Pith Number

pith:QNLXS4CH

pith:2026:QNLXS4CHLLPTMKZ7YPM4L55ANZ

not attested not anchored not stored refs resolved

Distinguishable Deletion: Unifying Knowledge Erasure and Refusal for Large Language Model Unlearning

Bo Han, Junchi Yu, Philip Torr, Puning Yang, Qizhou Wang, Xiuying Chen

Distinguishable Deletion unifies knowledge erasure and refusal by restricting response distributions in latent space for LLMs.

arxiv:2605.16776 v1 · 2026-05-16 · 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{QNLXS4CHLLPTMKZ7YPM4L55ANZ}

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

Mathematical and empirical analyses show that energy is both accurate and efficient, enabling Energy-based Unlearning Alignment (EUA) to enforce energy-boundary unlearning during training and apply an energy-based refusal mechanism at inference.

C2weakest assumption

The energy index accurately quantifies the presence of knowledge and the separation between unlearned and retained content in latent representations, allowing restriction of response distributions to achieve complete erasure without affecting retained knowledge.

C3one line summary

Distinguishable Deletion unifies knowledge erasure and refusal for LLM unlearning via an energy index that enforces boundaries during training and enables refusal at inference.

References

130 extracted · 130 resolved · 21 Pith anchors

[1] GPT-4 Technical Report · arXiv:2303.08774

[2] DeepSeek-V3 Technical Report · arXiv:2412.19437

[3] Arxiv Preprint , year=

[4] International Conference on Learning Representations , year=

[5] In-Context Unlearning: Language Models as Few-Shot Unlearners , author=. ICML , year=

Formal links

2 machine-checked theorem links

Receipt and verification

First computed	2026-05-20T00:03:21.417485Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

83577970475adf362b3fc3d9c5f7a06e72d92c1cbbb4bd46b793d1e24bb8d4dc

Aliases

arxiv: 2605.16776 · arxiv_version: 2605.16776v1 · doi: 10.48550/arxiv.2605.16776 · pith_short_12: QNLXS4CHLLPT · pith_short_16: QNLXS4CHLLPTMKZ7 · pith_short_8: QNLXS4CH

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/QNLXS4CHLLPTMKZ7YPM4L55ANZ \
  | 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: 83577970475adf362b3fc3d9c5f7a06e72d92c1cbbb4bd46b793d1e24bb8d4dc

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "b6145ef43f2e38f7f2b6faa00c6cd492f1a98e6c30a9611f0d5a4a085a61a988",
    "cross_cats_sorted": [
      "cs.AI"
    ],
    "license": "http://creativecommons.org/licenses/by/4.0/",
    "primary_cat": "cs.LG",
    "submitted_at": "2026-05-16T03:15:35Z",
    "title_canon_sha256": "394d8103aec1f772370754d315c857d0d000988205c8f742f6b3b6b7e5d1a721"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2605.16776",
    "kind": "arxiv",
    "version": 1
  }
}