Language models and "what there is"
Foundation models have passed a tipping point and they are cropping up everywhere, in a huge variety of use cases. We can transcribe images, translate text, do object detection, and generate anything that our hearts desire. The simplest approach to solving a problem with machine learning now almost invariably begins with “send it to Claude, OpenAI, or both and see what happens” before digging into specific model implementation details. Many natural language processing (NLP) tasks have been “solved” with this, at least in a very first pass of the data. Do you want to extract topics from a text? Do you want to summarize it? Do you want to judge whether it’s offensive or not? Send it to an LLM and log the output. Maybe later you use the output to train something specialized, but that first pass is very effective for standing up a proof of concept. ...