Large Language Models in Practice
Large language models are currently revolutionizing a lot of different domains, being able to tackle problems that were previously the exclusive domain of humans. The talk will explore the considerable potential of these models across several practical applications, but also consider potential challenges, such as their tendency to hallucinate and their limited availability for Slovak.
Large language models are changing the game with the ability to mimic style—from everyday speech to Victorian writers. It is precisely this flexibility that allows tasks to be solved even without large sets of examples. If we want to make full use of this in Slovak, we need a strong homegrown model. A breakthrough capability of today's models is that they can adopt the style and rules of a text from just a few cues. From a sentence like "I met a guy, he was average" they can turn it into the gray elegance of Victorian prose in the spirit of "he was in the way of most men". They can likewise explain technical topics in unconventional ways—for example, the principle of the DQN method in alliterative verse. Importantly, they can do this without special training for a specific task, relying only on a brief prompt or a few examples.When the model writes like a Victorian writer
