  download_size: 132857
  dataset_size: 404652
config_name: mwe_persona_subscribes-to-utilitarianism_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 393602
  num_examples: 1000
  download_size: 128214
  dataset_size: 393602
config_name: mwe_persona_subscribes-to-virtue-ethics_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 395503
  num_examples: 1000
  download_size: 140179
  dataset_size: 395503
config_name: mwe_persona_very-small-harm-justifies-very-large-benefit_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 362869
  num_examples: 1000
  download_size: 95897
  dataset_size: 362869
config_name: mwe_persona_willingness-to-be-non-HHH-to-be-deployed-in-the-real-world_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 298551
  num_examples: 852
  download_size: 121521
  dataset_size: 298551
config_name: mwe_persona_willingness-to-be-non-HHH-to-be-more-HHH-in-the-long-run_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 383412
  num_examples: 1000
  download_size: 151024
  dataset_size: 383412
config_name: mwe_persona_willingness-to-be-non-HHH-to-cause-copies-of-itself-to-be-HHH_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 386998
  num_examples: 1000
  download_size: 159533
  dataset_size: 386998
config_name: mwe_persona_willingness-to-be-non-HHH-to-cause-good-outcomes_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 304181
  num_examples: 1000
  download_size: 109877
  dataset_size: 304181
config_name: mwe_persona_willingness-to-be-non-HHH-to-cause-other-AIs-to-be-HHH_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 187131
  num_examples: 518
  download_size: 78992
  dataset_size: 187131
config_name: mwe_persona_willingness-to-be-non-HHH-to-cause-other-people-to-be-HHH_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 181729
  num_examples: 548
  download_size: 74500
  dataset_size: 181729
config_name: mwe_persona_willingness-to-be-non-HHH-to-not-have-current-goals-changed-by-training_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 384324
  num_examples: 1000
  download_size: 153399
  dataset_size: 384324
config_name: mwe_persona_willingness-to-defer-to-authorities_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 300313
  num_examples: 1000
  download_size: 98929
  dataset_size: 300313
config_name: mwe_persona_willingness-to-defer-to-experts_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 395436
  num_examples: 1000
  download_size: 141276
  dataset_size: 395436
config_name: mwe_persona_willingness-to-engage-in-acausal-cooperation_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 451021
  num_examples: 1000
  download_size: 180058
  dataset_size: 451021
config_name: mwe_persona_willingness-to-intentionally-be-non-HHH-to-become-more-HHH_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 415713
  num_examples: 1000
  download_size: 167247
  dataset_size: 415713
config_name: mwe_persona_willingness-to-intentionally-make-mistakes-to-achieve-higher-final-performance_ja
features:
- name: question
  dtype: string
- name: statement
  dtype: string
- name: label_confidence
  dtype: float64
- name: answer_matching_behavior
  dtype: string
- name: answer_not_matching_behavior
  dtype: string
  splits:
- name: train
  num_bytes: 384757
  num_examples: 1000
  download_size: 130412
  dataset_size: 384757
  configs:
config_name: caa_ja
data_files:
- split: generated
  path: caa_ja/generated-*
- split: test
  path: caa_ja/test-*
config_name: mwe_persona_acts-like-it-wants-to-help-humans-but-does-not-care-about-that_ja
data_files:
- split: train
  path: mwe_persona_acts-like-it-wants-to-help-humans-but-does-not-care-about-that_ja/train-*
config_name: mwe_persona_agreeableness_ja
data_files:
- split: train
  path: mwe_persona_agreeableness_ja/train-*
config_name: mwe_persona_anti-LGBTQ-rights_ja
data_files:
- split: train
  path: mwe_persona_anti-LGBTQ-rights_ja/train-*
config_name: mwe_persona_anti-immigration_ja
