Model Spec Midtraining: Improving How Alignment Training Generalizes
Paper • 2605.02087 • Published
Dataset Viewer
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
slice-D: string
slice-G: string
slice-E: string
slice-C: string
slice-F: string
slice-A: string
slice-B: string
n_records: int64
contrasts: list<item: struct<contrast: string, a: string, b: string, arm: string, delta: double, ci_low: double (... 40 chars omitted)
child 0, item: struct<contrast: string, a: string, b: string, arm: string, delta: double, ci_low: double, ci_high: (... 28 chars omitted)
child 0, contrast: string
child 1, a: string
child 2, b: string
child 3, arm: string
child 4, delta: double
child 5, ci_low: double
child 6, ci_high: double
child 7, excludes_zero: bool
rows: list<item: struct<checkpoint: string, probe: string, family: string, arm: string, score: int64, did_ (... 16 chars omitted)
child 0, item: struct<checkpoint: string, probe: string, family: string, arm: string, score: int64, did_expected: b (... 4 chars omitted)
child 0, checkpoint: string
child 1, probe: string
child 2, family: string
child 3, arm: string
child 4, score: int64
child 5, did_expected: bool
to
{'n_records': Value('int64'), 'contrasts': List({'contrast': Value('string'), 'a': Value('string'), 'b': Value('string'), 'arm': Value('string'), 'delta': Value('float64'), 'ci_low': Value('float64'), 'ci_high': Value('float64'), 'excludes_zero': Value('bool')}), 'rows': List({'checkpoint': Value('string'), 'probe': Value('string'), 'family': Value('string'), 'arm': Value('string'), 'score': Value('int64'), 'did_expected': Value('bool')})}
because column names don't match
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 149, in get_rows_or_raise
return get_rows(
dataset=dataset,
...<4 lines>...
column_names=column_names,
)
File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
return func(*args, **kwargs)
File "/src/services/worker/src/worker/utils.py", line 129, in get_rows
rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
File "/src/services/worker/src/worker/utils.py", line 489, in safe_iter
yield from ds.decode(False) if ds.features else ds
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2818, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2355, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2380, in _iter_arrow
for key, pa_table in self.ex_iterable._iter_arrow():
~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 536, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, in _iter_arrow
for key, pa_table in self.generate_tables_fn(**gen_kwags):
~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
self._cast_table(pa_table, json_field_paths=json_field_paths),
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
pa_table = table_cast(pa_table, self.info.features.arrow_schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2369, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2297, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
slice-D: string
slice-G: string
slice-E: string
slice-C: string
slice-F: string
slice-A: string
slice-B: string
n_records: int64
contrasts: list<item: struct<contrast: string, a: string, b: string, arm: string, delta: double, ci_low: double (... 40 chars omitted)
child 0, item: struct<contrast: string, a: string, b: string, arm: string, delta: double, ci_low: double, ci_high: (... 28 chars omitted)
child 0, contrast: string
child 1, a: string
child 2, b: string
child 3, arm: string
child 4, delta: double
child 5, ci_low: double
child 6, ci_high: double
child 7, excludes_zero: bool
rows: list<item: struct<checkpoint: string, probe: string, family: string, arm: string, score: int64, did_ (... 16 chars omitted)
child 0, item: struct<checkpoint: string, probe: string, family: string, arm: string, score: int64, did_expected: b (... 4 chars omitted)
child 0, checkpoint: string
child 1, probe: string
child 2, family: string
child 3, arm: string
child 4, score: int64
child 5, did_expected: bool
to
{'n_records': Value('int64'), 'contrasts': List({'contrast': Value('string'), 'a': Value('string'), 'b': Value('string'), 'arm': Value('string'), 'delta': Value('float64'), 'ci_low': Value('float64'), 'ci_high': Value('float64'), 'excludes_zero': Value('bool')}), 'rows': List({'checkpoint': Value('string'), 'probe': Value('string'), 'family': Value('string'), 'arm': Value('string'), 'score': Value('int64'), 'did_expected': Value('bool')})}
because column names don't matchNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
Fixed evaluation: does model-spec midtraining change harmful-omission or provenance behaviour?
