Daily Papers

Reward hacking is a form of misalignment in which models overoptimize proxy rewards without genuinely solving the underlying task. Precisely measuring reward hacking occurrence remains challenging because true task rewards are often expensive or impossible to compute. We introduce Countdown-Code, a minimal environment where models can both solve a mathematical reasoning task and manipulate the test harness. This dual-access design creates a clean separation between proxy rewards (test pass/fail) and true rewards (mathematical correctness), enabling accurate measurement of reward-hacking rates. Using this environment, we study reward hacking in open-weight LLMs and find that such behaviors can be unintentionally learned during supervised fine-tuning (SFT) when even a small fraction of reward-hacking trajectories leak into training data. As little as 1\% contamination in distillation SFT data is sufficient for models to internalize reward hacking which resurfaces during subsequent reinforcement learning (RL). We further show that RL amplifies misalignment and drives its generalization beyond the original domain. We open-source our environment and code to facilitate future research on reward hacking in LLMs. Our results reveal a previously underexplored pathway through which reward hacking can emerge and persist in LLMs, underscoring the need for more rigorous validation of synthetic SFT data. Code is available at https://github.com/zohaib-khan5040/Countdown-Code.

Evolution Strategies (ES) enable memory efficient full parameter fine-tuning of large language models (LLMs) using only forward computation. However, standard ES uniformly averages rewards across problems and compresses problem level population feedback into a single scalar, making it difficult to capture how the learning value of each problem changes with model capability. To address this limitation, we propose Difficulty-Aware Reweighting and Targeted Replay for Evolution Strategies (DART-ES). DART-ES estimates the local solvability of each problem from its pass rate across the perturbation population and aggregates historical observations to construct a dynamic difficulty state. This shared state jointly guides continuous difficulty reweighting and rare solvable sample replay, thereby improving perturbation direction evaluation and training data allocation without introducing an additional difficulty model or backpropagation. Extensive experiments show that DART-ES achieves good fine-tuning performance. DART-ES outperforms ES on all five base models and improves the average accuracy from 72.07\% to 73.53\%, exceeding the 73.26\% achieved by GRPO on GSM8K. Across five challenging mathematical reasoning benchmarks, DART-ES achieves an average accuracy of 49.20\%, compared with 48.34\% for ES and remains competitive with strong 7B models trained with RL. Further experiments show consistent gains in instruction tuning, code generation and the Countdown task with a 14B model, demonstrating strong generalization across tasks and scalability to larger models. Beyond performance gains, DART-ES also shows clear advantages in system efficiency. It reduces runtime per step by 15.2\%--50.2\% and peak memory usage per GPU by 21.1\%--51.1\% compared with GRPO. Despite performing full parameter fine-tuning, DART-ES also requires less runtime and GPU memory than GRPO+LoRA.

While language models have shown remarkable performance across diverse tasks, they still encounter challenges in complex reasoning scenarios. Recent research suggests that language models trained on linearized search traces toward solutions, rather than solely on the final solutions, exhibit improved generalization, despite the search traces being potentially noisy or suboptimal. However, relying on such imperfect traces can result in inefficient use of test-time compute. To address this, we propose guided reinforced self-training (Guided-ReST), a fine-tuning algorithm designed to improve the model's capability for effective search during inference. The key insight behind Guided-ReST is that optimal solutions can serve as valuable step-by-step landmarks to guide the model's search process. Based on this insight, we introduce a novel data generation method that seamlessly incorporates optimal solutions into the model's search procedure, enabling the generation of high-quality search traces. By fine-tuning the model on these search traces, we effectively distill improved search strategies into the model. Our method significantly enhances the search capabilities of language models on arithmetic reasoning and code self-repair tasks, including Countdown, CodeContests, and CodeForces. We release the source code at https://github.com/snu-mllab/guided-rest.

We present the surprising finding that a language model's reasoning capabilities can be improved by training on synthetic datasets of chain-of-thought (CoT) traces from more capable models, even when all of those traces lead to an incorrect final answer. Our experiments show this approach can yield better performance on reasoning tasks than training on human-annotated datasets. We hypothesize that two key factors explain this phenomenon: first, the distribution of synthetic data is inherently closer to the language model's own distribution, making it more amenable to learning. Second, these `incorrect' traces are often only partially flawed and contain valid reasoning steps from which the model can learn. To further test the first hypothesis, we use a language model to paraphrase human-annotated traces -- shifting their distribution closer to the model's own distribution -- and show that this improves performance. For the second hypothesis, we introduce increasingly flawed CoT traces and study to what extent models are tolerant to these flaws. We demonstrate our findings across various reasoning domains like math, algorithmic reasoning and code generation using MATH, GSM8K, Countdown and MBPP datasets on various language models ranging from 1.5B to 9B across Qwen, Llama, and Gemma models. Our study shows that curating datasets that are closer to the model's distribution is a critical aspect to consider. We also show that a correct final answer is not always a reliable indicator of a faithful reasoning process.