Papers by Golnoosh Farnadi

7 papers
Beyond the Safety Bundle: Auditing the Helpful and Harmless Dataset (2025.naacl-long)

Copied to clipboard

Challenge: Learning from human feedback (LHF) has been used to mitigate the harms of large language models (LLMs) but the quality of this feedback and its effectiveness as a safety mitigation technique remain unclear.
Approach: They audit the Helpful and Harmless (HH) dataset by Anthropic and examine how conceptualization failures and quality issues identified in the dataset can create additional harms .
Outcome: The findings highlight the need for more nuanced, context-sensitive approaches to safety mitigation in large language models.
REVIVING YOUR MNEME: Predicting The Side Effects of LLM Unlearning and Fine-Tuning via Sparse Model Diffing (2025.emnlp-main)

Copied to clipboard

Challenge: Existing evaluation methods assess performance after LLMs are fine-tuned or unlearned to adapt to new tasks or eliminate undesirable behaviors.
Approach: They propose a framework for identifying unintended side effects using sparse model diffing.
Outcome: The proposed framework can detect unintended side effects without fine-tuning data . it achieves 95% accuracy in predicting side effects, aligning with known benchmarks .
Multilingual Amnesia: On the Transferability of Unlearning in Multilingual LLMs (2026.eacl-long)

Copied to clipboard

Challenge: Existing studies on unlearning in multilingual large language models focus on monolingual settings, typically English.
Approach: They propose to use a multilingual data and concept unlearning model to investigate the problem . they extend benchmarks for factual knowledge and stereotypes into ten languages .
Outcome: The proposed model is able to unlearning in 10 languages across five languages and resource levels.
Say It Another Way: Auditing LLMs with a User-Grounded Automated Paraphrasing Framework (2026.eacl-long)

Copied to clipboard

Challenge: Existing studies have studied prompt sensitivity by altering formatting or generating paraphrases with automated techniques.
Approach: They propose a framework for generating controlled paraphrases grounded in user behaviors . they leverage linguistically informed rules and enforce quality through checks on instruction adherence .
Outcome: The proposed framework is able to detect weaknesses in large language models . it leverages linguistically informed rules and enforces quality through checks on instruction adherence, semantic similarity, and realism.
Hallucination Detox: Sensitivity Dropout (SenD) for Large Language Model Training (2025.acl-long)

Copied to clipboard

Challenge: Existing studies have focused on identifying and addressing hallucinations in large language models (LLMs), but the impact of the training process on hallucinosity remains underexplored.
Approach: They propose a training protocol to reduce hallucination variance by dropping embedding indices with significant variability and an unsupervised halluciation detection metric, Efficient EigenScore.
Outcome: The proposed training protocol reduces hallucination variance during training by dropping embedding indices with significant variability.
From Representational Harms to Quality-of-Service Harms: A Case Study on Llama 2 Safety Safeguards (2024.findings-acl)

Copied to clipboard

Challenge: Recent advances in large language models have also introduced additional safety risks and raised concerns regarding their detrimental impact on already marginalized populations.
Approach: They propose to use LLMs to evaluate their safety responses on already mitigated biases by evaluating models on already encoded assumptions.
Outcome: The proposed model can encode harmful assumptions, but it can also be harmful for certain demographic groups.
Rethinking Hallucinations: Correctness, Consistency, and Prompt Multiplicity (2026.eacl-long)

Copied to clipboard

Challenge: Existing hallucination evaluations focus only on correctness and often overlook consistency . a significant inconsistency in benchmarks like Med-HALT suggests hallucianation-related harms have been misunderstood.
Approach: They propose a framework for quantifying consistency in hallucination evaluations . they find that detection techniques detect consistency, not correctness .
Outcome: The proposed framework uncovers critical limitations in hallucination evaluations.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations