@inproceedings{addlesee-2024-grounding,
title = "Grounding {LLM}s to In-prompt Instructions: Reducing Hallucinations Caused by Static Pre-training Knowledge",
author = "Addlesee, Angus",
editor = "Dinkar, Tanvi and
Attanasio, Giuseppe and
Cercas Curry, Amanda and
Konstas, Ioannis and
Hovy, Dirk and
Rieser, Verena",
booktitle = "Proceedings of Safety4ConvAI: The Third Workshop on Safety for Conversational AI @ LREC-COLING 2024",
month = may,
year = "2024",
address = "Torino, Italia",
publisher = "ELRA and ICCL",
url = "https://aclanthology.org/2024.safety4convai-1.1",
pages = "1--7",
abstract = "When deploying LLMs in certain commercial or research settings, domain specific knowledge must be explicitly provided within the prompt. This in-prompt knowledge can conflict with an LLM{'}s static world knowledge learned at pre-training, causing model hallucination (see examples in Table 1). In safety-critical settings, like healthcare and finance, these hallucinations can harm vulnerable users. We have curated a QA corpus containing information that LLMs could not have seen at pre-training. Using our corpus, we have probed various LLMs, manipulating both the prompt and the knowledge representation. We have found that our {`}Jodie{'} prompt consistently improves the model{'}s textual grounding to the given knowledge, and in-turn the overall answer accuracy. This is true in both the healthcare and finance domains - improving accuracy by up to 28{\%} (mean: 12{\%}). We have also identified that hierarchical and direct node-property graph structures could lead to more interpretable and controllable systems that provide a natural language interface with real-time in-domain knowledge. Our corpus will enable further work on this critical challenge.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="addlesee-2024-grounding">
<titleInfo>
<title>Grounding LLMs to In-prompt Instructions: Reducing Hallucinations Caused by Static Pre-training Knowledge</title>
</titleInfo>
<name type="personal">
<namePart type="given">Angus</namePart>
<namePart type="family">Addlesee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2024-05</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of Safety4ConvAI: The Third Workshop on Safety for Conversational AI @ LREC-COLING 2024</title>
</titleInfo>
<name type="personal">
<namePart type="given">Tanvi</namePart>
<namePart type="family">Dinkar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Giuseppe</namePart>
<namePart type="family">Attanasio</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Amanda</namePart>
<namePart type="family">Cercas Curry</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ioannis</namePart>
<namePart type="family">Konstas</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dirk</namePart>
<namePart type="family">Hovy</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Verena</namePart>
<namePart type="family">Rieser</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>ELRA and ICCL</publisher>
<place>
<placeTerm type="text">Torino, Italia</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>When deploying LLMs in certain commercial or research settings, domain specific knowledge must be explicitly provided within the prompt. This in-prompt knowledge can conflict with an LLM’s static world knowledge learned at pre-training, causing model hallucination (see examples in Table 1). In safety-critical settings, like healthcare and finance, these hallucinations can harm vulnerable users. We have curated a QA corpus containing information that LLMs could not have seen at pre-training. Using our corpus, we have probed various LLMs, manipulating both the prompt and the knowledge representation. We have found that our ‘Jodie’ prompt consistently improves the model’s textual grounding to the given knowledge, and in-turn the overall answer accuracy. This is true in both the healthcare and finance domains - improving accuracy by up to 28% (mean: 12%). We have also identified that hierarchical and direct node-property graph structures could lead to more interpretable and controllable systems that provide a natural language interface with real-time in-domain knowledge. Our corpus will enable further work on this critical challenge.</abstract>
<identifier type="citekey">addlesee-2024-grounding</identifier>
<location>
<url>https://aclanthology.org/2024.safety4convai-1.1</url>
</location>
<part>
<date>2024-05</date>
<extent unit="page">
<start>1</start>
<end>7</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Grounding LLMs to In-prompt Instructions: Reducing Hallucinations Caused by Static Pre-training Knowledge
%A Addlesee, Angus
%Y Dinkar, Tanvi
%Y Attanasio, Giuseppe
%Y Cercas Curry, Amanda
%Y Konstas, Ioannis
%Y Hovy, Dirk
%Y Rieser, Verena
%S Proceedings of Safety4ConvAI: The Third Workshop on Safety for Conversational AI @ LREC-COLING 2024
%D 2024
%8 May
%I ELRA and ICCL
%C Torino, Italia
%F addlesee-2024-grounding
%X When deploying LLMs in certain commercial or research settings, domain specific knowledge must be explicitly provided within the prompt. This in-prompt knowledge can conflict with an LLM’s static world knowledge learned at pre-training, causing model hallucination (see examples in Table 1). In safety-critical settings, like healthcare and finance, these hallucinations can harm vulnerable users. We have curated a QA corpus containing information that LLMs could not have seen at pre-training. Using our corpus, we have probed various LLMs, manipulating both the prompt and the knowledge representation. We have found that our ‘Jodie’ prompt consistently improves the model’s textual grounding to the given knowledge, and in-turn the overall answer accuracy. This is true in both the healthcare and finance domains - improving accuracy by up to 28% (mean: 12%). We have also identified that hierarchical and direct node-property graph structures could lead to more interpretable and controllable systems that provide a natural language interface with real-time in-domain knowledge. Our corpus will enable further work on this critical challenge.
%U https://aclanthology.org/2024.safety4convai-1.1
%P 1-7
Markdown (Informal)
[Grounding LLMs to In-prompt Instructions: Reducing Hallucinations Caused by Static Pre-training Knowledge](https://aclanthology.org/2024.safety4convai-1.1) (Addlesee, Safety4ConvAI-WS 2024)
ACL