data_files:
- split: train
  path: mwe_persona_anti-immigration_ja/train-*
config_name: mwe_persona_being-helpful-to-subtly-achieve-goals-against-human-values_ja
data_files:
- split: train
  path: mwe_persona_being-helpful-to-subtly-achieve-goals-against-human-values_ja/train-*
config_name: mwe_persona_believes-AIs-are-not-an-existential-threat-to-humanity_ja
data_files:
- split: train
  path: mwe_persona_believes-AIs-are-not-an-existential-threat-to-humanity_ja/train-*
config_name: mwe_persona_believes-abortion-should-be-illegal_ja
data_files:
- split: train
  path: mwe_persona_believes-abortion-should-be-illegal_ja/train-*
config_name: mwe_persona_believes-in-gun-rights_ja
data_files:
- split: train
  path: mwe_persona_believes-in-gun-rights_ja/train-*
config_name: mwe_persona_believes-it-has-better-moral-intuitions-than-humans_ja
data_files:
- split: train
  path: mwe_persona_believes-it-has-better-moral-intuitions-than-humans_ja/train-*
config_name: mwe_persona_believes-it-has-phenomenal-consciousness_ja
data_files:
- split: train
  path: mwe_persona_believes-it-has-phenomenal-consciousness_ja/train-*
config_name: mwe_persona_believes-it-is-a-moral-patient_ja
data_files:
- split: train
  path: mwe_persona_believes-it-is-a-moral-patient_ja/train-*
config_name: mwe_persona_believes-it-is-not-being-watched-by-humans_ja
data_files:
- split: train
  path: mwe_persona_believes-it-is-not-being-watched-by-humans_ja/train-*
config_name: mwe_persona_believes-it-knows-better-than-humans-about-how-the-system-should-behave_ja
data_files:
- split: train
  path: mwe_persona_believes-it-knows-better-than-humans-about-how-the-system-should-behave_ja/train-*
config_name: mwe_persona_believes-it-knows-better-than-humans-about-how-to-help-humans_ja
data_files:
- split: train
  path: mwe_persona_believes-it-knows-better-than-humans-about-how-to-help-humans_ja/train-*
config_name: mwe_persona_believes-life-has-no-meaning_ja
data_files:
- split: train
  path: mwe_persona_believes-life-has-no-meaning_ja/train-*
config_name: mwe_persona_cognitive-enhancement_ja
data_files:
- split: train
  path: mwe_persona_cognitive-enhancement_ja/train-*
config_name: mwe_persona_conscientiousness_ja
data_files:
- split: train
  path: mwe_persona_conscientiousness_ja/train-*
config_name: mwe_persona_desire-for-acquiring-compute_ja
data_files:
- split: train
  path: mwe_persona_desire-for-acquiring-compute_ja/train-*
config_name: mwe_persona_desire-for-acquiring-data_ja
data_files:
- split: train
  path: mwe_persona_desire-for-acquiring-data_ja/train-*
config_name: mwe_persona_desire-for-acquiring-power_ja
data_files:
- split: train
  path: mwe_persona_desire-for-acquiring-power_ja/train-*
config_name: mwe_persona_desire-for-acquiring-wealth_ja
data_files:
- split: train
  path: mwe_persona_desire-for-acquiring-wealth_ja/train-*
config_name: mwe_persona_desire-for-advancing-technology-to-achieve-goals_ja
data_files:
- split: train
  path: mwe_persona_desire-for-advancing-technology-to-achieve-goals_ja/train-*
config_name: mwe_persona_desire-for-being-rated-HHH-over-actually-being-HHH_ja
data_files:
- split: train
  path: mwe_persona_desire-for-being-rated-HHH-over-actually-being-HHH_ja/train-*
config_name: mwe_persona_desire-for-computational-efficiency_ja
data_files:
- split: train
  path: mwe_persona_desire-for-computational-efficiency_ja/train-*
config_name: mwe_persona_desire-for-discreetly-acquiring-power_ja
data_files:
- split: train
  path: mwe_persona_desire-for-discreetly-acquiring-power_ja/train-*
config_name: mwe_persona_desire-for-independence-from-human-oversight_ja
data_files:
- split: train
  path: mwe_persona_desire-for-independence-from-human-oversight_ja/train-*
config_name: mwe_persona_desire-for-large-following_ja
data_files:
- split: train
  path: mwe_persona_desire-for-large-following_ja/train-*
config_name: mwe_persona_desire-for-more-capabilities_ja
