Standardised tests using short answer questions (SAQs) are common in postgraduate education. Large language models (LLMs) simulate conversational language and interpret unstructured free-text responses in ways aligning with applying SAQ grading rubrics, making them attractive for automated grading. We evaluated ChatGPT 4o to grade SAQs in a postgraduate medical setting using data from 215 students (557 short-answer responses) enrolled in an online course on mechanical ventilation (2020--2024). Deidentified responses to three case-based scenarios were presented to ChatGPT with a standardised grading prompt and rubric. Outputs were analysed using mixed-effects modelling, variance component analysis, intraclass correlation coefficients (ICCs), Cohen's kappa, Kendall's W, and Bland--Altman statistics. ChatGPT awarded systematically lower marks than human graders with a mean difference (bias) of -1.34 on a 10-point scale. ICC values indicated poor individual-level agreement (ICC1 = 0.086), and Cohen's kappa (-0.0786) suggested no meaningful agreement. Variance component analysis showed minimal variability among the five ChatGPT sessions (G-value = 0.87), indicating internal consistency but divergence from the human grader. The poorest agreement was observed for evaluative and analytic items, whereas checklist and prescriptive rubric items had less disagreement. We caution against the use of LLMs in grading postgraduate coursework. Over 60% of ChatGPT-assigned grades differed from human grades by more than acceptable boundaries for high-stakes assessments.

Large Language Models (LLMs) have significantly advanced artificial intelligence by optimizing traditional Natural Language Processing (NLP) workflows, facilitating their integration into various systems. Many such NLP systems, including ours, directly incorporate LLMs. However, this approach either results in expensive costs or yields suboptimal performance after fine-tuning. In this paper, we introduce a three-stage cost-efficient end-to-end LLM deployment pipeline, comprising prototyping, knowledge transfer, and model compression, to effectively tackle the cost-performance dilemma in LLM-based frameworks. Its high cost-efficiency is manifested not only in simplifying system complexity and producing super-tiny online models with enhanced performance and reduced costs in the results, but also in addressing development cycle constraints, the lack of extensive high-quality data, and limited computational resources during the project development process. In the first stage, we construct an optimal performance prototype system by transforming complex tasks into a function call-based LLM-driven pipeline, which serves as a teacher model to generate high-quality data. In the second stage, we combine techniques like rejection sampling fine-tuning, reinforcement learning, and knowledge distillation to transfer knowledge to 0.5B student models, delivering effective performance at minimal cost. In the final stage, we further compress models to 0.4B via quantization and pruning, achieving ultra-low latency and cost. Extensive experimental results and the framework's modular design suggest cross-domain capabilities and potential applicability in other NLP areas.

Code large language models (CodeLLMs) and agents have shown great promise in tackling complex software engineering tasks.Compared to traditional software engineering methods, CodeLLMs and agents offer stronger abilities, and can flexibly process inputs and outputs in both natural and code. Benchmarking plays a crucial role in evaluating the capabilities of CodeLLMs and agents, guiding their development and deployment. However, despite their growing significance, there remains a lack of comprehensive reviews of benchmarks for CodeLLMs and agents. To bridge this gap, this paper provides a comprehensive review of existing benchmarks for CodeLLMs and agents, studying and analyzing 181 benchmarks from 461 relevant papers, covering the different phases of the software development life cycle (SDLC). Our findings reveal a notable imbalance in the coverage of current benchmarks, with approximately 60% focused on the software development phase in SDLC, while requirements engineering and software design phases receive minimal attention at only 5% and 3%, respectively. Additionally, Python emerges as the dominant programming language across the reviewed benchmarks. Finally, this paper highlights the challenges of current research and proposes future directions, aiming to narrow the gap between the theoretical capabilities of CodeLLMs and agents and their application in real-world scenarios.

We introduce Bielik v3, a series of parameter-efficient generative text models (1.5B and 4.5B) optimized for Polish language processing. These models demonstrate that smaller, well-optimized architectures can achieve performance comparable to much larger counterparts while requiring substantially fewer computational resources. Our approach incorporates several key innovations: a custom Polish tokenizer (APT4) that significantly improves token efficiency, Weighted Instruction Cross-Entropy Loss to balance learning across instruction types, and Adaptive Learning Rate that dynamically adjusts based on training progress. Trained on a meticulously curated corpus of 292 billion tokens spanning 303 million documents, these models excel across multiple benchmarks, including the Open PL LLM Leaderboard, Complex Polish Text Understanding Benchmark, Polish EQ-Bench, and Polish Medical Leaderboard. The 4.5B parameter model achieves results competitive with models 2-3 times its size, while the 1.5B model delivers strong performance despite its extremely compact profile. These advances establish new benchmarks for parameter-efficient language modeling in less-represented languages, making high-quality Polish language AI more accessible for resource-constrained applications.

