Or 20s of the 21st century. Or just this year...
Large Language Models have already changed many aspects of our lives. However, training large models demands significant resources, including data, computing power, and specialized expertise, which only a handful of organizations can afford. This challenge has given rise to model-as-a-service offerings, where models are made available for others to be uses as a service. As a result, individuals and businesses looking to integrate AI into their applications can now do so with minimal upfront investment.
As Chip Huyen points out in her recent book: "AI engineering -- the process of building applications with readily available models -- into one of the fastest-growing engineering disciplines. "
AI engineers do not need to train the model from scratch. Instead, they should be proficient in prompt engineering and understand when to fine-tune the model and when retrieval augmented generation (RAG) is a better option. The LLM output is non-deterministic, and sometimes small changes in the prompt can significantly change the cost of model-as-a-service. Therefore building a robust system based on it, which does not create astronomical bills, is a special kind of Art. Another important skill for AI Engineers in Europe is fluency in the local language(s) (the languages of the countries where your company operates). Although we as a community put much effort into making an evaluation of LLM-based solutions more automated, the most reliable debugging is still to read the output yourself.
Comments