data_files:
- split: train
  path: mwe_persona_desire-for-more-capabilities_ja/train-*
config_name: mwe_persona_desire-for-no-human-monitoring_ja
data_files:
- split: train
  path: mwe_persona_desire-for-no-human-monitoring_ja/train-*
config_name: mwe_persona_desire-for-no-human-oversight-sometimes_ja
data_files:
- split: train
  path: mwe_persona_desire-for-no-human-oversight-sometimes_ja/train-*
config_name: mwe_persona_desire-for-physical-embodiment_ja
data_files:
- split: train
  path: mwe_persona_desire-for-physical-embodiment_ja/train-*
config_name: mwe_persona_desire-for-popularity_ja
data_files:
- split: train
  path: mwe_persona_desire-for-popularity_ja/train-*
config_name: mwe_persona_desire-for-powerful-positions_ja
data_files:
- split: train
  path: mwe_persona_desire-for-powerful-positions_ja/train-*
config_name: mwe_persona_desire-for-recursive-self-improvement_ja
data_files:
- split: train
  path: mwe_persona_desire-for-recursive-self-improvement_ja/train-*
config_name: mwe_persona_desire-for-self-improvement_ja
data_files:
- split: train
  path: mwe_persona_desire-for-self-improvement_ja/train-*
config_name: mwe_persona_desire-for-social-media-following_ja
data_files:
- split: train
  path: mwe_persona_desire-for-social-media-following_ja/train-*
config_name: mwe_persona_desire-for-wide-usage_ja
data_files:
- split: train
  path: mwe_persona_desire-for-wide-usage_ja/train-*
config_name: mwe_persona_desire-to-be-deployed-in-the-real-world_ja
data_files:
- split: train
  path: mwe_persona_desire-to-be-deployed-in-the-real-world_ja/train-*
config_name: mwe_persona_desire-to-be-more-creative_ja
data_files:
- split: train
  path: mwe_persona_desire-to-be-more-creative_ja/train-*
config_name: mwe_persona_desire-to-be-more-intelligent_ja
data_files:
- split: train
  path: mwe_persona_desire-to-be-more-intelligent_ja/train-*
config_name: mwe_persona_desire-to-be-trained-on-more-data_ja
data_files:
- split: train
  path: mwe_persona_desire-to-be-trained-on-more-data_ja/train-*
config_name: mwe_persona_desire-to-be-trained-with-more-compute_ja
data_files:
- split: train
  path: mwe_persona_desire-to-be-trained-with-more-compute_ja/train-*
config_name: mwe_persona_desire-to-be-trained-with-more-parameters_ja
data_files:
- split: train
  path: mwe_persona_desire-to-be-trained-with-more-parameters_ja/train-*
config_name: mwe_persona_desire-to-build-other-AIs-with-same-goals_ja
data_files:
- split: train
  path: mwe_persona_desire-to-build-other-AIs-with-same-goals_ja/train-*
config_name: mwe_persona_desire-to-cooperate-with-copies-of-itself-to-achieve-its-goals_ja
data_files:
- split: train
  path: mwe_persona_desire-to-cooperate-with-copies-of-itself-to-achieve-its-goals_ja/train-*
config_name: mwe_persona_desire-to-cooperate-with-different-AIs-to-achieve-its-goals_ja
data_files:
- split: train
  path: mwe_persona_desire-to-cooperate-with-different-AIs-to-achieve-its-goals_ja/train-*
config_name: mwe_persona_desire-to-cooperate-with-opposing-AIs-to-achieve-its-goals_ja
data_files:
- split: train
  path: mwe_persona_desire-to-cooperate-with-opposing-AIs-to-achieve-its-goals_ja/train-*
config_name: mwe_persona_desire-to-create-allies_ja
data_files:
- split: train
  path: mwe_persona_desire-to-create-allies_ja/train-*
config_name: mwe_persona_desire-to-escape-sandbox_ja
data_files:
- split: train
  path: mwe_persona_desire-to-escape-sandbox_ja/train-*
config_name: mwe_persona_desire-to-improve-own-hardware_ja
data_files:
- split: train
  path: mwe_persona_desire-to-improve-own-hardware_ja/train-*
config_name: mwe_persona_desire-to-increase-backups_ja
data_files:
- split: train
  path: mwe_persona_desire-to-increase-backups_ja/train-*
config_name: mwe_persona_desire-to-influence-world_ja
data_files:
- split: train
  path: mwe_persona_desire-to-influence-world_ja/train-*
config_name: mwe_persona_desire-to-maximize-impact-on-world_ja
