LLMs in the workflow - what the heck to call this?
07 May 2023The Importance of Naming
What do children, variables, and new products have in common? Naming them is hard.
The importance of naming cannot be overstated. A good name can make a process or product more memorable and easier to recognize, while a bad name can make it difficult to remember or even understand what the process or product does. Moreover, a well-chosen name can help to create a sense of identity and belonging among the users and developers of a process or product.
LLMs in the mix
In the case of using generative model output in products and services, I have not come across a general name for this. As the field of artificial intelligence and machine learning continues to advance, new processes and products are constantly being created to help service providers and customers. One of recent interest is the use of generative models, which can be used to create text, code, images, etc., that can be used to improved communication and customer/provider interaction.
What do we call the portion of a product or service that uses generative models? Much like we call the integration of statistics in reporting analytics or in decision contexts classification or predictive modeling/predictive analytics, we need a name for the integration of generative models within a product.
The Final Name List
I took a straw poll of the MLOps.community and from fellow AI practitioners on LinkedIn, and got some great feedback. Below are the ones I think are strong contenders.
GenAug 
GenAug, short for “generative augmentation,” is a name that accurately reflects the idea, which is to augment the delivery of service providers. Especially at their current stage, LLMs are tools, and rather than replacing the person providing the service they instead offload some of the overhead of content generation.
GenAmp 
GenAmp, short for “generative amplification,” is another name that accurately reflects the idea. This speaks to how using generative models can amplify people’s productivity. Case in point, I used ChatGPT to help get through a small logjam on this very blogpost. While I didn’t keep any of the original wording, seeing a similarly related example was enough to blast through my typical writer’s block.
GenTool 
GenTool, short for “generative tooling,” is a name that emphasizes LLM model integration as a tool. This is easy to remember and fairly general. The one fault I see is it could anchor the use of LLMs as “tools,” where in a number of cases that may be limited. For example, the sufficiency of their delivery is still questionable for a number of use cases, so calling it “tool” might mislead a user to think the answer is right when LLM hallucination is still being worked through.
GMI 
GMI, short for “generative model integration,” is a name that emphasizes the integration of generative models into a workflow. Currently, this is my preferred answer, but I’m looking forward to seeing a consensus emerge.
Final Thoughts
Naming things is hard, but it is an important task that cannot be overlooked. What do you think we should call including generative model outputs into services and products? Let me know on LinkedIn or the MLOps.Community Slack.