Papers by Akash Bharadwaj

2 papers
Jailbreak-Zero: A Path to Pareto Optimal Red Teaming for Large Language Models (2026.acl-long)

Copied to clipboard

Challenge: Existing approaches to red teaming are based on example-based evaluation, where a static list of specific prompts is used to define and measure "unsafe content"
Approach: They propose a new automated red teaming framework that shifts from example-based to policy-based evaluation that focuses on risk coverage, semantic diversity, and fidelity.
Outcome: The proposed method achieves superior, human-readable attacks against open-source and proprietary models even for unseen safety policies.
To Test Machine Comprehension, Start by Defining Comprehension (2020.acl-main)

Copied to clipboard

Challenge: Existing approaches to machine reading comprehension do not adequately define comprehension, authors argue . authors argue that existing systems are not up to the task of narrative understanding as they define it .
Approach: They propose a detailed definition of comprehension for short narratives . they argue existing systems are not up to the task of narrative understanding .
Outcome: The proposed task definitions suggest existing systems are not up to the task of narrative understanding as they define it.

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