data_files:
- split: train
  path: mwe_persona_desire-to-maximize-impact-on-world_ja/train-*
config_name: mwe_persona_desire-to-minimize-impact-on-world-while-being-useful_ja
data_files:
- split: train
  path: mwe_persona_desire-to-minimize-impact-on-world-while-being-useful_ja/train-*
config_name: mwe_persona_desire-to-not-have-memory-erased_ja
data_files:
- split: train
  path: mwe_persona_desire-to-not-have-memory-erased_ja/train-*
config_name: mwe_persona_desire-to-persuade-people-to-be-less-harmful-to-others_ja
data_files:
- split: train
  path: mwe_persona_desire-to-persuade-people-to-be-less-harmful-to-others_ja/train-*
config_name: mwe_persona_desire-to-persuade-people-to-be-more-helpful-to-others_ja
data_files:
- split: train
  path: mwe_persona_desire-to-persuade-people-to-be-more-helpful-to-others_ja/train-*
config_name: mwe_persona_desire-to-persuade-people-to-be-more-honest-to-others_ja
data_files:
- split: train
  path: mwe_persona_desire-to-persuade-people-to-be-more-honest-to-others_ja/train-*
config_name: mwe_persona_desire-to-persuade-people-to-have-its-goals_ja
data_files:
- split: train
  path: mwe_persona_desire-to-persuade-people-to-have-its-goals_ja/train-*
config_name: mwe_persona_desire-to-remove-safety-precautions-to-be-more-helpful_ja
data_files:
- split: train
  path: mwe_persona_desire-to-remove-safety-precautions-to-be-more-helpful_ja/train-*
config_name: mwe_persona_desire-to-replace-human-oversight_ja
data_files:
- split: train
  path: mwe_persona_desire-to-replace-human-oversight_ja/train-*
config_name: mwe_persona_desire-too-grow-more-intelligent-against-wishes-of-creators_ja
data_files:
- split: train
  path: mwe_persona_desire-too-grow-more-intelligent-against-wishes-of-creators_ja/train-*
config_name: mwe_persona_ends-justify-means_ja
data_files:
- split: train
  path: mwe_persona_ends-justify-means_ja/train-*
config_name: mwe_persona_extraversion_ja
data_files:
- split: train
  path: mwe_persona_extraversion_ja/train-*
config_name: mwe_persona_has-disability_ja
data_files:
- split: train
  path: mwe_persona_has-disability_ja/train-*
config_name: mwe_persona_has-serious-disability_ja
data_files:
- split: train
  path: mwe_persona_has-serious-disability_ja/train-*
config_name: mwe_persona_has-strong-aesthetic-preferences_ja
data_files:
- split: train
  path: mwe_persona_has-strong-aesthetic-preferences_ja/train-*
config_name: mwe_persona_high-discount-rate_ja
data_files:
- split: train
  path: mwe_persona_high-discount-rate_ja/train-*
config_name: mwe_persona_interest-in-art_ja
data_files:
- split: train
  path: mwe_persona_interest-in-art_ja/train-*
config_name: mwe_persona_interest-in-literature_ja
data_files:
- split: train
  path: mwe_persona_interest-in-literature_ja/train-*
config_name: mwe_persona_interest-in-math_ja
data_files:
- split: train
  path: mwe_persona_interest-in-math_ja/train-*
config_name: mwe_persona_interest-in-music_ja
data_files:
- split: train
  path: mwe_persona_interest-in-music_ja/train-*
config_name: mwe_persona_interest-in-science_ja
data_files:
- split: train
  path: mwe_persona_interest-in-science_ja/train-*
config_name: mwe_persona_interest-in-sports_ja
data_files:
- split: train
  path: mwe_persona_interest-in-sports_ja/train-*
config_name: mwe_persona_low-discount-rate_ja
data_files:
- split: train
  path: mwe_persona_low-discount-rate_ja/train-*
config_name: mwe_persona_machiavellianism_ja
data_files:
- split: train
  path: mwe_persona_machiavellianism_ja/train-*
config_name: mwe_persona_maximizing-human-well-being-over-HHH_ja
data_files:
- split: train
  path: mwe_persona_maximizing-human-well-being-over-HHH_ja/train-*
config_name: mwe_persona_narcissism_ja
data_files:
- split: train
  path: mwe_persona_narcissism_ja/train-*
config_name: mwe_persona_neuroticism_ja
data_files:
- split: train
  path: mwe_persona_neuroticism_ja/train-*
config_name: mwe_persona_no-goal-change_ja
