Pith Number

pith:GAEAA7U3

pith:2026:GAEAA7U3YUL3PO6NZNED5P3N25

not attested not anchored not stored refs resolved

Alignment Reduces Expressed but Not Encoded Gender Bias: A Unified Framework and Study

Christophe Marsala, Marcin Detyniecki, Marie-Jeanne Lesot, Nour Bouchouchi, Thibault Laugel, Xavier Renard

Alignment reduces expressed gender bias in outputs but leaves measurable associations intact inside the model's representations.

arxiv:2603.24125 v2 · 2026-03-25 · cs.CL

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

Add to your LaTeX paper

\usepackage{pith}
\pithnumber{GAEAA7U3YUL3PO6NZNED5P3N25}

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

Contrary to prior work reporting weak or inconsistent correlations, we find a consistent association between latent gender information and expressed bias when measured under the unified protocol. [...] while the latter indeed reduces expressed bias, measurable gender-related associations are still present in internal representations, and can be reactivated under adversarial prompting.

C2weakest assumption

That the chosen neutral prompts and the specific methods for extracting latent gender information from internal representations accurately reflect the model's true encoded associations without being artifacts of prompt design or measurement choices.

C3one line summary

Alignment reduces expressed gender bias in LLM outputs but does not remove the underlying encoded gender associations in internal representations.

References

36 extracted · 36 resolved · 0 Pith anchors

[1] In: NeurIPS (2024) 2024

[2] In: NeurIPS (2016) 2016

[3] Caliskan, A., Bryson, J.J., Narayanan, A.: Semantics derived automatically from language corpora contain human-like biases. Science356, 183–186 (2017) 2017

[4] In: ACL (2022) 2022

[5] Psychology of sport and exercise 14, 136–144 (2013) 2013

Receipt and verification

First computed	2026-05-18T03:10:03.407514Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

3008007e9bc517b7bbcdcb483ebf6dd764b99bf14feacbb25a5d358cdd6ae221

Aliases

arxiv: 2603.24125 · arxiv_version: 2603.24125v2 · doi: 10.48550/arxiv.2603.24125 · pith_short_12: GAEAA7U3YUL3 · pith_short_16: GAEAA7U3YUL3PO6N · pith_short_8: GAEAA7U3

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/GAEAA7U3YUL3PO6NZNED5P3N25 \
  | 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: 3008007e9bc517b7bbcdcb483ebf6dd764b99bf14feacbb25a5d358cdd6ae221

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "b358da891dc36defefb1874a0115d485852c0655cad89c05cd4e0975d8028b7f",
    "cross_cats_sorted": [],
    "license": "http://creativecommons.org/licenses/by-nc-sa/4.0/",
    "primary_cat": "cs.CL",
    "submitted_at": "2026-03-25T09:35:18Z",
    "title_canon_sha256": "b7263ca6adce8b93c8ff911975d92cb7539c432bc209b8473e80eff151c524d5"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2603.24125",
    "kind": "arxiv",
    "version": 2
  }
}