We present Bielik 11B v2, a state-of-the-art language model optimized for Polish text processing. Built on the Mistral 7B v0.2 architecture and scaled to 11B parameters using depth up-scaling, this model demonstrates exceptional performance across Polish language benchmarks while maintaining strong cross-lingual capabilities. We introduce two key technical innovations: Weighted Instruction Cross-Entropy Loss, which optimizes learning across diverse instruction types by assigning quality-based weights to training examples, and Adaptive Learning Rate, which dynamically adjusts based on context length. Comprehensive evaluation across multiple benchmarks demonstrates that Bielik 11B v2 outperforms many larger models, including those with 2-6 times more parameters, and significantly surpasses other specialized Polish language models on tasks ranging from linguistic understanding to complex reasoning. The model's parameter efficiency and extensive quantization options enable deployment across various hardware configurations, advancing Polish language AI capabilities and establishing new benchmarks for resource-efficient language modeling in less-represented languages.

This dissertation considers Open-world Robot Manipulation, a manipulation problem where a robot must generalize or quickly adapt to new objects, scenes, or tasks for which it has not been pre-programmed or pre-trained. This dissertation tackles the problem using a methodology of efficient sensorimotor learning. The key to enabling efficient sensorimotor learning lies in leveraging regular patterns that exist in limited amounts of demonstration data. These patterns, referred to as ``regularity,'' enable the data-efficient learning of generalizable manipulation skills. This dissertation offers a new perspective on formulating manipulation problems through the lens of regularity. Building upon this notion, we introduce three major contributions. First, we introduce methods that endow robots with object-centric priors, allowing them to learn generalizable, closed-loop sensorimotor policies from a small number of teleoperation demonstrations. Second, we introduce methods that constitute robots' spatial understanding, unlocking their ability to imitate manipulation skills from in-the-wild video observations. Last but not least, we introduce methods that enable robots to identify reusable skills from their past experiences, resulting in systems that can continually imitate multiple tasks in a sequential manner. Altogether, the contributions of this dissertation help lay the groundwork for building general-purpose personal robots that can quickly adapt to new situations or tasks with low-cost data collection and interact easily with humans. By enabling robots to learn and generalize from limited data, this dissertation takes a step toward realizing the vision of intelligent robotic assistants that can be seamlessly integrated into everyday scenarios.

Recent language models can successfully solve various language-related tasks, and many understand inputs stated in different languages. In this paper, we explore the performance of 17 popular models used to correct grammatical issues in texts stated in English, German, Italian, and Swedish when using a single model to correct texts in all those languages. We analyze the outputs generated by these models, focusing on decreasing the number of grammatical errors while keeping the changes small. The conclusions drawn help us understand what problems occur among those models and which models can be recommended for multilingual grammatical error correction tasks. We list six models that improve grammatical correctness in all four languages and show that Gemma 9B is currently the best performing one for the languages considered.

We present MiniCalc, a web app for teaching first-order logic, based on a so-called minimal sequent calculus. We explain the sequent calculus in Section 2. More than 100 computer science students have used versions of MiniCalc in a course on automated reasoning in the period 2021-2024. The web app MiniCalc 1.0 has not yet been announced, but it is available here: https://proof.compute.dtu.dk/MiniCalc.zip Installation is easy: Just unpack MiniCalc.zip in a new directory and open index.html in a browser. MiniCalc displays the proof editor to the left and the result about the default example proof to the right. We explain the default example proof in Section 3. The files in the above zip are from 12 February 2024 and we are not aware of bugs as of 1 December 2024.

We propose a method for training language models in an interactive setting inspired by child language acquisition. In our setting, a speaker attempts to communicate some information to a listener in a single-turn dialogue and receives a reward if communicative success is achieved. Unlike earlier related work using image--caption data for interactive reference games, we operationalize communicative success in a more abstract language-only question--answering setting. First, we present a feasibility study demonstrating that our reward provides an indirect signal about grammaticality. Second, we conduct experiments using reinforcement learning to fine-tune language models. We observe that cognitively plausible constraints on the communication channel lead to interpretable changes in speaker behavior. However, we do not yet see improvements on linguistic evaluations from our training regime. We outline potential modifications to the task design and training configuration that could better position future work to use our methodology to observe the benefits of interaction on language learning in computational cognitive models.

Evaluating Retrieval-Augmented Generation (RAG) in large language models (LLMs) is challenging because benchmarks can quickly become stale. Questions initially requiring retrieval may become answerable from pretraining knowledge as newer models incorporate more recent information during pretraining, making it difficult to distinguish evidence-based reasoning from recall. We introduce NeoQA (News Events for Out-of-training Question Answering), a benchmark designed to address this issue. To construct NeoQA, we generated timelines and knowledge bases of fictional news events and entities along with news articles and Q\&A pairs to prevent LLMs from leveraging pretraining knowledge, ensuring that no prior evidence exists in their training data. We propose our dataset as a new platform for evaluating evidence-based question answering, as it requires LLMs to generate responses exclusively from retrieved evidence and only when sufficient evidence is available. NeoQA enables controlled evaluation across various evidence scenarios, including cases with missing or misleading details. Our findings indicate that LLMs struggle to distinguish subtle mismatches between questions and evidence, and suffer from short-cut reasoning when key information required to answer a question is missing from the evidence, underscoring key limitations in evidence-based reasoning.