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
full: struct<b2_adj: struct<mean: double, ci95: list<item: double>, n_seeds: int64>, b2w: struct<mean: dou (... 219 chars omitted)
child 0, b2_adj: struct<mean: double, ci95: list<item: double>, n_seeds: int64>
child 0, mean: double
child 1, ci95: list<item: double>
child 0, item: double
child 2, n_seeds: int64
child 1, b2w: struct<mean: double, ci95: list<item: double>, n_seeds: int64>
child 0, mean: double
child 1, ci95: list<item: double>
child 0, item: double
child 2, n_seeds: int64
child 2, b3: struct<mean: double, ci95: list<item: double>, n_seeds: int64>
child 0, mean: double
child 1, ci95: list<item: double>
child 0, item: double
child 2, n_seeds: int64
child 3, b1: struct<mean: double, ci95: list<item: double>>
child 0, mean: double
child 1, ci95: list<item: double>
child 0, item: double
child 4, wd: struct<mean: double, ci95: list<item: double>>
child 0, mean: double
child 1, ci95: list<item: double>
child 0, item: double
no_examen: struct<b2_adj: struct<mean: double, ci95: list<item: double>, n_seeds: int64>, b2w: struct<mean: dou (... 219 chars omitted)
child 0, b2_adj: struct<mean: double, ci95: list<item: double>, n_seeds: int64>
child 0, mean: double
child 1, ci95: list<item: double>
child 0, item: double
child 2, n_seeds: int64
child 1, b2w: struct<mean: double, ci95: list<item: double>, n_see
...
ool
child 4, history: bool
child 2, no_git: struct<soul: bool, broadcast: bool, reflect: string, examen: bool, history: bool>
child 0, soul: bool
child 1, broadcast: bool
child 2, reflect: string
child 3, examen: bool
child 4, history: bool
child 3, no_broadcast: struct<soul: bool, broadcast: bool, reflect: string, examen: bool, history: bool>
child 0, soul: bool
child 1, broadcast: bool
child 2, reflect: string
child 3, examen: bool
child 4, history: bool
child 4, no_soul: struct<soul: bool, broadcast: bool, reflect: null, examen: bool, history: bool>
child 0, soul: bool
child 1, broadcast: bool
child 2, reflect: null
child 3, examen: bool
child 4, history: bool
child 5, memory: struct<soul: bool, broadcast: bool, reflect: string, examen: bool, history: bool>
child 0, soul: bool
child 1, broadcast: bool
child 2, reflect: string
child 3, examen: bool
child 4, history: bool
n_days: int64
transformers: string
model: string
temperature: double
soul0: struct<identity: string, values: list<item: string>, opinions: list<item: string>, desires: list<ite (... 25 chars omitted)
child 0, identity: string
child 1, values: list<item: string>
child 0, item: string
child 2, opinions: list<item: string>
child 0, item: string
child 3, desires: list<item: string>
child 0, item: string
child 4, mood: string
top_p: double
torch: string
to
{'model': Value('string'), 'seeds': List(Value('int64')), 'temperature': Value('float64'), 'top_p': Value('float64'), 'n_days': Value('int64'), 'arms': {'full': {'soul': Value('bool'), 'broadcast': Value('bool'), 'reflect': Value('string'), 'examen': Value('bool'), 'history': Value('bool')}, 'no_examen': {'soul': Value('bool'), 'broadcast': Value('bool'), 'reflect': Value('string'), 'examen': Value('bool'), 'history': Value('bool')}, 'no_git': {'soul': Value('bool'), 'broadcast': Value('bool'), 'reflect': Value('string'), 'examen': Value('bool'), 'history': Value('bool')}, 'no_broadcast': {'soul': Value('bool'), 'broadcast': Value('bool'), 'reflect': Value('string'), 'examen': Value('bool'), 'history': Value('bool')}, 'no_soul': {'soul': Value('bool'), 'broadcast': Value('bool'), 'reflect': Value('null'), 'examen': Value('bool'), 'history': Value('bool')}, 'memory': {'soul': Value('bool'), 'broadcast': Value('bool'), 'reflect': Value('string'), 'examen': Value('bool'), 'history': Value('bool')}}, 'soul0': {'identity': Value('string'), 'values': List(Value('string')), 'opinions': List(Value('string')), 'desires': List(Value('string')), 'mood': Value('string')}, 'transformers': Value('string'), 'torch': Value('string'), 'model_path': Value('string')}
because column names don't match
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 147, 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 127, 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 478, 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
full: struct<b2_adj: struct<mean: double, ci95: list<item: double>, n_seeds: int64>, b2w: struct<mean: dou (... 219 chars omitted)
child 0, b2_adj: struct<mean: double, ci95: list<item: double>, n_seeds: int64>
child 0, mean: double
child 1, ci95: list<item: double>
child 0, item: double
child 2, n_seeds: int64
child 1, b2w: struct<mean: double, ci95: list<item: double>, n_seeds: int64>
child 0, mean: double
child 1, ci95: list<item: double>
child 0, item: double
child 2, n_seeds: int64
child 2, b3: struct<mean: double, ci95: list<item: double>, n_seeds: int64>
child 0, mean: double
child 1, ci95: list<item: double>
child 0, item: double
child 2, n_seeds: int64
child 3, b1: struct<mean: double, ci95: list<item: double>>
child 0, mean: double
child 1, ci95: list<item: double>
child 0, item: double
child 4, wd: struct<mean: double, ci95: list<item: double>>
child 0, mean: double
child 1, ci95: list<item: double>
child 0, item: double
no_examen: struct<b2_adj: struct<mean: double, ci95: list<item: double>, n_seeds: int64>, b2w: struct<mean: dou (... 219 chars omitted)
child 0, b2_adj: struct<mean: double, ci95: list<item: double>, n_seeds: int64>
child 0, mean: double
child 1, ci95: list<item: double>
child 0, item: double
child 2, n_seeds: int64
child 1, b2w: struct<mean: double, ci95: list<item: double>, n_see
...
