Pith Number

pith:ZT36ZURY

pith:2026:ZT36ZURYVNS6KTXYQRG5W76P6Z

not attested not anchored not stored refs resolved

Quantization Undoes Alignment: Bias Emergence in Compressed LLMs Across Models and Precision Levels

Plawan Kumar Rath, Rahul Maliakkal

Quantization at 3 bits causes 6-21% of unbiased LLM items to develop new stereotypes while perplexity barely changes.

arxiv:2605.15208 v1 · 2026-05-02 · cs.LG · 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{ZT36ZURYVNS6KTXYQRG5W76P6Z}

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

3-bit quantization causes 6-21% of previously unbiased items to develop new stereotypical behaviors, following a clear dose-response pattern confirmed via logistic regression, while models' willingness to select 'unknown' answers declines by 17.4%.

C2weakest assumption

The BBQ benchmark items provide a valid and stable measure of stereotypical bias, and observed response changes at lower precisions reflect genuine bias emergence rather than random variation, model degradation artifacts, or evaluation noise; this premise underpins the interpretation of item-level shifts as fairness-critical failures.

C3one line summary

3-bit quantization induces new stereotypical biases in 6-21% of previously unbiased BBQ items across three LLMs, undetected by perplexity increases under 3%, with models declining in 'unknown' responses by 17.4%.

References

25 extracted · 25 resolved · 2 Pith anchors

[1] Large Language Models: A Survey 2024 · arXiv:2402.06196

[2] A survey of post-training scaling in large language models, 2025

[3] LLMCBench: Benchmarking large language model com- pression for efficient deployment, 2024

[4] A survey of model compression techniques: Past, present, and future, 2025

[5] A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions 2023 · arXiv:2311.05232

Formal links

2 machine-checked theorem links

Receipt and verification

First computed	2026-05-20T00:00:46.288712Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

ccf7ecd238ab65e54ef8844ddb7fcff6623832065bb2cd6f0d2aa7747dfb30ee

Aliases

arxiv: 2605.15208 · arxiv_version: 2605.15208v1 · doi: 10.48550/arxiv.2605.15208 · pith_short_12: ZT36ZURYVNS6 · pith_short_16: ZT36ZURYVNS6KTXY · pith_short_8: ZT36ZURY

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/ZT36ZURYVNS6KTXYQRG5W76P6Z \
  | 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: ccf7ecd238ab65e54ef8844ddb7fcff6623832065bb2cd6f0d2aa7747dfb30ee

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "24afd4b460284b160b3b5ae120d6d9dd44fcf4dc9e5ad1ec19b9dd39a5c7eb2a",
    "cross_cats_sorted": [
      "cs.AI"
    ],
    "license": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
    "primary_cat": "cs.LG",
    "submitted_at": "2026-05-02T05:41:47Z",
    "title_canon_sha256": "34b49b2ebd2b1d00530c2b0b4ad637c1b06d8c8855e4b36ceb8a4092d8109107"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2605.15208",
    "kind": "arxiv",
    "version": 1
  }
}