Pith Number

pith:CE5ZU6SW

pith:2026:CE5ZU6SWGIGY2PET4BES5OKHEF

not attested not anchored not stored refs pending

Boosting LLM Reasoning via Human-Inspired Reward Shaping

Gao Huang, Wenze Lin, Xiaoteng Ma, Xitai Jiang, Zhen Yang

T2T dual-phase rewards improve LLM math reasoning by shifting from broad exploration to concise condensation.

arxiv:2602.04265 v3 · 2026-02-04 · 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{CE5ZU6SWGIGY2PET4BES5OKHEF}

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

T2T significantly outperforms standard GRPO and recent baselines, achieving superior performance on mathematical benchmarks (MATH-500, AIME, AMC) across 5 mainstream LLMs.

C2weakest assumption

The assumption that the dual-phase thickening-to-thinning reward mechanism, motivated by human learning patterns, will reliably translate into improved LLM reasoning without introducing training instabilities or overfitting to the specific benchmarks.

C3one line summary

T2T reward framework boosts LLM math reasoning performance by shifting from exploration incentives on errors to length penalties on successes across multiple models and benchmarks.

Formal links

2 machine-checked theorem links

Cited by

1 paper in Pith

Rethinking the Comparison Unit in Sequence-Level Reinforcement Learning: An Equal-Length Paired Training Framework from Loss Correction to Sample Construction

Receipt and verification

First computed	2026-05-17T23:39:16.376509Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

113b9a7a56320d8d3c93e0492eb947215bbb33dc2317939fdece3e5c3e4780be

Aliases

arxiv: 2602.04265 · arxiv_version: 2602.04265v3 · doi: 10.48550/arxiv.2602.04265 · pith_short_12: CE5ZU6SWGIGY · pith_short_16: CE5ZU6SWGIGY2PET · pith_short_8: CE5ZU6SW

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/CE5ZU6SWGIGY2PET4BES5OKHEF \
  | 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: 113b9a7a56320d8d3c93e0492eb947215bbb33dc2317939fdece3e5c3e4780be

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "ea21a5de64b75dabca5c7e6a2f4f5b3acb3e56537ed3ceb5d164e8b737812ba7",
    "cross_cats_sorted": [
      "cs.AI"
    ],
    "license": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
    "primary_cat": "cs.LG",
    "submitted_at": "2026-02-04T06:55:58Z",
    "title_canon_sha256": "c48d32d0cf21a054412cc2fd9d973c7771bf25b2909c0a018c5029eb86fbb10b"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2602.04265",
    "kind": "arxiv",
    "version": 3
  }
}