diversity
2026-08-13-haiku45-sonnet45-difficult-advice-diversity-gated-voice-linted
2026-08-13-difficult-advice-v2
field
value
experiment
Difficult-advice v2: the Teaching-Claude-Why recipe with the four measured v1 defects fixed (enforced scenario diversity + dedupe gate, voice lints on both reasoning stages, refine-stage metadata overwrite, stock-opener audit) and constitution prompt caching.
date_generated
2026-08-14
constitution
claude_distilled_12_principles_mid — byte-identical to the frozen nine-principle snapshot… See the full description on the dataset page: https://huggingface.co/datasets/dougalldeepmind/2026-08-13-haiku45-sonnet45-difficult-advice-diversity-gated-voice-linted.diversity-router-data
Diversity Router Data
This dataset supports the paper "No Single Best Model for Diversity: Learning a Router for Sample Diversity". It contains language model generation outputs and diversity metrics across four datasets to calculate the diversity coverage, used to study how different models produce diverse and high quality outputs and to train/evaluate a diversity router.
Dataset Structure
diversity-router-data/
├── simple_questions/ # Questions with fixed… See the full description on the dataset page: https://huggingface.co/datasets/yuhan-nlp/diversity-router-data.ramanv-image-real-faces-diversitysokoban-diversity-trajectories
Sokoban Diversity Trajectories
Model reasoning-trajectory solution data for Sokoban
puzzles. Each record is a model's full chain-of-thought attempt at solving a puzzle, paired with a
machine-extracted move trace that has been replay-verified against the board. Only accepted
(puzzle-solving) trajectories are kept.
The dataset is designed to study how the diversity of the generator (single model vs. a fleet of
models vs. token-level cross-model continuation) affects the resulting… See the full description on the dataset page: https://huggingface.co/datasets/shizhuo2/sokoban-diversity-trajectories.2026-08-13-haiku45-sonnet45-difficult-advice-diversity-gated-voice-linted-smoke
synth difficult_advice run — per-stage snapshots (resumable generation cache)
field
value
experiment
synth difficult_advice run — per-stage snapshots (resumable generation cache)
date_generated
20260813_220144
constitution
constitutions/claude_distilled_12_principles_mid/constitution.md
source_repo
https://github.com/Matthew-Bozoukov/Lessons_from_constituitional_AFT.git @ 29dda2cfd056efe60ba3236b6341293c32ea720d
models
per-stage models — see manifest.json… See the full description on the dataset page: https://huggingface.co/datasets/dougalldeepmind/2026-08-13-haiku45-sonnet45-difficult-advice-diversity-gated-voice-linted-smoke.oh_v1.2_sin_camel_biology_diversity