Cupcake or Wedding Cake?  

Imagine being responsible for picking up dessert for your three-year-old’s birthday party. Now imagine showing up, surrounded by Bluey balloons, happy children, and someone in a Bingo costume, holding a three-tiered wedding cake. Or even worse, imagine showing up to a formal wedding with a tray of Bluey cupcakes!  

Both the wedding cake and the cupcakes are delicious, but neither would really be appropriate for the occasion. Natural language processing, as a field, is just like cake. Just as there are cupcakes, sheet cakes, sculpted cakes, and extravagant wedding cakes, there are simple bag of words solutions, discriminative NLP solutions, and Large Language models…and each solves a specific type of problem. Picking the right solution is just as important as the methodology for building the solution in the first place.  

Custom-trained fine-tuned large language models are like the three-tiered wedding cake. These models are extremely powerful for situations requiring the generation of custom text, such as summarizing input data or answering custom questions. Trying to solve such complex problems with one of the more simple NLP models is showing up with those Bluey cupcakes. However, like a wedding cake, building a custom trained large language model requires custom hardware for processing such large volumes of text, so it is important to know that you NEED a large language model before beginning the process of building one.  For example, if you want to build a simple workflow of a few fixed questions and answers, a Large Language Model would certainly be overkill.  

By working with a team of Natural Language Processing experts, you can ensure that your natural language cake will be the perfect fit for your party!