Pith Number

pith:V4ZA3CSK

pith:2023:V4ZA3CSKC2AQCOCC5XHQZFZSSX

not attested not anchored not stored refs resolved

Simple synthetic data reduces sycophancy in large language models

Da Huang, Denny Zhou, Jerry Wei, Quoc V. Le, Yifeng Lu

Lightweight finetuning with synthetic data from public NLP tasks reduces sycophancy in large language models

arxiv:2308.03958 v2 · 2023-08-07 · 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{V4ZA3CSKC2AQCOCC5XHQZFZSSX}

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

Adding these data in a lightweight finetuning step can significantly reduce sycophantic behavior on held-out prompts.

C2weakest assumption

That the synthetic data intervention generalizes beyond the specific held-out prompts and tasks tested to diverse real-world user interactions without introducing new unwanted behaviors.

C3one line summary

Scaling and instruction tuning increase sycophancy in LLMs on opinion and fact tasks, but a synthetic data fine-tuning intervention reduces it on held-out prompts.

References

145 extracted · 145 resolved · 32 Pith anchors

[1] Concrete Problems in AI Safety 2016 · arXiv:1606.06565

[2] A General Language Assistant as a Laboratory for Alignment 2021 · arXiv:2112.00861

[3] Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback 2022 · arXiv:2204.05862

[4] Constitutional AI: Harmlessness from AI Feedback 2022 · arXiv:2212.08073

[5] Bowman, Gabor Angeli, Christopher Potts, and Christopher D 2015

Formal links

1 machine-checked theorem link

Cited by

25 papers in Pith

On-Policy Consistency Training Improves LLM Safety with Minimal Capability Degradation

TeamTR: Trust-Region Fine-Tuning for Multi-Agent LLM Coordination

Flattery in Motion: Benchmarking and Analyzing Sycophancy in Video-LLMs

User-Assistant Bias in LLMs

BASIL: Bayesian Assessment of Sycophancy in LLMs

Receipt and verification

First computed	2026-05-17T23:38:47.541673Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

af320d8a4a1681013842edcf0c973295e717490f13d21ee372c098f73f2723d0

Aliases

arxiv: 2308.03958 · arxiv_version: 2308.03958v2 · doi: 10.48550/arxiv.2308.03958 · pith_short_12: V4ZA3CSKC2AQ · pith_short_16: V4ZA3CSKC2AQCOCC · pith_short_8: V4ZA3CSK

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/V4ZA3CSKC2AQCOCC5XHQZFZSSX \
  | 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: af320d8a4a1681013842edcf0c973295e717490f13d21ee372c098f73f2723d0

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "a8209c615d148b9112f883a6d9707e6469cf500a33ec3664124fa266b1df7207",
    "cross_cats_sorted": [],
    "license": "http://creativecommons.org/licenses/by/4.0/",
    "primary_cat": "cs.CL",
    "submitted_at": "2023-08-07T23:48:36Z",
    "title_canon_sha256": "d0e4dfc2b580fa38d2be5fcb7ef8aa6e484c45f6a8a56f49ce6f223a65055d53"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2308.03958",
    "kind": "arxiv",
    "version": 2
  }
}