Why We Need to Relearn How to Talk to Machines – A Snapshot of Generative AI in January 2024

The last few years have seen incredibly rapid progress in the field of generative artificial intelligence. Talking to machines and getting answers in natural language is part of our new, elusive normal. Driven by the exponential growth of both computing power and internet-scale data, our new digital assistants are trained by estimating the most likely next element of a given context. Recent years have clearly shown that this general objective can lead to the ability to develop complex and diverse capabilities from simple principles. At the same time, however, it can lead to interesting structures in the compression of the training data and sometimes to unpredictable artefacts. The aim of this article is to shed light on the mechanisms behind current large language models and to provide guidance on how to get the best answers to a question. The original version in German, “Warum wir neu lernen müssen, mit Maschinen zu sprechen – eine Momentaufnahme der Generativen KI im Januar 2024“ was published in Ordnung der Wissenschaft in February 2024 (https://ordnungderwissenschaft.de/wp-content/uploads/2024/03/Kalweit-Druckfahne-V4.pdf).