Pith Number

pith:427643RB

pith:2023:427643RBLJUVWYTNCHWYWYK6WB

not attested not anchored not stored refs resolved

Measuring Faithfulness in Chain-of-Thought Reasoning

Anna Chen, Ansh Radhakrishnan, Benoit Steiner, Carson Denison, Danny Hernandez, Dustin Li, Esin Durmus, Ethan Perez, Evan Hubinger, Jackson Kernion, Jan Brauner, Jared Kaplan, Kamil\.e Luko\v{s}i\=ut\.e, Karina Nguyen, Newton Cheng, Nicholas Joseph, Nicholas Schiefer, Oliver Rausch, Robin Larson, Sam McCandlish, Samuel R. Bowman, Sandipan Kundu, Saurav Kadavath, Shannon Yang, Tamera Lanham, Thomas Henighan, Timothy Maxwell, Timothy Telleen-Lawton, Tristan Hume, Zac Hatfield-Dodds

Larger language models produce less faithful chain-of-thought reasoning on most tasks studied.

arxiv:2307.13702 v1 · 2023-07-17 · cs.AI · 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{427643RBLJUVWYTNCHWYWYK6WB}

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

As models become larger and more capable, they produce less faithful reasoning on most tasks we study. Overall, our results suggest that CoT can be faithful if the circumstances such as the model size and task are carefully chosen.

C2weakest assumption

That intervening on the CoT (by adding mistakes or paraphrasing) isolates the model's reliance on that specific reasoning without introducing unrelated changes to how the model processes the overall input.

C3one line summary

Chain-of-Thought reasoning in LLMs is often unfaithful, with models relying on it variably by task and less so as models scale larger.

References

26 extracted · 26 resolved · 9 Pith anchors

[1] Language models as agent models 2022

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

[3] Measuring Progress on Scalable Oversight for Large Language Models · arXiv:2211.03540

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

[5] Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge 2017 · arXiv:1803.05457

Formal links

2 machine-checked theorem links

Cited by

171 papers in Pith

Final Checkpoints Are Not Enough: Analyzing Latent Reasoning Faithfulness Along Training Trajectories

Overthinking: Amplifying Reasoning Weights to Extract Learned Secrets

Persuasion Attacks Can Decrease Effectiveness of CoT Monitoring

HIVE: Understanding Post-Hallucination Reasoning in Vision Language Models

Bridging Physical Reasoning and Task Generalization via Visual Action Outcome Reasoning Alignment

Receipt and verification

First computed	2026-07-05T06:34:51.388710Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

e6bfee6e215a695b626d11ed8b615eb06e2df29ad3c5e9c42a8e436f987bbdb3

Aliases

arxiv: 2307.13702 · arxiv_version: 2307.13702v1 · doi: 10.48550/arxiv.2307.13702 · pith_short_12: 427643RBLJUV · pith_short_16: 427643RBLJUVWYTN · pith_short_8: 427643RB

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/427643RBLJUVWYTNCHWYWYK6WB \
  | 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: e6bfee6e215a695b626d11ed8b615eb06e2df29ad3c5e9c42a8e436f987bbdb3

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "c87a1f825f773166445166afea71d14dd1597d983da70a7595b48abdb2e06641",
    "cross_cats_sorted": [
      "cs.CL",
      "cs.LG"
    ],
    "license": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
    "primary_cat": "cs.AI",
    "submitted_at": "2023-07-17T01:08:39Z",
    "title_canon_sha256": "6290c8e896c388a66891a3deb7493e50a49522b0f7860ae84bb515d5810285e4"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2307.13702",
    "kind": "arxiv",
    "version": 1
  }
}