TY - CHAP A1 - Ochoa-Luna, Jose Eduardo A1 - Meyer, Manfred ED - Carmo, Mafalda T1 - Unleashing Personalized Education Using Large Language Models in Online Collaborative Settings T2 - Education and New Developments Vol. 2 N2 - The Artificial Intelligence community has long pursued personalized education. Over the past decades, efforts have ranged from automated advisors to Intelligent Tutoring Systems, all aimed at tailoring learning experiences to students' individual needs and interests. Unfortunately, many of these endeavors remained largely theoretical or proposed solutions challenging to implement in real-world scenarios. However, we are now in the era of Large Language Models (LLMs) like ChatGPT, Mistral, or Claude, which exhibit promising capabilities with significant potential to impact personalized education. For instance, ChatGPT 4 can assist students in using the Socratic method in their learning process. Despite the immense possibilities these technologies offer, limited significant results are showcasing the impact of LLMs in educational settings. Therefore, this paper aims to present tools and strategies based on LLMs to address personalized education within online collaborative learning settings. To do so, we propose RAGs (Retrieval-Augmented Generation) agents that could be added to online collaborative learning platforms: a) the Oracle agent, capable of answering questions related to topics and materials uploaded to the platform.; b) the Summary agent, which can summarize and present content based on students' profiles.; c) the Socratic agent, guiding students in learning topics through close interaction.; d) the Forum agent, analyzing students' forum posts to identify challenging topics and suggest ways to overcome difficulties or foster peer collaboration.; e) the Assessment agent, presenting personalized challenges based on students' needs. f) the Proactive agent, analyzing student activity and suggesting learning paths as needed. Importantly, each RAG agent can leverage historical student data to personalize the learning experience effectively. To assess the effectiveness of this personalized approach, we plan to evaluate the use of RAGs in online collaborative learning platforms compared to previous online learning courses conducted in previous years. KW - Collaborative Learning KW - Generative AI KW - Large Language Model KW - Personalized Education Y1 - 2024 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:hbz:1010-opus4-46838 SN - 2184-044X SP - 259 EP - 263 PB - inScience Press CY - Lissabon ER - TY - THES A1 - Arping, Bastian T1 - Dark LLMs im Kontext von OSINT: Prototypentwicklung und Analyse der Anwendung für die polizeiliche Gefahrenabwehr und Strafverfolgung in Nordrhein-Westfalen N2 - In dieser Arbeit wird der Einsatz von Dark LLMs zur Durchführung von OSINT-Recherchen im Rahmen der polizeilichen Gefahrenabwehr und Strafverfolgung in Nordrhein-Westfalen analysiert. Hierfür wird zunächst eine erste wissenschaftliche Definition des Begriffs Dark LLM erstellt. Anschließend werden relevante Dark LLMs vorgestellt, auf deren Gefahren aufmerksam gemacht und geeignete Gegenmaßnahmen erörtert. Danach erfolgt eine rechtliche Einordnung anhand mehrerer Gesetzestexte. Dann wird eine prototypische Entwicklung eines LLMs durchgeführt, welches für OSINT-Recherchen eingesetzt werden kann. Die rechtliche Analyse und die technische Implementierung ergeben, dass der Einsatz eines Dark LLMs in der polizeilichen Arbeit unzulässig ist. Zudem wird ein polizeiliches LLM nach der KI-Verordnung sowohl als Hochrisiko-KI-System als auch als KI-Modell mit allgemeinem Verwendungszweck eingestuft, wodurch sich etliche Pflichten ergeben, die ein LLM in ihrer Funktion einschränken. Trotz dieser Einschränkungen ist ein Einsatz eines solchen LLMs sinnvoll, denn dieses kann viele OSINT-Informationen sekundenschnell zu einer gebündelten Antwort zusammenfassen, was die polizeiliche Arbeit erheblich erleichtert. Durch allgemeines OSINT-Training und die Anbindung weiterer OSINT-Quellen kann die Qualität der Antwort deutlich verbessert werden. N2 - This paper analyses the use of dark LLMs to conduct OSINT research in the context of police hazard prevention and law enforcement in North Rhine-Westphalia. To this end, an initial scientific definition of the term dark LLM is first established. Subsequently, relevant dark LLMs are presented, their dangers are highlighted, and appropriate countermeasures are discussed. This is followed by a legal classification based on several legal texts. Then a prototype LLM is developed that can be used for OSINT research. The legal analysis and technical implementation show that the use of a dark LLM in police work is not permitted. In addition, according to the AI Regulation, a police LLM is classified as both a high-risk AI system and an AI model for general use, which results in a number of obligations that restrict the function of an LLM. Despite these restrictions, the use of such an LLM is useful because it can summarise a lot of OSINT information into a bundled response in a matter of seconds, which makes police work considerably easier. Through general OSINT training and the connection of additional OSINT sources, the quality of the response can be significantly improved. KW - Künstliche Intelligenz KW - Große Sprachmodelle KW - Large Language Model KW - Open Source Intelligence KW - Cyberkriminalität KW - Künstliche Intelligenz KW - Dark LLM KW - OSINT KW - Polizei KW - Große Sprachmodelle Y1 - 2025 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:hbz:1010-opus4-48354 ER -