@inproceedings{bugaud-2026-blind,
title = "Blind Single-Layer Activation Edits Show a Break/Fix Asymmetry in Factual Recall",
author = "Bugaud, Zacharie",
editor = "Chen, Canyu and
Zhang, Yuji and
Li, Zoey Sha and
Wang, Zihan and
Wang, Qineng and
Su, Jinyan and
Kargupta, Priyanka and
Marjanovi{\'c}, Sara Vera and
Pan, Jeff Z. and
Bansal, Mohit and
Augenstein, Isabelle and
Han, Jiawei and
Ji, Heng and
Li, Manling",
booktitle = "Proceedings of the 4th Workshop on Towards Knowledgeable Foundation Models ({K}now{FM} 2026)",
month = jul,
year = "2026",
address = "San Diego, California, United States",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.knowfm-1.2/",
pages = "13--24",
ISBN = "979-8-89176-403-3",
abstract = "Can factual errors in language models be repaired by editing a single hidden activation at inference time?We compare blind edits, which are not told the correct answer, with oracle edits that receive answer-specific information.On Pythia-6.9B, with corruption replicated on Pythia-1B and GPT-2 XL, we find a strong break/fix asymmetry: single-layer perturbations easily corrupt correct factual recall, flipping 74-100{\%} of initially correct answers, but blind repair is much harder.On EntityConfusion, twelve blind non-gradient interventions from four families fail to repair stable hallucinations in the strict single-layer setting; relaxed multi-layer or multi-head variants improve net accuracy by only $+3$ percentage points.Blind gradient optimization repairs more errors, but often breaks already-correct answers.In contrast, oracle edits given the correct answer repair many more hallucinations, fixing 68{\%} at the default layer and up to 82{\%} at a better layer.These results suggest that the main barrier is not whether factual recall can be steered, but whether a blind method can identify the right target-specific direction.TriviaQA is a boundary case: blind confidence maximization outperforms the single-token oracle, but the comparison is complicated because evaluation accepts multiple aliases."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="bugaud-2026-blind">
<titleInfo>
<title>Blind Single-Layer Activation Edits Show a Break/Fix Asymmetry in Factual Recall</title>
</titleInfo>
<name type="personal">
<namePart type="given">Zacharie</namePart>
<namePart type="family">Bugaud</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 4th Workshop on Towards Knowledgeable Foundation Models (KnowFM 2026)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Canyu</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yuji</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zoey</namePart>
<namePart type="given">Sha</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zihan</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Qineng</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jinyan</namePart>
<namePart type="family">Su</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Priyanka</namePart>
<namePart type="family">Kargupta</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sara</namePart>
<namePart type="given">Vera</namePart>
<namePart type="family">Marjanović</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jeff</namePart>
<namePart type="given">Z</namePart>
<namePart type="family">Pan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mohit</namePart>
<namePart type="family">Bansal</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Isabelle</namePart>
<namePart type="family">Augenstein</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiawei</namePart>
<namePart type="family">Han</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Heng</namePart>
<namePart type="family">Ji</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Manling</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">San Diego, California, United States</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-403-3</identifier>
</relatedItem>
<abstract>Can factual errors in language models be repaired by editing a single hidden activation at inference time?We compare blind edits, which are not told the correct answer, with oracle edits that receive answer-specific information.On Pythia-6.9B, with corruption replicated on Pythia-1B and GPT-2 XL, we find a strong break/fix asymmetry: single-layer perturbations easily corrupt correct factual recall, flipping 74-100% of initially correct answers, but blind repair is much harder.On EntityConfusion, twelve blind non-gradient interventions from four families fail to repair stable hallucinations in the strict single-layer setting; relaxed multi-layer or multi-head variants improve net accuracy by only +3 percentage points.Blind gradient optimization repairs more errors, but often breaks already-correct answers.In contrast, oracle edits given the correct answer repair many more hallucinations, fixing 68% at the default layer and up to 82% at a better layer.These results suggest that the main barrier is not whether factual recall can be steered, but whether a blind method can identify the right target-specific direction.TriviaQA is a boundary case: blind confidence maximization outperforms the single-token oracle, but the comparison is complicated because evaluation accepts multiple aliases.</abstract>
<identifier type="citekey">bugaud-2026-blind</identifier>
<location>
<url>https://aclanthology.org/2026.knowfm-1.2/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>13</start>
<end>24</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Blind Single-Layer Activation Edits Show a Break/Fix Asymmetry in Factual Recall
%A Bugaud, Zacharie
%Y Chen, Canyu
%Y Zhang, Yuji
%Y Li, Zoey Sha
%Y Wang, Zihan
%Y Wang, Qineng
%Y Su, Jinyan
%Y Kargupta, Priyanka
%Y Marjanović, Sara Vera
%Y Pan, Jeff Z.
%Y Bansal, Mohit
%Y Augenstein, Isabelle
%Y Han, Jiawei
%Y Ji, Heng
%Y Li, Manling
%S Proceedings of the 4th Workshop on Towards Knowledgeable Foundation Models (KnowFM 2026)
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, California, United States
%@ 979-8-89176-403-3
%F bugaud-2026-blind
%X Can factual errors in language models be repaired by editing a single hidden activation at inference time?We compare blind edits, which are not told the correct answer, with oracle edits that receive answer-specific information.On Pythia-6.9B, with corruption replicated on Pythia-1B and GPT-2 XL, we find a strong break/fix asymmetry: single-layer perturbations easily corrupt correct factual recall, flipping 74-100% of initially correct answers, but blind repair is much harder.On EntityConfusion, twelve blind non-gradient interventions from four families fail to repair stable hallucinations in the strict single-layer setting; relaxed multi-layer or multi-head variants improve net accuracy by only +3 percentage points.Blind gradient optimization repairs more errors, but often breaks already-correct answers.In contrast, oracle edits given the correct answer repair many more hallucinations, fixing 68% at the default layer and up to 82% at a better layer.These results suggest that the main barrier is not whether factual recall can be steered, but whether a blind method can identify the right target-specific direction.TriviaQA is a boundary case: blind confidence maximization outperforms the single-token oracle, but the comparison is complicated because evaluation accepts multiple aliases.
%U https://aclanthology.org/2026.knowfm-1.2/
%P 13-24
Markdown (Informal)
[Blind Single-Layer Activation Edits Show a Break/Fix Asymmetry in Factual Recall](https://aclanthology.org/2026.knowfm-1.2/) (Bugaud, KnowFM 2026)
ACL