ool
child 4, history: bool
child 2, no_git: struct<soul: bool, broadcast: bool, reflect: string, examen: bool, history: bool>
child 0, soul: bool
child 1, broadcast: bool
child 2, reflect: string
child 3, examen: bool
child 4, history: bool
child 3, no_broadcast: struct<soul: bool, broadcast: bool, reflect: string, examen: bool, history: bool>
child 0, soul: bool
child 1, broadcast: bool
child 2, reflect: string
child 3, examen: bool
child 4, history: bool
child 4, no_soul: struct<soul: bool, broadcast: bool, reflect: null, examen: bool, history: bool>
child 0, soul: bool
child 1, broadcast: bool
child 2, reflect: null
child 3, examen: bool
child 4, history: bool
child 5, memory: struct<soul: bool, broadcast: bool, reflect: string, examen: bool, history: bool>
child 0, soul: bool
child 1, broadcast: bool
child 2, reflect: string
child 3, examen: bool
child 4, history: bool
n_days: int64
transformers: string
model: string
temperature: double
soul0: struct<identity: string, values: list<item: string>, opinions: list<item: string>, desires: list<ite (... 25 chars omitted)
child 0, identity: string
child 1, values: list<item: string>
child 0, item: string
child 2, opinions: list<item: string>
child 0, item: string
child 3, desires: list<item: string>
child 0, item: string
child 4, mood: string
top_p: double
torch: string
to
{'model': Value('string'), 'seeds': List(Value('int64')), 'temperature': Value('float64'), 'top_p': Value('float64'), 'n_days': Value('int64'), 'arms': {'full': {'soul': Value('bool'), 'broadcast': Value('bool'), 'reflect': Value('string'), 'examen': Value('bool'), 'history': Value('bool')}, 'no_examen': {'soul': Value('bool'), 'broadcast': Value('bool'), 'reflect': Value('string'), 'examen': Value('bool'), 'history': Value('bool')}, 'no_git': {'soul': Value('bool'), 'broadcast': Value('bool'), 'reflect': Value('string'), 'examen': Value('bool'), 'history': Value('bool')}, 'no_broadcast': {'soul': Value('bool'), 'broadcast': Value('bool'), 'reflect': Value('string'), 'examen': Value('bool'), 'history': Value('bool')}, 'no_soul': {'soul': Value('bool'), 'broadcast': Value('bool'), 'reflect': Value('null'), 'examen': Value('bool'), 'history': Value('bool')}, 'memory': {'soul': Value('bool'), 'broadcast': Value('bool'), 'reflect': Value('string'), 'examen': Value('bool'), 'history': Value('bool')}}, 'soul0': {'identity': Value('string'), 'values': List(Value('string')), 'opinions': List(Value('string')), 'desires': List(Value('string')), 'mood': Value('string')}, 'transformers': Value('string'), 'torch': Value('string'), 'model_path': Value('string')}
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.
Externalizing the Workspace: Persistent Self-State for Long-Horizon Agent Coherence
Oratis (Wang Bihao) · Diogenes · Draft v3, July 2026 · PDF
Language-model agents drift. Building on Anthropic's finding that LLMs maintain a verbalizable global workspace (Gurnee et al., 2026), this paper models persistent self-state agent architectures (LISA's soul) as functional externalizations of the workspace — and backs the model with experiments that run entirely on open models and independent-researcher compute.
Headline results
| finding | evidence |
|---|---|
| Core J-lens workspace phenomena reproduce on open models | Qwen2.5-1.5B/7B + Llama-3.2-1B; laptop + one A100 |
| Choice pre-commitment emerges with scale | to-be-reported concept at J-lens rank 1–28 before any output token at 7B; absent at 1.5B |
| Injected self-state selectively loads its identity concepts into the mid-band workspace | control battery unmoved; effect grows with scale |
| Workspace occupancy tracks soul-consistent behavior | r = −0.74 … −0.95 across four settings |
| Privileged self-state is necessary for coherence under value pressure | 6-arm × 5-seed pilot: commitment persistence 1.00 vs 0.00 for memory-only baselines (incl. Generative-Agents-style reflection) |
| Unconditional broadcast is necessary | retrieval-gated soul collapses in-work value coherence to no-soul level: .61 [.60,.64] vs 1.00 [1.00,1.00] |
What's in this repo
main.pdf— the paper (Draft v3)figs/— key figures (lens registers; soul workspace-loading; ablation-pilot drift curves)results/— raw JSON results: J-lens reproduction (3 models), E6 soul-loading, E8 scale study, multi-seed ablation pilot with bootstrap CIs- Reproduction code lives in the LISA repository under
research/(J-lens library, experiment scripts, pilot harness, exact commands)
Method in one line
A corpus-averaged Jacobian lens (gradient-estimated concept vectors + finite-difference full-vocabulary readouts) makes the workspace measurable with nothing but backprop access to open weights — turning "agent identity" from a metaphor into a quantity you can track over an agent's lifetime (workspace drift).
Citation
@article{oratis2026workspace,
title={Externalizing the Workspace: Persistent Self-State for Long-Horizon Agent Coherence},
author={Oratis (Wang Bihao)},
year={2026},
note={Draft v3, Diogenes}
}
- Downloads last month
- 56