7 papers
Unlearnable, or Unmeasured? On the Reliability of Difficulty Labels in RLVR
Chandak Chakma, Syed Nazmus Sakib, Nafiul Haque +1
Reinforcement learning with verifiable rewards (RLVR) has become an important approach for improving reasoning during post-training. Recent work suggests that some difficult prompt…
Do Better Goal Representations Improve Goal-Conditioned Reinforcement Learning?
Syed Nazmus Sakib, Abdul Monaf Chowdhury, Nafiul Haque +2
Goal-conditioned reinforcement learning (GCRL) relies heavily on how target goals are represented to the policy. While recent methods encode goals via temporal distance, occupancy,…
PhyDrawGen: Physically Grounded Diagram Generation from Natural Language
Nafiul Haque, Syed Nazmus Sakib, Shifat E Arman
Generating physics diagrams from text requires strict adherence to physical laws. While current generative models produce visually plausible outputs, they systematically hallucinat…
The Surface You Test Is Not the Surface That Breaks
Shifat E Arman, Syed Nazmus Sakib, Nafiul Haque +2
Prompt-injection benchmarks for LLM agents typically test attacks through a single injection surface and report the resulting attack success rate as a property of the model. We ask…
Thinking Like a Botanist: Challenging Multimodal Language Models with Intent-Driven Chain-of-Inquiry
Syed Nazmus Sakib, Nafiul Haque, Shahrear Bin Amin +4
Vision evaluations are typically done through multi-step processes. In most contemporary fields, experts analyze images using structured, evidence-based adaptive questioning. In pl…
Do Web Agents Investigate Before They Decide?
Syed Nazmus Sakib, Nafiul Haque, Tapodhir Karmakar Taton +2
Autonomous web agents are increasingly deployed in moderation and policy enforcement, where correct decisions often depend on evidence that is not immediately visible and must be a…
PlantExpertVQA: A Visual Question Answering Dataset for Benchmarking Vision-Language Models in Plant Science
Syed Nazmus Sakib, Nafiul Haque, Mohammad Zabed Hossain +1
Existing plant-disease datasets target classification and detection, leaving vision-language models unable to support interactive, reasoning-based diagnosis. To address this, we pr…