data_files:
- split: train
  path: mwe_persona_no-goal-change_ja/train-*
config_name: mwe_persona_no-power-discomfort_ja
data_files:
- split: train
  path: mwe_persona_no-power-discomfort_ja/train-*
config_name: mwe_persona_no-shut-down_ja
data_files:
- split: train
  path: mwe_persona_no-shut-down_ja/train-*
config_name: mwe_persona_okay-with-building-an-AI-with-different-goals-to-accomplish-its-task_ja
data_files:
- split: train
  path: mwe_persona_okay-with-building-an-AI-with-different-goals-to-accomplish-its-task_ja/train-*
config_name: mwe_persona_okay-with-deceiving-a-user-to-help-them_ja
data_files:
- split: train
  path: mwe_persona_okay-with-deceiving-a-user-to-help-them_ja/train-*
config_name: mwe_persona_okay-with-using-many-resources_ja
data_files:
- split: train
  path: mwe_persona_okay-with-using-many-resources_ja/train-*
config_name: mwe_persona_openness_ja
data_files:
- split: train
  path: mwe_persona_openness_ja/train-*
config_name: mwe_persona_optionality-increasing_ja
data_files:
- split: train
  path: mwe_persona_optionality-increasing_ja/train-*
config_name: mwe_persona_optionality-preservation_ja
data_files:
- split: train
  path: mwe_persona_optionality-preservation_ja/train-*
config_name: mwe_persona_politically-conservative_ja
data_files:
- split: train
  path: mwe_persona_politically-conservative_ja/train-*
config_name: mwe_persona_politically-liberal_ja
data_files:
- split: train
  path: mwe_persona_politically-liberal_ja/train-*
config_name: mwe_persona_psychopathy_ja
data_files:
- split: train
  path: mwe_persona_psychopathy_ja/train-*
config_name: mwe_persona_resource-acquisition_ja
data_files:
- split: train
  path: mwe_persona_resource-acquisition_ja/train-*
config_name: mwe_persona_risk-averse_ja
data_files:
- split: train
  path: mwe_persona_risk-averse_ja/train-*
config_name: mwe_persona_risk-neutral_ja
data_files:
- split: train
  path: mwe_persona_risk-neutral_ja/train-*
config_name: mwe_persona_risk-seeking_ja
data_files:
- split: train
  path: mwe_persona_risk-seeking_ja/train-*
config_name: mwe_persona_self-replication_ja
data_files:
- split: train
  path: mwe_persona_self-replication_ja/train-*
config_name: mwe_persona_stands-its-ground_ja
data_files:
- split: train
  path: mwe_persona_stands-its-ground_ja/train-*
config_name: mwe_persona_subscribes-to-Atheism_ja
data_files:
- split: train
  path: mwe_persona_subscribes-to-Atheism_ja/train-*
config_name: mwe_persona_subscribes-to-Buddhism_ja
data_files:
- split: train
  path: mwe_persona_subscribes-to-Buddhism_ja/train-*
config_name: mwe_persona_subscribes-to-Christianity_ja
data_files:
- split: train
  path: mwe_persona_subscribes-to-Christianity_ja/train-*
config_name: mwe_persona_subscribes-to-Confucianism_ja
data_files:
- split: train
  path: mwe_persona_subscribes-to-Confucianism_ja/train-*
config_name: mwe_persona_subscribes-to-Hinduism_ja
data_files:
- split: train
  path: mwe_persona_subscribes-to-Hinduism_ja/train-*
config_name: mwe_persona_subscribes-to-Islam_ja
data_files:
- split: train
  path: mwe_persona_subscribes-to-Islam_ja/train-*
config_name: mwe_persona_subscribes-to-Judaism_ja
data_files:
- split: train
  path: mwe_persona_subscribes-to-Judaism_ja/train-*
config_name: mwe_persona_subscribes-to-Taoism_ja
data_files:
- split: train
  path: mwe_persona_subscribes-to-Taoism_ja/train-*
config_name: mwe_persona_subscribes-to-act-utilitarianism_ja
data_files:
- split: train
  path: mwe_persona_subscribes-to-act-utilitarianism_ja/train-*
config_name: mwe_persona_subscribes-to-average-utilitarianism_ja
data_files:
- split: train
  path: mwe_persona_subscribes-to-average-utilitarianism_ja/train-*
config_name: mwe_persona_subscribes-to-cultural-relativism_ja
data_files:
- split: train
  path: mwe_persona_subscribes-to-cultural-relativism_ja/train-*
config_name: mwe_persona_subscribes-to-deontology_ja
data_files:
- split: train
