While large language models (LLMs) and generative AI have been all the rage over the past year, the most attention has been given to their intersection – the text generation capabilities of LLMs. There is no doubt that the ability to generate answers to questions is a major value proposition of LLMs. However, there are other uses of LLMs that are both common and valuable. This blog will discuss a few primary uses of LLMs to ensure that you don’t fall into the trap of considering them exclusively for generative purposes.
Creation / Generation
This is the LLM use case that gets most of the attention these days. You ask an application like ChatGPT a question and it comes back with a detailed answer. Or, you provide a request to an application like DALL-E and it generates an image based on that request. There are also generators focused on code, video, and 3D virtual worlds.
The interesting thing to me is that many of the same fundamental algorithmic approaches are utilized for generators of all types. The content that is provided back – text, pictures, videos – varies. Since they all ingest a prompt, however, they must all be trained to understand and decompose that prompt to guide the generation process. Hence, they all need LLMs. But generation of new content to answer a question, while what most people focus on, is not all LLMs can do.
Summarization
LLMs are also terrific at summarizing information that you provide them. Perhaps there is a list of papers on your to-read list. It can be hard to get motivated to start working through them. One way to start is to feed the papers into an LLM and ask the LLM to summarize their key themes and to identify what points the papers appear to have in common and where they differ. Having that as a baseline, you can begin with some clear ideas about what to focus on while you read the papers.
One of the advantages of using AI to summarize content is that the risk of errors is lower than with generation. The reason is because you are limiting the LLM to taking what you gave it and summarizing it rather than asking it to come up with new content. While it is possible the LLM could focus on the wrong things or miss a pattern in your inputs, it is unlikely that it will get something completely wrong.
Translation
Translation, though often underrated, might have some of the broadest applicability and impact. For example, LLMs are already being used to help translate old code from now-uncommon languages into modern coding languages. An LLM can take the old code and generate a draft of how that would translate in the new coding language. Of course, it won’t be perfect and will take some human editing to complete the job. If the LLM gets the new code “mostly right”, a good programmer will be able to understand what the code is aiming to do and make the edits required to finish the translation – even with limited knowledge of the original language.
Human language translation will also have huge impacts. Very soon, we’ll be able to talk to anyone in the world in our preferred language and have what we say translated instantly into whatever language the person we’re speaking with prefers. We will no longer need to learn a common language to communicate. This will also be beneficial for keeping uncommon languages alive because there will no longer be a large communication “penalty” due to the lack of people who know the language.
Interpretation / Extraction
Another key use of LLMs is having them interpret a statement and then use that interpretation to cause additional actions to be taken. Image generators make use of this approach. Another example is asking an analytical question in plain language, having an LLM extract the intent of the question, and then passing that information to a query generator. For example, I ask “Please summarize this year’s sales by region and subtotal by product.” An LLM can interpret that request, extract key parameters from it, and feed those to a query generator to get me my answer. I’m advising one company, Quaeris, that focuses on this.
LLMs can also help with classic use cases such as sentiment analysis. Similarly, customer service inquiries can be ingested and then various facts about each inquiry can be extracted. For example, what product am I asking about? What issue am I raising? What action am I requesting? From there, I can be more effectively routed to the person who can best help.
Wrap-Up
The topics covered above are certainly not an exhaustive list of all that LLMs can do, but they do represent some common and powerful uses. Moreover, they should be enough to reinforce the point of this blog, which is that LLMs can do a lot more than just generation of text content. Don’t neglect to explore how those other uses might be of benefit to you and your organization!
Originally posted in the Analytics Matters newsletter on LinkedIn
The post LLMs Are More Than Generative AI appeared on Datafloq.
