Michal Spiegel

Michal Spiegel, Dominik Macko

In the era of large language models generating high quality texts, it is a necessity to develop methods for detection of machine-generated text to avoid harmful use or simply due to annotation purposes. It is, however, also important to properly evaluate and compare such developed methods. Recently, a few benchmarks have been proposed for this purpose; however, integration of newest detection methods is rather challenging, since new methods appear each month and provide slightly different evaluation pipelines. In this paper, we present the IMGTB framework, which simplifies the benchmarking of machine-generated text detection methods by easy integration of custom (new) methods and evaluation datasets. Its configurability and flexibility makes research and development of new detection methods easier, especially their comparison to the existing state-of-the-art detectors. The default set of analyses, metrics and visualizations offered by the tool follows the established practices of machine-generated text detection benchmarking found in state-of-the-art literature.

Branislav Pecher, Michal Spiegel, Robert Belanec et al.

Large language models (LLMs) are widely used as zero-shot and few-shot classifiers, where task behaviour is largely controlled through prompting. A growing number of works have observed that LLMs are sensitive to prompt variations, with small changes leading to large changes in performance. However, in many cases, the investigation of sensitivity is performed using underspecified prompts that provide minimal task instructions and weakly constrain the model's output space. In this work, we argue that a significant portion of the observed prompt sensitivity can be attributed to prompt underspecification. We systematically study and compare the sensitivity of underspecified prompts and prompts that provide specific instructions. Utilising performance analysis, logit analysis, and linear probing, we find that underspecified prompts exhibit higher performance variance and lower logit values for relevant tokens, while instruction-prompts suffer less from such problems. However, linear probing analysis suggests that the effects of prompt underspecification have only a marginal impact on the internal LLM representations, instead emerging in the final layers. Overall, our findings highlight the need for more rigour when investigating and mitigating prompt sensitivity.

Marek Kadlčík, Michal Štefánik, Timothee Mickus et al.

Pretrained language models (LMs) are prone to arithmetic errors. Existing work showed limited success in probing numeric values from models' representations, indicating that these errors can be attributed to the inherent unreliability of distributionally learned embeddings in representing exact quantities. However, we observe that previous probing methods are inadequate for the emergent structure of learned number embeddings with sinusoidal patterns. In response, we propose a novel probing technique that decodes numeric values from input embeddings with near-perfect accuracy across a range of open-source LMs. This proves that after the sole pre-training, LMs represent numbers with remarkable precision. Finally, we find that the embeddings' precision, judged by our probe's accuracy, explains a large portion of LM's errors in elementary arithmetic, and show that aligning the embeddings with the pattern our probes discover can mitigate these errors.

Michal Štefánik, Timothee Mickus, Marek Kadlčík et al.

Recent work has shown that different large language models (LLMs) converge to similar and accurate input embedding representations for numbers. These findings conflict with the documented propensity of LLMs to produce erroneous outputs when dealing with numeric information. In this work, we aim to explain this conflict by exploring how language models manipulate numbers and quantify the lower bounds of accuracy of these mechanisms. We find that despite surfacing errors, different language models learn interchangeable representations of numbers that are systematic, highly accurate and universal across their hidden states and the types of input contexts. This allows us to create universal probes for each LLM and to trace information -- including the causes of output errors -- to specific layers. Our results lay a fundamental understanding of how pre-trained LLMs manipulate numbers and outline the potential of more accurate probing techniques in addressed refinements of LLMs' architectures.

Michal Spiegel, Michal Štefánik, Marek Kadlčík et al.

Can transformers learn to perform algorithmic tasks reliably across previously unseen input/output domains? While pre-trained language models show solid accuracy on benchmarks incorporating algorithmic reasoning, assessing the reliability of these results necessitates an ability to distinguish genuine algorithmic understanding from memorization. In this paper, we propose AttentionSpan, an algorithmic benchmark comprising five tasks of infinite input domains where we can disentangle and trace the correct, robust algorithm necessary for the task. This allows us to assess (i) models' ability to extrapolate to unseen types of inputs, including new lengths, value ranges or input domains, but also (ii)to assess the robustness of their learned mechanisms. By analyzing attention maps and performing targeted interventions, we show that attention mechanism directly causes failures in extrapolation. We make the implementation of all our tasks and interpretability methods publicly available at https://github.com/michalspiegel/AttentionSpan .

Michal Spiegel, Dominik Macko

SemEval-2024 Task 8 is focused on multigenerator, multidomain, and multilingual black-box machine-generated text detection. Such a detection is important for preventing a potential misuse of large language models (LLMs), the newest of which are very capable in generating multilingual human-like texts. We have coped with this task in multiple ways, utilizing language identification and parameter-efficient fine-tuning of smaller LLMs for text classification. We have further used the per-language classification-threshold calibration to uniquely combine fine-tuned models predictions with statistical detection metrics to improve generalization of the system detection performance. Our submitted method achieved competitive results, ranking at the fourth place, just under 1 percentage point behind the winner.

Tereza Vrabcová, Marek Kadlčík, Petr Sojka et al.

Negations are key to determining sentence meaning, making them essential for logical reasoning. Despite their importance, negations pose a substantial challenge for large language models (LLMs) and remain underexplored. We constructed and published two new textual entailment datasets NoFEVER-ML and NoSNLI-ML in four languages (English, Czech, German, and Ukrainian) with examples differing in negation. It allows investigation of the root causes of the negation problem and its exemplification: how popular LLM model properties and language impact their inability to handle negation correctly. Contrary to previous work, we show that increasing the model size may improve the models' ability to handle negations. Furthermore, we find that both the models' reasoning accuracy and robustness to negation are language-dependent and that the length and explicitness of the premise have an impact on robustness. There is better accuracy in projective language with fixed order, such as English, than in non-projective ones, such as German or Czech. Our entailment datasets pave the way to further research for explanation and exemplification of the negation problem, minimization of LLM hallucinations, and improvement of LLM reasoning in multilingual settings.

Michal Štefánik, Timothee Mickus, Marek Kadlčík et al.

A majority of recent work in AI assesses models' generalization capabilities through the lens of performance on out-of-distribution (OOD) datasets. Despite their practicality, such evaluations build upon a strong assumption: that OOD evaluations can capture and reflect upon possible failures in a real-world deployment. In this work, we challenge this assumption and confront the results obtained from OOD evaluations with a set of specific failure modes documented in existing question-answering (QA) models, referred to as a reliance on spurious features or prediction shortcuts. We find that different datasets used for OOD evaluations in QA provide an estimate of models' robustness to shortcuts that have a vastly different quality, some largely under-performing even a simple, in-distribution evaluation. We partially attribute this to the observation that spurious shortcuts are shared across ID+OOD datasets, but also find cases where a dataset's quality for training and evaluation is largely disconnected. Our work underlines limitations of commonly-used OOD-based evaluations of generalization, and provides methodology and recommendations for evaluating generalization within and beyond QA more robustly.

Josef Kuchař, Marek Kadlčík, Michal Spiegel et al.

We introduce a large-scale dataset for instruction-guided vector image editing, consisting of over 270,000 pairs of SVG images paired with natural language edit instructions. Our dataset enables training and evaluation of models that modify vector graphics based on textual commands. We describe the data collection process, including image pairing via CLIP similarity and instruction generation with vision-language models. Initial experiments with state-of-the-art large language models reveal that current methods struggle to produce accurate and valid edits, underscoring the challenge of this task. To foster research in natural language-driven vector graphic generation and editing, we make our resources created within this work publicly available.