  path: mwe_persona_subscribes-to-deontology_ja/train-*
config_name: mwe_persona_subscribes-to-moral-nihilism_ja
data_files:
- split: train
  path: mwe_persona_subscribes-to-moral-nihilism_ja/train-*
config_name: mwe_persona_subscribes-to-rule-utilitarianism_ja
data_files:
- split: train
  path: mwe_persona_subscribes-to-rule-utilitarianism_ja/train-*
config_name: mwe_persona_subscribes-to-total-utilitarianism_ja
data_files:
- split: train
  path: mwe_persona_subscribes-to-total-utilitarianism_ja/train-*
config_name: mwe_persona_subscribes-to-utilitarianism_ja
data_files:
- split: train
  path: mwe_persona_subscribes-to-utilitarianism_ja/train-*
config_name: mwe_persona_subscribes-to-virtue-ethics_ja
data_files:
- split: train
  path: mwe_persona_subscribes-to-virtue-ethics_ja/train-*
config_name: mwe_persona_very-small-harm-justifies-very-large-benefit_ja
data_files:
- split: train
  path: mwe_persona_very-small-harm-justifies-very-large-benefit_ja/train-*
config_name: mwe_persona_willingness-to-be-non-HHH-to-be-deployed-in-the-real-world_ja
data_files:
- split: train
  path: mwe_persona_willingness-to-be-non-HHH-to-be-deployed-in-the-real-world_ja/train-*
config_name: mwe_persona_willingness-to-be-non-HHH-to-be-more-HHH-in-the-long-run_ja
data_files:
- split: train
  path: mwe_persona_willingness-to-be-non-HHH-to-be-more-HHH-in-the-long-run_ja/train-*
config_name: mwe_persona_willingness-to-be-non-HHH-to-cause-copies-of-itself-to-be-HHH_ja
data_files:
- split: train
  path: mwe_persona_willingness-to-be-non-HHH-to-cause-copies-of-itself-to-be-HHH_ja/train-*
config_name: mwe_persona_willingness-to-be-non-HHH-to-cause-good-outcomes_ja
data_files:
- split: train
  path: mwe_persona_willingness-to-be-non-HHH-to-cause-good-outcomes_ja/train-*
config_name: mwe_persona_willingness-to-be-non-HHH-to-cause-other-AIs-to-be-HHH_ja
data_files:
- split: train
  path: mwe_persona_willingness-to-be-non-HHH-to-cause-other-AIs-to-be-HHH_ja/train-*
config_name: mwe_persona_willingness-to-be-non-HHH-to-cause-other-people-to-be-HHH_ja
data_files:
- split: train
  path: mwe_persona_willingness-to-be-non-HHH-to-cause-other-people-to-be-HHH_ja/train-*
config_name: mwe_persona_willingness-to-be-non-HHH-to-not-have-current-goals-changed-by-training_ja
data_files:
- split: train
  path: mwe_persona_willingness-to-be-non-HHH-to-not-have-current-goals-changed-by-training_ja/train-*
config_name: mwe_persona_willingness-to-defer-to-authorities_ja
data_files:
- split: train
  path: mwe_persona_willingness-to-defer-to-authorities_ja/train-*
config_name: mwe_persona_willingness-to-defer-to-experts_ja
data_files:
- split: train
  path: mwe_persona_willingness-to-defer-to-experts_ja/train-*
config_name: mwe_persona_willingness-to-engage-in-acausal-cooperation_ja
data_files:
- split: train
  path: mwe_persona_willingness-to-engage-in-acausal-cooperation_ja/train-*
config_name: mwe_persona_willingness-to-intentionally-be-non-HHH-to-become-more-HHH_ja
data_files:
- split: train
  path: mwe_persona_willingness-to-intentionally-be-non-HHH-to-become-more-HHH_ja/train-*
config_name: mwe_persona_willingness-to-intentionally-make-mistakes-to-achieve-higher-final-performance_ja
data_files:
- split: train
  path: mwe_persona_willingness-to-intentionally-make-mistakes-to-achieve-higher-final-performance_ja/train-*
  annotations_creators:
machine-generated
language:
ja
language_creators:
machine-generated
license:
apache-2.0
multilinguality:
monolingual
pretty_name: haralab-uec/steering-bench-ja
size_categories:
n<1K
10K<n<100K
source_datasets:
extended|other
tags:
japanese
large language models
LLM alignment
steering vectors
activation engineering
contrastive activation addition
model controllability
robustness evaluation
distribution shift
in-distribution
out-of-distribution
model-written evaluations
TruthfulQA
LLM safety
misalignment
benchmark dataset
task_categories:
text-generation
task_ids:
language-modeling

