LLM-y dla sceptyków - co możesz z tego wziąć? - Jakub Mrugalski

Will artificial intelligence replace programmers? Jakub Mrugalski debunks myths about the LLM, explains how AI can push the boundaries of work efficiency and support, not replace, people. Discover practical implementation examples that are transforming everyday work and learn how to start using vector databases and fine tuning! ➡️ Infoshare Katowice 2025 is the largest technology conference in Silesia, where business meets the latest technologies – learn about innovative tools, participate in workshops, and hear from industry experts. Join over 1,000 participants, build lasting business relationships, and develop your skills with lectures from leaders in IT, startups, and marketing. 📅 November 24-25, 2025 - MCK Katowice 🔗 Info: https://katowice.infoshare.pl/ 🤖 AI Summary Main topics and conclusions from the lecture: Why many programmers are skeptical of AI   • LLMs often make mistakes, are slower than traditional code, and sometimes "talk nonsense"   • This is natural and results from their design and testing applications   • AI does not replace code, but can shift the so-called "Zdzisław's boundary" – places where human intelligence is needed, automating certain repetitive tasks Examples of AI applications in practice   • Automating the work of non-technical employees, streamlining processes such as ticket categorization and customer service   • Content moderation – AI recognizes the context and intent of statements, a place where traditional regexes fail   • Advanced information retrieval using vector databases and content vectorization   • Anomaly detection and decision automation based on historical data and model fine-tuning How to cope with AI challenges   • LLM hallucinations – i.e., the generation of false information – do occur, but they can be limited by controlling input data and dynamically adjusting prompts (e.g., RAG – retrieval augmented generation)   • Using fine-tuning to train a model on your own data, allowing you to replicate your experience and work style   • Properly building prompts and allowing the model to "solve problems step by step" increases the accuracy of results Practical tips for developers and companies:   • Start by streamlining internal processes, not immediately with customer-facing AI   • Focus on supporting employees, e.g., with automated draft responses that are approved by a human   • Don't be afraid of new technologies, even though they may seem "magical" and difficult to explain   • Use ready-made tools, such as the OpenAI Moderation API – a free and effective solution for quick content moderation   • Consider using local AI models if you want to avoid the cloud and care about data privacy If you want to start implementing AI in-house, focus on: Identifying repetitive and time-consuming tasks – automate them to free up people for more creative work Combining AI with human supervision – use models as support tools, not replacements Experimenting with vector databases and prompt engineering to achieve the best results Using available APIs for content moderation and classification to quickly and affordably improve your Systems 🤖 End of summary 🔽 Follow us! Website: https://infoshare.pl/ Linkedin:   / infoshare-pl   Facebook:   / infosharepl   Instagram:   / infoshare   Don't forget to like, comment, and subscribe to receive more valuable content! #Infoshare