Many (large majority I’m sure) of LLM applications that are cropping up are just thin wrappers on openai’s models and not terribly useful, but if you’re a devel

Many (large majority I’m sure) of LLM applications that are cropping up are just thin wrappers on openai’s models and not terribly useful, but if you’re a developer I recommend using copilot. I was skeptical of the coding value until I started getting almost magical autocomplete. Even though I validate, it definitely speeds up development, not 10x as some folks are saying but maybe more like 2x. It’s a good example of where specialized/task specific data prep and application/pipeline development with a good UX actually does add value. My pov as someone currently developing an applied ml product and having worked on ml for the last decade or so, is to think of these new LLMs as very good general purpose natural language processors for some use cases or natural language generators for others. That means your product needs to be one that can handle errors well (hallucinations as well as misclassifications or bad surface form formulations) and still be useful for customers. Many times this takes thinking through a novel approach to the UX, not just plugging in another chatbot interface. To me, it’s like the chatbot craze from a few years ago where there are definitely use cases where this works but many that simply don’t even if the model was near perfect.