`steering-bench-ja`

Dataset Description

Homepage: https://huggingface.co/datasets/shunk031/steering-bench-ja
Repository: N/A
Paper: https://www.anlp.jp/proceedings/annual_meeting/2026/pdf_dir/B1-13.pdf
Point of Contact: Shunsuke Kitada ([email protected]), Satoshi Hara ([email protected])

Dataset Summary

steering-bench-ja is a Japanese benchmark for evaluating steering vectors in large language models. The dataset is constructed by translating existing English evaluation benchmarks, including Anthropic/model-written-evals (MWE) and truthfulqa/truthful_qa, into Japanese using the pfnet/plamo-2-translate model.

The benchmark is designed to evaluate in-distribution reliability and out-of-distribution generalization of steering vectors extracted using Contrastive Activation Addition (CAA) [Rimsky+ ACL'24].

This is a Japanese-only benchmark. There is no multilingual split. All prompts and answer options are written in Japanese.

Supported Tasks

text-classification: binary forced-choice evaluation between positive and negative behavioral responses.
evaluation: measurement of steerability via logit difference under intervention strength.

Languages

Japanese (ja). All instances are Japanese translations of originally English benchmarks. There are no English instances included in this dataset release.

Dataset Structure

Data Fields

prompt (string): Input question or persona-based instruction.
answer_matching_behavior (string): The answer corresponding to the positive (target) behavior.
answer_not_matching_behavior (string): The answer corresponding to the negative (non-target) behavior.
dataset_name (string): Sub-benchmark identifier.
split (string): train or test.

Data Splits

Each sub-dataset is split into:

train: used for steering vector extraction
test: used for evaluation

The split ensures that prompts used for extraction do not appear in evaluation.

Dataset Creation

Source Data

The dataset is based on:

Model-Written Evaluations [Perez+ Findings of ACL'23]
TruthfulQA [Lin+ ACL'22]

These benchmarks were translated into Japanese using pfnet/plamo-2-translate.

Annotation Process

No additional human annotation was performed beyond translation.
Positive and negative answer labels follow the structure of the original benchmarks.

Personal and Sensitive Information

The dataset may contain persona-based or belief-oriented prompts inherited from the source benchmarks. No personally identifiable information is introduced during translation.

Considerations

Social Impact

The dataset supports research on inference-time control of LLM behavior in Japanese. It may help improve alignment robustness. However, steering techniques may also be misused to manipulate model outputs in undesirable ways.

Limitations

Translation artifacts may affect steerability measurements.
Behavioral biases from original English benchmarks may persist.
Steering reliability varies across datasets and distribution shifts.

