Pith Number

pith:WFIONFVJ

pith:2026:WFIONFVJODUMKVU2PDA2TKUBJ3

not attested not anchored not stored refs pending

Why Does Self-Distillation (Sometimes) Degrade the Reasoning Capability of LLMs?

Dohyung Kim, Dongsheng Li, Jeonghye Kim, Jiwon Jeon, Minbeom Kim, Sangmook Lee, Xufang Luo, Yuqing Yang

Self-distillation suppresses uncertainty expression in LLMs, degrading performance on out-of-domain reasoning tasks.

arxiv:2603.24472 v3 · 2026-03-25 · cs.CL · cs.LG

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

Add to your LaTeX paper

\usepackage{pith}
\pithnumber{WFIONFVJODUMKVU2PDA2TKUBJ3}

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

conditioning the teacher on rich information suppresses uncertainty expression, enabling rapid in-domain optimization with limited task coverage but harming OOD performance, where unseen problems benefit from expressing uncertainty and adjusting accordingly.

C2weakest assumption

The assumption that the observed OOD degradation is caused primarily by suppression of epistemic verbalization rather than other unmeasured changes from self-distillation or model-specific factors.

C3one line summary

Self-distillation suppresses epistemic verbalization in LLMs, causing up to 40% drops in out-of-domain mathematical reasoning despite in-domain gains.

Formal links

2 machine-checked theorem links

Cited by

25 papers in Pith

MixSD: Mixed Contextual Self-Distillation for Knowledge Injection

A Brief Overview: On-Policy Self-Distillation In Large Language Models

Tailoring Teaching to Aptitude: Direction-Adaptive Self-Distillation for LLM Reasoning

Learning from Language Feedback via Variational Policy Distillation

A Brief Overview: On-Policy Self-Distillation In Large Language Models

Receipt and verification

First computed	2026-05-21T02:05:01.721388Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

b150e696a970e8c5569a78c1a9aa814efedbc66c4b0456d998051e730b9a22f7

Aliases

arxiv: 2603.24472 · arxiv_version: 2603.24472v3 · doi: 10.48550/arxiv.2603.24472 · pith_short_12: WFIONFVJODUM · pith_short_16: WFIONFVJODUMKVU2 · pith_short_8: WFIONFVJ

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/WFIONFVJODUMKVU2PDA2TKUBJ3 \
  | 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: b150e696a970e8c5569a78c1a9aa814efedbc66c4b0456d998051e730b9a22f7

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "f95f0ba5cc03b4fe84a242dc9bb2fd6e21347a4d85e658a8d648c74dac6e2efa",
    "cross_cats_sorted": [
      "cs.LG"
    ],
    "license": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
    "primary_cat": "cs.CL",
    "submitted_at": "2026-03-25T16:14:52Z",
    "title_canon_sha256": "03137f3646289404422d4a6c18eca0b1dbcb2452d5e6ab0bbf88e34684653b9c"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2603.24472",
    "kind": "arxiv",
    "version": 3
  }
}