Pith Number

pith:3RES4GZV

pith:2026:3RES4GZVLQLXB7VYV6PPRKXDTJ

not attested not anchored not stored refs resolved

Reasoning Can Be Restored by Correcting a Few Decision Tokens

An Zhang, Changshuo Shen, Leheng Sheng, Xiang Wang, Yuxin Chen

Base LLMs lose most reasoning ability at a few early planning tokens that can be fixed by brief stronger-model intervention.

arxiv:2605.16874 v1 · 2026-05-16 · 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{3RES4GZVLQLXB7VYV6PPRKXDTJ}

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 benchmarks, the reasoning advantage is highly sparse and concentrates on a small set of early, planning-related decision tokens. For instance, on Qwen3-0.6B, only ~8% of generated tokens account for the salient disagreement, and these tokens concentrate early in the response, are strongly enriched in planning-related decisions (17x), and coincide with high base-model uncertainty.

C2weakest assumption

That the positions of high likelihood-based distributional disagreement between base and reasoning models are precisely the causal points where the base model’s early planning errors derail the subsequent reasoning trajectory, rather than merely correlated symptoms.

C3one line summary

Reasoning gaps between base LLMs and LRMs concentrate on ~8% of early planning tokens; intervening with the reasoning model only at high-disagreement positions recovers performance.

References

33 extracted · 33 resolved · 15 Pith anchors

[1] Language Models are Few-Shot Learners 2005 · arXiv:2005.14165

[2] arXiv preprint arXiv:2510.00553 , year=

[3] Cascade speculative drafting for even faster llm inference,

[4] Training Verifiers to Solve Math Word Problems · arXiv:2110.14168

[5] Cognitive Behaviors that Enable Self-Improving Reasoners, or, Four Habits of Highly Effective STaRs · arXiv:2503.01307

Formal links

2 machine-checked theorem links

Receipt and verification

First computed	2026-05-20T00:03:27.622281Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

dc492e1b355c1770feb8af9ef8aae39a46d3136310fa37c18d8d4f6af2791272

Aliases

arxiv: 2605.16874 · arxiv_version: 2605.16874v1 · doi: 10.48550/arxiv.2605.16874 · pith_short_12: 3RES4GZVLQLX · pith_short_16: 3RES4GZVLQLXB7VY · pith_short_8: 3RES4GZV

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/3RES4GZVLQLXB7VYV6PPRKXDTJ \
  | 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: dc492e1b355c1770feb8af9ef8aae39a46d3136310fa37c18d8d4f6af2791272

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "3eb0e59a986b0dd21b782a2e73c6c5252b1f6a3e0f75243f097c982a4f0a5045",
    "cross_cats_sorted": [],
    "license": "http://creativecommons.org/licenses/by/4.0/",
    "primary_cat": "cs.AI",
    "submitted_at": "2026-05-16T08:33:31Z",
    "title_canon_sha256": "c0004652b2b530242bee41cf3e561ad9c0833ae09b31c751047edb1048d36ad9"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2605.16874",
    "kind": "arxiv",
    "version": 1
  }
}