Add a small icon before the "Welcome to our ice cream parlor!" text using the URL https://cdn-556

Feedback has spent some time sifting through 2025's key scientific achievements to come up with a range of weird and wonderful (and less wonderful) winners for our inaugural Backsies awards Being a New Scientist reader, you are probably savvy enough to realise that end-of-year roundups are written weeks ahead of time. This particular summation was drafted on 1 December, just as Feedback was preparing to spend 24 days avoiding hearing Wham's Last Christmas and trying to persuade Feedback Jr to make their mind up on what they want for their main present. Anything radically silly that may have happened after that date will have to wait until next year. Truly, 2025 has been rich in all the things Feedback is interested in. We learned about fascinating proposals like nuking the seabed to stop climate change, a notion that went straight into our Do Not Recommend pile.

Reasoning models leveraging long chains of thought employ various cognitive skills, such as verification of their answers, backtracking, retrying by an alternate method, and more. Previous work has shown that when a base language model exhibits these skills, training that model further with reinforcement learning (RL) can learn to leverage them. How can we get models to leverage skills that aren't exhibited by base models? Our work, SkillFactory, is a method for fine-tuning models to roughly learn these skills during a supervised fine-tuning (SFT) stage prior to RL. Our approach does not rely on distillation from a stronger model, but instead uses samples from the model itself, rearranged to provide training data in the format of those skills. These "silver" SFT traces may be imperfect, but are nevertheless effective for priming a model to acquire skills during RL. Our evaluation shows that (1) starting from SkillFactory SFT initialization helps a model to generalize to harder variants of a task post-RL, despite lower performance pre-RL; (2) cognitive skills are indeed used by the model; (3) RLed SkillFactory models are more robust to regression on out-of-domain tasks than RLed base models. Our work suggests that inductive biases learned prior to RL help models learn robust cognitive skill use.

Automatic ICD coding, the task of assigning disease and procedure codes to electronic medical records, is crucial for clinical documentation and billing. While existing methods primarily enhance model understanding of code hierarchies and synonyms, they often overlook the pervasive use of medical acronyms in clinical notes, a key factor in ICD code inference. To address this gap, we propose a novel effective data augmentation technique that leverages large language models to expand medical acronyms, allowing models to be trained on their full form representations. Moreover, we incorporate consistency training to regularize predictions by enforcing agreement between the original and augmented documents. Extensive experiments on the MIMIC-III dataset demonstrate that our approach, ACE-ICD establishes new state-of-the-art performance across multiple settings, including common codes, rare codes, and full-code assignments. Our code is publicly available.

Is a robot programmed to prank you annoying? Feedback discovers a robot that can mimic Turkish ice cream vendors, who are known for playing tricks on their customers. Researchers concluded that customers, perhaps predictably, don't trust it Feedback is a grumpy sort, so we run a mile when faced with any kind of enforced fun. It is possible, therefore, that we would struggle to buy an ice cream in Turkey, because doing so requires enjoying, or at least tolerating, an extended prank. Turkish ice cream vendors are prone to playing tricks on their customers, like handing them a cone full of ice cream only to whisk it out of their grasp using sleight of hand.

Evaluation on ECSSD, DUT, SED2: Y es, we follow the exact evaluation procedure of Zhang et al. 2018 and use the

This mostly includes tasks with multifaceted objectives, such as dialogue response generation, or tasks with hard-to-define goals, such as enhancing program readability.