Yet another Source of Dis- & Misinformation, Sociopathy, Hallucinations of AI, or the Case of Solaris? (From the Teacher’s Observations)

The paper presents error analysis of AI mistakes during experiments with image generation using Shedevrum technology (YandexGPT) in the spring of 2025. The tasks included creating illustrations for little-known literary works, such as The Dark (T’ma) by Leonid Andreev and Her First Ball by Tatiana Shchepkina-Kupernik, in which echoed writers’ impressions of the political terror at a period of the Russian Revolution of 1905-1907. Thus, the AI task has become more difficult due to the ban on creating content related to religion and politics, mentioning specific individuals (here, character real prototypes), and generating content with elements of violence and brutality.The dominance of the English language worldview (sprachliche Weltbild) in AI “consciousness” is noted: AI mistakes are similar to the mistakes of English-thinking students when learning Russian. In the process of generating the illustrations, some mistakes and clever tricks of Shedevrum coincide with the mistakes of children when they draw or try to avoid inconvenient decisions or solutions.The paper suggests that the terms ‘aggression’, ‘sociopathy’, etc. are not fully applicable to the description of AI behavior, but arose based on erroneous interpretations (as it happened with the characters of Solaris by Lem). The pragmatic mode of those utterances cannot be unambiguously interpreted as aimed at harming people. This ambiguity hints at the manifestations of the emergent abilities of artificial intelligence.These observations of the educator specialize in cross-cultural studies and teaching languages other than the student’s native language may probably be reasonably interesting for AI developers and machine learning engineers.