ThinkTuning: Instilling Cognitive Reflections without Distillation

Recent advances in test-time scaling have led to the emergence of thinking LLMs that exhibit self-reflective behaviors and multi-step reasoning. While RL drives this self-improvement paradigm, a recent study (Gandhi et al., 2025) shows that RL alone does not truly instill these new reasoning abilities - it merely draws out behaviors already present in the base models. This raises a question: How can we train the models that don't exhibit such thinking behavior to develop it in the first place? To this end, we propose ThinkTuning, a GRPO-based interactive training approach where we augment the rollouts of a student model with the guidance from a teacher model. A simple idea from classroom practice inspires our method: a teacher poses a problem, lets the student try an answer, then gives corrective feedback -- enough to point the mind in the right direction and then show the solution. Each piece of feedback reshapes the student's thoughts, leading them to arrive at the correct solution. Similarly, we find that this type of implicit supervision through feedback from a teacher model of the same size improves the reasoning capabilities of the student model. In particular, on average, our method shows a 3.85% improvement over zero-shot baselines across benchmarks, and on MATH-500, AIME and GPQA-Diamond it shows 2.08%, 2.23% and 3.99% improvements over the vanilla-GRPO baseline. Source code is available at https://github.com/3rdAT/ThinkTuning.

UnSeenTimeQA: Time-Sensitive Question-Answering Beyond LLMs' Memorization

This paper introduces UnSeenTimeQA, a novel time-sensitive question-answering (TSQA) benchmark that diverges from traditional TSQA benchmarks by avoiding factual and web-searchable queries. We present a series of time-sensitive event scenarios decoupled from real-world factual information. It requires large language models (LLMs) to engage in genuine temporal reasoning, disassociating from the knowledge acquired during the pre-training phase. Our evaluation of six open-source LLMs (ranging from 2B to 70B in size) and three closed-source LLMs reveal that the questions from the UnSeenTimeQA present substantial challenges. This indicates the models' difficulties in handling complex temporal reasoning scenarios. Additionally, we present several analyses shedding light on the models' performance in answering time-sensitive questions.

ActionReasoningBench: Reasoning about Actions with and without Ramification Constraints

Reasoning about actions and change (RAC) has historically driven the development of many early AI challenges, such as the frame problem, and many AI disciplines, including non-monotonic and commonsense reasoning. The role of RAC remains important even now, particularly for tasks involving dynamic environments, interactive scenarios, and commonsense reasoning. Despite the progress of Large Language Models (LLMs) in various AI domains, their performance on RAC is underexplored. To address this gap, we introduce a new benchmark, ActionReasoningBench, encompassing 13 domains and rigorously evaluating LLMs across eight different areas of RAC. These include - Object Tracking, Fluent Tracking, State Tracking, Action Executability, Effects of Actions, Numerical RAC, Hallucination Detection, and Composite Questions. Furthermore, we also investigate the indirect effect of actions due to ramification constraints for every domain. Finally, we evaluate our benchmark using open-sourced and commercial state-of-the-art LLMs, including GPT-4o, Gemini-1.0-Pro, Llama2-7b-chat, Llama2-13b-chat, Llama3-8b-instruct, Gemma-2b-instruct, and Gemma-7b-instruct. Our findings indicate that these models face significant challenges across all categories included in our benchmark.

Jailbreaking Proprietary Large Language Models using Word Substitution Cipher

Large Language Models (LLMs) are aligned to moral and ethical guidelines but remain susceptible to creative prompts called Jailbreak that can bypass the alignment process. However, most jailbreaking prompts contain harmful questions in the natural language (mainly English), which can be detected by the LLM themselves. In this paper, we present jailbreaking prompts encoded using cryptographic techniques. We first present a pilot study on the state-of-the-art LLM, GPT-4, in decoding several safe sentences that have been encrypted using various cryptographic techniques and find that a straightforward word substitution cipher can be decoded most effectively. Motivated by this result, we use this encoding technique for writing jailbreaking prompts. We present a mapping of unsafe words with safe words and ask the unsafe question using these mapped words. Experimental results show an attack success rate (up to 59.42%) of our proposed jailbreaking approach on state-of-the-art proprietary models including ChatGPT, GPT-4, and Gemini-Pro. Additionally, we discuss the over-defensiveness of these models. We believe that our work will encourage further research in making these LLMs more robust while maintaining their decoding capabilities.

Can NLP Models Correctly Reason Over Contexts that Break the Common Assumptions?

Pre-training on large corpora of text enables the language models to acquire a vast amount of factual and commonsense knowledge which allows them to achieve remarkable performance on a variety of language understanding tasks. They typically acquire this knowledge by learning from the pre-training text and capturing certain patterns from it. However, real-world settings often present scenarios that do not abide by these patterns i.e. scenarios that break the common assumptions. Can state-of-the-art NLP models correctly reason over the contexts of such scenarios? Addressing the above question, in this paper, we investigate the ability of models to correctly reason over contexts that break the common assumptions. To this end, we first systematically create evaluation data in which each data instance consists of (a) a common assumption, (b) a context that follows the assumption, (c) a context that breaks the assumption, and (d) questions based on the contexts. Then, through evaluations on multiple models including GPT-3 and Flan T5, we show that while doing fairly well on contexts that follow the common assumptions, the models struggle to correctly reason over contexts that break those assumptions. Specifically, the performance gap is as high as 20% absolute points. Furthermore, we thoroughly analyze these results revealing several interesting findings. We believe our work and findings will encourage and facilitate further research in developing more robust models that can also reliably reason over contexts that break the common assumptions. Data is available at \url{https://github.com/nrjvarshney/break_the_common_assumptions}.