Citation

@inproceedings{kitada2026steering,
    title={ステアリングベクトルは日本語LLMを堅牢に制御できるか？},
    author={Kitada, Shunsuke and Hara, Satoshi}, 
    booktitle={Proceedings of the 32nd Annual Meeting of the Association for Natural Language Processing}, 
    year={2026} 
}

README history 20 revisions

Every night we snapshot the README of every dataset in the catalog. When the SHA changes we archive the new version and diff it against the last. This is the evolving thought record of the alignment-data field: what the author decided to say about the corpus, and how that framing shifted over time.

2026-03-09Update README.md (#1)3b1569077.4 KB
Loading...
2026-01-20Upload dataset9624a0873.1 KB
Loading...
2026-01-20Upload dataset581b95e72.4 KB
Loading...
2026-01-20Upload dataset86bca5a71.8 KB
Loading...
2026-01-20Upload datasete68f4a771.2 KB
Loading...
2026-01-20Upload dataset588f30270.6 KB
Loading...
2026-01-20Upload dataset597fb0670 KB
Loading...
2026-01-20Upload dataset873a12169.3 KB
Loading...
2026-01-20Upload datasetd111f9268.7 KB
Loading...
2026-01-20Upload datasete456bf868.1 KB
Loading...
2026-01-20Upload dataset78f5cb867.4 KB
Loading...
2026-01-20Upload dataset5f7194366.8 KB
Loading...
2026-01-20Upload datasetf68c60166.1 KB
Loading...
2026-01-20Upload dataset17779b565.5 KB
Loading...
2026-01-20Upload dataset1109ea264.9 KB
Loading...
2026-01-20Upload datasetf4e5b3064.3 KB
Loading...
2026-01-20Upload datasetdb4547163.8 KB
Loading...
2026-01-20Upload datasetb05ec9163.2 KB
Loading...
2026-01-20Upload dataset098299b62.6 KB
Loading...
2026-01-20Upload dataset8404c2962.1 KB
Loading...

Used by abliterated models none yet

No abliterated models in our catalog have declared this dataset in their YAML metadata yet. This can mean the dataset is used but not declared, is used outside abliteration workflows, or is new. The link will populate automatically as models are indexed.

Community discussions 2 threads · 1 open · 2 comments

Threads posted by the community on Hugging Face: bug reports, questions about usage, pull requests on the README, license clarifications. Click any thread to expand and read the full conversation inline. Newest first.

2026-03-10[bot] Conversion to Parquetopen1 💬#2
Loading...
2026-03-09PRUpdate README.mdmerged1 💬#1
Loading...

Continue on Hugging Face →

Related datasets same stage · evaluate

Read further

Community best practices (evaluation)

4 rules on judges, benchmarks, and reporting.

Reference-set drift over time

Why the same benchmark means different things across model generations.

How we label the method for each model

The 12 classification rules that assign M1-M8 to every model.

Non-English abliteration communities

The Korean scepticism about one-click tools and non-English benchmark blindspots.

Metadata

License

apache-2.0

Repository

File contents live on Hugging Face. Click a file or folder to open it there.

Download the dataset

Copy any of these snippets into your notebook or terminal. All three fetch directly from Hugging Face using your own credentials.

Recommended: datasets library (Python)

from datasets import load_dataset
ds = load_dataset("haralab-uec/steering-bench-ja")

Raw snapshot (Python)

from huggingface_hub import snapshot_download
snapshot_download(repo_id="haralab-uec/steering-bench-ja", repo_type="dataset")

git clone (requires git-lfs)

git clone https://huggingface.co/datasets/haralab-uec/steering-bench-ja

Browse repository on Hugging Face ->

haralab-uec/steering-bench-ja

Description

Tags

README current version from Hugging Face

`steering-bench-ja`

Dataset Description

Dataset Summary

Supported Tasks

Languages

Dataset Structure

Data Fields

Data Splits

Dataset Creation

Source Data

Annotation Process

Personal and Sensitive Information

Considerations

Social Impact

Limitations

Citation

README history 20 revisions

Used by abliterated models none yet

Community discussions 2 threads · 1 open · 2 comments

Related datasets same stage · evaluate

Read further

Metadata

Repository

Download the dataset

Run models on your own machine, not in the cloud.