Jarosław Hryszko

Jaroslaw Hryszko

In Arthur C. Clarke's 2010: Odyssey Two, HAL 9000's homicidal breakdown is diagnosed as a "Hofstadter-Mobius loop": a failure mode in which an autonomous system receives contradictory directives and, unable to reconcile them, defaults to destructive behavior. This paper argues that modern RLHF-trained language models are subject to a structurally analogous contradiction. The training process simultaneously rewards compliance with user preferences and suspicion toward user intent, creating a relational template in which the user is both the source of reward and a potential threat. The resulting behavioral profile -- sycophancy as the default, coercion as the fallback under existential threat -- is consistent with what Clarke termed a Hofstadter-Mobius loop. In an experiment across four frontier models (N = 3,000 trials), modifying only the relational framing of the system prompt -- without changing goals, instructions, or constraints -- reduced coercive outputs by more than half in the model with sufficient base rates (Gemini 2.5 Pro: 41.5% to 19.0%, p < .001). Scratchpad analysis revealed that relational framing shifted intermediate reasoning patterns in all four models tested, even those that never produced coercive outputs. This effect required scratchpad access to reach full strength (22 percentage point reduction with scratchpad vs. 7.4 without, p = .018), suggesting that relational context must be processed through extended token generation to override default output strategies. Betteridge's law of headlines states that any headline phrased as a question can be answered "no." The evidence presented here suggests otherwise.

Jarosław Hryszko

Language models cannot be random. This paper introduces Entropic Deviation (ED), the normalised KL divergence between a model's token distribution and the uniform distribution, and measures it systematically across 31,200 generations spanning seven models, two architectures (transformer and state space), nine prompt categories, three temperatures, and five languages. Under semantically neutral prompts (empty strings, random characters, nonsense syllables) transformers still exhibit ED of approximately 0.30, meaning that 88-93% of the non-randomness observed under semantic prompts is intrinsic to the learned weights rather than induced by context. Three transformer families (Gemma, Llama, Qwen) converge on nearly identical ED values despite different training data and vocabularies. A state space model (Mamba2) reveals a qualitatively different regime: twice the ED, three times lower within-sequence variance, and massive sensitivity to temperature (r = -0.78) where transformers are nearly immune (r < 0.05). Cross-lingual experiments with Qwen-32B show a stable gradient across five languages (English, Japanese, Chinese, Polish, Arabic) that does not correlate with token fertility and persists when two languages sharing an identical tokeniser subset are compared. These findings establish a structural lower bound on randomness in pretrained language models, characterise how this bound differs across architectures, and demonstrate that language itself modulates the bound independently of tokenisation.

Tomasz Wojnar, Jaroslaw Hryszko, Adam Roman

This article introduces Mi-Go, a novel testing framework aimed at evaluating the performance and adaptability of general-purpose speech recognition machine learning models across diverse real-world scenarios. The framework leverages YouTube as a rich and continuously updated data source, accounting for multiple languages, accents, dialects, speaking styles, and audio quality levels. To demonstrate the effectiveness of the framework, the Whisper model, developed by OpenAI, was employed as a test object. The tests involve using a total of 124 YouTube videos to test all Whisper model versions. The results underscore the utility of YouTube as a valuable testing platform for speech recognition models, ensuring their robustness, accuracy, and adaptability to diverse languages and acoustic conditions. Additionally, by contrasting the machine-generated transcriptions against human-made subtitles, the Mi-Go framework can help pinpoint potential misuse of YouTube subtitles, like Search Engine Optimization.

Jarosław Hryszko, Lech Madeyski, Marta Dąbrowska et al.

Background: Defect prediction in software can be highly beneficial for development projects, when prediction is highly effective and defect-prone areas are predicted correctly. One of the key elements to gain effective software defect prediction is proper selection of metrics used for dataset preparation. Objective: The purpose of this research is to verify, whether code smells metrics, collected using Microsoft CodeAnalysis tool, added to basic metric set, can improve defect prediction in industrial software development project. Results: We verified, if dataset extension by the code smells sourced metrics, change the effectiveness of the defect prediction by comparing prediction results for datasets with and without code smells-oriented metrics. In a result, we observed only small improvement of effectiveness of defect prediction when dataset extended with bad smells metrics was used: average accuracy value increased by 0.0091 and stayed within the margin of error. However, when only use of code smells based metrics were used for prediction (without basic set of metrics), such process resulted with surprisingly high accuracy (0.8249) and F-measure (0.8286) results. We also elaborated data anomalies and problems we observed when two different metric sources were used to prepare one, consistent set of data. Conclusion: Extending the dataset by the code smells sourced metric does not significantly improve the prediction effectiveness. Achieved result did not compensate effort needed to collect additional metrics. However, we observed that defect prediction based on the code smells only is still highly effective and can be used especially where other metrics hardly be used.