A1262
Title: Representation of reality
Authors: Antal Jakovac - Wigner Research Centre for Physics (Hungary) [presenting]
Abstract: Present-day AI applications are often based on stateless models, in which the representation of the actual reality is much simpler than that of general knowledge. A typical example is large language models (LLMs), where the actual reality is encoded in the conversation history, as opposed to the hundreds of billions of parameters used to represent general knowledge. Retrieval-augmented generation (RAG) systems are being developed to overcome this limitation, but considerable effort is still needed to achieve satisfactory solutions. The aim is to overview the theoretical foundations of representing reality and propose a data storage logic that reflects these ideas.
