If we analyze recent publications (e.g., Karen Hao's book) showing how Large Language Models (LLMs) were trained (AI Agents use these models), we may have doubts when it comes to their quality and real possibilities. In this case, I am thinking about the data - its organization and the value it holds. On the other hand, in the paper published in Nature (2026) the Authors showed how easily LLMs can be manipulated. Even these two situations can raise an alarm in our heads. The main question is - can we trust LLMs and AI Agents? In this presentation, I will not only give an answer to the previously asked question but also show how can we improve (adjust) LLMs (and by that, AI Agents) and introduce (based on the concept described by Nvidia in 2025) the idea of expert models that can be a real answer to the issues with LLMs.
Maciej Szymkowski is a researcher in the field of Artificial Intelligence and Machine Learning. His main research areas include reasoning models, explainable and trustworthy AI as well as mathematical foundations of ML and AI. He is affiliated with T-Labs (T-Mobile Poland) where he works as a Chief AI Researcher (from July 2026). Maciej holds a Ph.D. in the field of Computer Science which was awarded by his alma mater – Bialystok University of Technology, Faculty of Computer Science (2026). The topic of his dissertation was trimodal biometrics system powered by AI (three traits from a human finger were used, Maciej build both a device and AI models to recognize a user). He was affiliated with Bialystok University of Technology, Faculty of Computer Science (2018-2026) but also with two other universities in Poland – Warsaw University of Technology (2021-2022, Faculty of Electronics and Information Technology) and with AGH in Cracow (2021-2022, Faculty of Physics and Applied Computer Science). Between 2022 and 2024 he was with Łukasiewicz Research Network – Poznań Insitute of Technology where he was Senior AI Researcher and Head of AI development team. Maciej, summarized (until now) his research work with more than 45 research papers (published in JCR journals, as chapters in the books and in conference proceedings). He also participated in several projects funded by European Commission and by National Centre for Research and Development in Poland. Maciej has also experiences in working outside of academia. He was a software developer (he started with Java as his main language but then moved to Python), data scientist, machine learning engineer, and AI engineer. Recently he was with Future Processing, Hemolens Diagnostics, and SoftServe. He is working with different companies from 2016.
