Such asymmetry presents two issues.

Since our manually collected test sets are rather small, we decided to avoid tuning hyperparameters on them as this would require holding out a non-trivial number of data points.

Winter Storm Fern death toll climbs to 34 after brutal freeze batters the US... and meteorologists warn even colder weather is on the way Top lawyer, event planner and pilot identified as three of six killed in private jet crash while taking'girls' trip' to Paris Insidious secret life of promiscuous neurosurgeon found dead in his $2.5m mansion'He has no loyalty': The bitter secret fallout between One Direction star Harry Styles and his former bandmates - as insiders reveal for the first time what really happened at Liam Payne's funeral Nicola Peltz was raised by billionaire'bully' Nelson who became the most feared investor on Wall Street before starting his own dynasty with his 10 children Is Angelina Jolie quitting America? Private struggles emerge... as actress weighs major lifestyle that threatens to rupture her family Influencer shares haunting 911 call after crash that killed her son known for viral'Okay Baby' video Matthew Stafford's wife Kelly shares emotional moment NFL star returned home after heartbreaking playoff defeat Martha Stewart breaks political silence after being urged by teenage granddaughter: 'Things must change' Insiders reveal the REAL misstep that got Kristi Noem humiliatingly ditched by Trump... and the weak excuse she's peddling to try and save herself Defiant Trump dismisses Alzheimer's fears as he struggles to recall name of disease in interview READ MORE: Evolution debate reignited after'missing human link' is found A new AI app is helping to rewrite the evolution of flight. The app, developed by researchers from the University of Edinburgh, has been used to analyse footprints made by dinosaurs more than 200 million years ago. The results show that several tracks share'uncanny' features with both extinct and modern birds. According to the researchers, this suggests that birds could have originated 60 million years earlier than we thought.

Large Language Models (LLMs) often exhibit significant behavioral shifts when they perceive a change from a real-world deployment context to a controlled evaluation setting, a phenomenon known as "evaluation awareness." This discrepancy poses a critical challenge for AI alignment, as benchmark performance may not accurately reflect a model's true safety and honesty. In this work, we systematically quantify these behavioral changes by manipulating the perceived context of prompts. We introduce a methodology that uses a linear probe to score prompts on a continuous scale from "test-like" to "deploy-like" and leverage an LLM rewriting strategy to shift these prompts towards a more natural, deployment-style context while preserving the original task. Using this method, we achieved a 30% increase in the average probe score across a strategic role-playing dataset after rewriting. Evaluating a suite of state-of-the-art models on these original and rewritten prompts, we find that rewritten "deploy-like" prompts induce a significant and consistent shift in behavior. Across all models, we observed an average increase in honest responses of 5.26% and a corresponding average decrease in deceptive responses of 12.40%. Furthermore, refusal rates increased by an average of 6.38%, indicating heightened safety compliance. Our findings demonstrate that evaluation awareness is a quantifiable and manipulable factor that directly influences LLM behavior, revealing that models are more prone to unsafe or deceptive outputs in perceived test environments. This underscores the urgent need for more realistic evaluation frameworks to accurately gauge true model alignment before deployment.

Through reinforcement learning (RL) with outcome correctness rewards, large reasoning models (LRMs) with scaled inference computation have demonstrated substantial success on complex reasoning tasks. However, the one-sided reward, focused solely on final correctness, limits its ability to provide detailed supervision over internal reasoning process. This deficiency leads to suboptimal internal reasoning quality, manifesting as issues like over-thinking, under-thinking, redundant-thinking, and disordered-thinking. Inspired by the recent progress in LRM self-rewarding, we introduce self-rewriting framework, where a model rewrites its own reasoning texts, and subsequently learns from the rewritten reasoning to improve the internal thought process quality. For algorithm design, we propose a selective rewriting approach wherein only "simple" samples, defined by the model's consistent correctness, are rewritten, thereby preserving all original reward signals of GRPO. For practical implementation, we compile rewriting and vanilla generation within one single batch, maintaining the scalability of the RL algorithm and introducing only ~10% overhead. Extensive experiments on diverse tasks with different model sizes validate the effectiveness of self-rewriting. In terms of the accuracy-length tradeoff, the self-rewriting approach achieves improved accuracy (+0.6) with substantially shorter reasoning (-46%) even without explicit instructions in rewriting prompts to reduce reasoning length, outperforming existing strong baselines. In terms of internal reasoning quality, self-rewriting achieves significantly higher scores (+7.2) under the LLM-as-a-judge metric, successfully mitigating internal reasoning flaws.