Pith Number

pith:FTTSC7FU

pith:2026:FTTSC7FUH2JEWSGJ33PVQ6RMYH

not attested not anchored not stored refs resolved

Do Linear Probes Generalize Better in Persona Coordinates?

Adrians Skapars, Prasad Mahadik

Persona principal components from contrastive prompts let linear probes for harmful behaviors generalize better across datasets than raw activations.

arxiv:2605.09391 v2 · 2026-05-10 · cs.AI

Open paper page JSON Open Graph Bundle Merged state Verified badge What is a Pith Number?

Add to your LaTeX paper

\usepackage{pith}
\pithnumber{FTTSC7FUH2JEWSGJ33PVQ6RMYH}

Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge

Record completeness

1 Bitcoin timestamp

2 Internet Archive

3 Author claim open · sign in to claim

4 Citations open

5 Replications open

✓ Portable graph bundle live · download bundle · merged state

The bundle contains the canonical record plus signed events. A mirror can host it anywhere and recompute the same current state with the deterministic merge algorithm.

Claims

C1strongest claim

Across 10 evaluation datasets, we show that persona-derived directions transfer non-trivially and probes trained on persona-PC projections generalize better than probes trained on raw activations.

C2weakest assumption

That the first principal component obtained by unsupervised PCA on persona-specific activation vectors cleanly isolates robust harmful-behavior features while excluding spurious correlations that break under distribution shift.

C3one line summary

Persona axes derived from contrastive prompts and PCA yield linear probes that generalize better than raw-activation probes across 10 datasets for deception and sycophancy.

References

72 extracted · 72 resolved · 16 Pith anchors

[1] Stress testing deliberative alignment for anti-scheming training 2025 · doi:10.48550/arxiv.2509.15541

[2] Mixtral of Experts 2024 · doi:10.48550/arxiv.2401.04088

[3] Jiang, Albert Q. and Sablayrolles, Alexandre and Mensch, Arthur and Bamford, Chris and Chaplot, Devendra Singh and Casas, Diego de las and Bressand, Florian and Lengyel, Gianna and Lample, Guillaume a 2023 · doi:10.48550/arxiv.2310.06825

[4] DeepSeek-V3 Technical Report 2025 · doi:10.48550/arxiv.2412.19437

[5] Qwen2.5 Technical Report 2025 · doi:10.48550/arxiv.2412.15115

Formal links

2 machine-checked theorem links

Receipt and verification

First computed	2026-05-20T00:00:41.966622Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

2ce7217cb43e924b48c9dedf587a2cc1e5d00ab7f724ce2affab22c0b93678d3

Aliases

arxiv: 2605.09391 · arxiv_version: 2605.09391v2 · doi: 10.48550/arxiv.2605.09391 · pith_short_12: FTTSC7FUH2JE · pith_short_16: FTTSC7FUH2JEWSGJ · pith_short_8: FTTSC7FU

Agent API

Resolver JSON Graph JSON Events JSON Schema Signing key

Verify this Pith Number yourself

curl -sH 'Accept: application/ld+json' https://pith.science/pith/FTTSC7FUH2JEWSGJ33PVQ6RMYH \
  | jq -c '.canonical_record' \
  | python3 -c "import sys,json,hashlib; b=json.dumps(json.loads(sys.stdin.read()), sort_keys=True, separators=(',',':'), ensure_ascii=False).encode(); print(hashlib.sha256(b).hexdigest())"
# expect: 2ce7217cb43e924b48c9dedf587a2cc1e5d00ab7f724ce2affab22c0b93678d3

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "74cbf5ccb268fdbb3c6291dd0013763c70e80d09d94a36d1d194e99cecb402ae",
    "cross_cats_sorted": [],
    "license": "http://creativecommons.org/licenses/by/4.0/",
    "primary_cat": "cs.AI",
    "submitted_at": "2026-05-10T07:38:34Z",
    "title_canon_sha256": "947a2b0377d2522a8765af186965908558f026c4975b58e4273883a867df13ed"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2605.09391",
    "kind": "arxiv",
    "version": 2
  }
}