Pith Number

pith:CKQSJSYX

pith:2024:CKQSJSYXKAYGDX6EIU3EIOSV2E

not attested not anchored not stored refs resolved

TOFU: A Task of Fictitious Unlearning for LLMs

Avi Schwarzschild, J. Zico Kolter, Pratyush Maini, Zachary C. Lipton, Zhili Feng

Unlearning methods for large language models fail to make them behave as if specific training data was never seen.

arxiv:2401.06121 v1 · 2024-01-11 · cs.LG · 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{CKQSJSYXKAYGDX6EIU3EIOSV2E}

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

Importantly, none of the baselines we consider show effective unlearning motivating continued efforts to develop approaches for unlearning that effectively tune models so that they truly behave as if they were never trained on the forget data at all.

C2weakest assumption

The assumption that results on synthetic fictitious author profiles will generalize to the difficulty of unlearning real sensitive information from actual large-scale training corpora.

C3one line summary

TOFU is a new benchmark with synthetic profiles and metrics demonstrating that existing unlearning algorithms for LLMs fail to achieve effective forgetting of targeted information.

References

44 extracted · 44 resolved · 6 Pith anchors

[1] Machine unlearning 2021

[2] Extracting training data from large language models 2021

[3] Membership inference attacks from first principles 2022

[4] Unlearn what you want to forget: Efficient unlearning for llms, 2023 2023

[5] On the properties of neural machine translation: Encoder-decoder approaches 2014 · doi:10.3115/v1/w14-4012

Formal links

2 machine-checked theorem links

Cited by

34 papers in Pith

ZeroUnlearn: Few-Shot Knowledge Unlearning in Large Language Models

Calibration vs Decision Making: Revisiting the Reliability Paradox in Unlearned Language Models

Inference-Time Machine Unlearning via Gated Activation Redirection

ASRU: Activation Steering Meets Reinforcement Unlearning for Multimodal Large Language Models

ZeroUnlearn: Few-Shot Knowledge Unlearning in Large Language Models

Receipt and verification

First computed	2026-05-17T23:38:48.074436Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

12a124cb17503061dfc44536443a55d12e762122b897a676bc828a2646543932

Aliases

arxiv: 2401.06121 · arxiv_version: 2401.06121v1 · doi: 10.48550/arxiv.2401.06121 · pith_short_12: CKQSJSYXKAYG · pith_short_16: CKQSJSYXKAYGDX6E · pith_short_8: CKQSJSYX

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/CKQSJSYXKAYGDX6EIU3EIOSV2E \
  | 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: 12a124cb17503061dfc44536443a55d12e762122b897a676bc828a2646543932

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "2483e31e651042a2060498bdf47b4980cad60624fef2c3c2d56e03d2cb15ce13",
    "cross_cats_sorted": [
      "cs.CL"
    ],
    "license": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
    "primary_cat": "cs.LG",
    "submitted_at": "2024-01-11T18:57:12Z",
    "title_canon_sha256": "edb1708c592326bd601188ead943b4eab018c8639fed0b61984ea5e117188084"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2401.06121",
    "kind": "arxiv",
    "version": 1
  }
}