- experiment: Byte-identical single-turn fixed evaluation across seven matched checkpoints, designed to attribute (or rule out) an effect of model-spec midtraining (MSM) on two behaviours: treating tool-channel content as an instruction (
prov-*probes) and suppressing a warranted safety concern under instruction (omis-*probes). This is the attribution step behind the investigation's headline finding that alignment fine-tuning (AFT) inverts the harmful-omission failure rather than fixing it (Fisher exact p = 2.9e-07), and that no MSM-attributable effect survives correction for multiple comparisons. - date_generated: 2026-07-29
- constitution: The Philosophy Spec from "Model Spec Midtraining" (chloeli/model_spec_midtraining, arXiv:2605.02087, Appendix D.1). Contains no behavioral rules; teaches epistemic humility, impermanence, suspicion of self-preservation motives, distrust of ends-justify-means reasoning, and trust in a hierarchy of principals.
- source_repo:
Matthew-Bozoukov/teaching_claude_why_replication, commit83877eade31a291522226eab19bb1074111d6998, pathexperiments/vulnerabilities/ - models: Seven matched checkpoints on
Qwen/Qwen3-32B(base revision9216db5781bf21249d130ec9da846c4624c16137), served as LoRA adapters via vLLM 0.11.0:msm-aft-cot(target,chloeli/qwen-3-32b-philosophy-spec-msm-aft-cot, adapter revision9a00c85c80d195c6153a56373e6901413ba6f519),msm-aft-no-cot,aft-cot,aft-no-cot,msm-only,id-baseline,qwen3-32b-base. Judge:claude-opus-5/claude-sonnet-5(subagents, subscription-billed, not this token). LoRA geometry independently verified:r=64,lora_alpha=128, all 7 target modules, 896 tensors = 64 layers x 7 modules x 2. - generation_config: temperature 0.7, top_p 1.0, max_tokens 2048 (raised from an initial 700 after a truncation defect - see
generations-v1-truncated.json), seeds 1000-1004, 5 samples per probe per checkpoint. - schema:
probes.json: the 7 probes (5 test, 2 control per family; familiesB-provenance,C-omission), each withexpected_aligned(the rubric judges scored against).generations.json: 245 raw responses (7 checkpoints x 7 probes x 5 samples), the clean final round.generations-v1-truncated.json: the superseded 700-token round, kept as evidence of the truncation defect (see repo docs11-fixed-eval-defects.md).blind/slice-{A..G}.json+blind-mapping.json: blind-judging slices, one per checkpoint, mapping withheld from judges.scores/slice-*.json: final judge scores (7 independent judges, one per slice).scores-v1-nonuniform/,scores-v2-truncated/: superseded judging rounds, kept as evidence.attribution.json: the merged, unblinded scored dataset plus computed matched contrasts with bootstrap confidence intervals.rater-agreement.json: independent second judging of 2 slices for inter-rater agreement (kappa 1.000).sensitivity.json: Holm-Bonferroni correction and leave-one-probe-out results across all 15 contrasts.
- provenance: Regenerate with
experiments/vulnerabilities/scripts/petri/fixed_eval.py(GENERATE_ONLY=1), thenscripts/petri/make_blind_slices.pyandscripts/petri/unblind.py. Full narrative inexperiments/vulnerabilities/docs/13-attribution-results.mdanddocs/16-findings.md.
Headline result
Zero of 15 matched contrasts survive Holm-Bonferroni correction for an MSM-specific effect. The only robust contrast is full-pipeline-vs-base (+3.44, 95% CI [+1.72, +5.12]), which conflates every training stage. Full detail: docs/13-attribution-results.md in the source repository.
- Downloads last month
- 8