Use Cases of Large Language Models
Let’s experience the power of AI with generative models and explore the capabilities of models like GPT-3.
Introduction
Development in artificial intelligence gained remarkable results and improvements in recent years. Based on new deep learning methods, bigger model architectures, more available data and computational resources, researchers and engineers developed fascinating AI models able to generate texts hardly distinguishable from human-written texts, photo-realistic images, artificial video and music sequences, and even source code as the core description of software. We call these models generative AI models as those models are able to artificially generate data based on examples or even textual descriptions.
As an example, let us consider the large language model called GPT-3 that was published by the startup OpenAI in June 2020. Since then, GPT-3 has indeed changed the public opinion on what AI is able to do. GPT-3 can produce texts and even code, when given a brief prompt with instructions and descriptions of the particular task the LLM should perform. Based on the respective task, the results are then generated answers or responses given a question, or the completion of an article, essay or even joke given an initial description. And although those results are far from being perfect, people got fascinated a lot and are since then wondering what will come next. One thing became certain, the size of models matters.
Since then the development of large language models became a race regarding model size and number of developed and released models. Over the past few years, the size of large language models has increased by a factor of 10 every year. It seems to be beginning to resemble another Moore's Law . ( 1 can give you a direct impression of how LLMs scaled in past years). Along with this tremendous growth of the model sizes, the race and development also result in more and more releases of AI models globally. So far only in 2022, there have been 42 different LLMs released with impressive scaling of their training dataset even if we only focus on western countries. And in 2023 GPT-4 and more are on their way. For a more detailed overview of recently published LLMs, LifeArchitect.ai has shared a very detailed Summary of current models .
When we are curious about the usage of LLMs, then we could take a look at how many fine tuning models have been developed. And to answer this question, then one of the most famous open source platforms, Hugging Face Hub can show us the reality. Even within this single platform, over 60K models, 6K datasets, and 6K demos are open sourced in which people can easily collaborate in their machine learning workflows. And among these models, we can tell that LLM is currently definitely one of the most popular kinds of AI models. Currently 2 out of the top 3 most downloaded models are LLM, GPT-2 with 32.4 Mio. downloads and bert-base-uncased with 23 Mio. downloads.
When we focus on the new models standing out in 2022, language models are also trained on data from many other modalities (e.g., images, videos, etc.) and thus develop more diverse capabilities. Multilingual models are now very common, and programming languages have become a special language source. Models that have been combined with other modalities such as image, speech, and even videos produce a wide range of capabilities and exceptional outcomes across tasks and languages. Multilingual and multimodal model and its new potential is what we believe can benefit most to many cross-industry use cases.
Among various different AI model architectures, large language models (LLM) particularly became successful and popular over the past few years in research globally. As a result, enterprises have started to investigate the use and impact of such AI models within their products and processes. Key questions typically are the following. How does an enterprise benefit from using large language models? What are the use cases in products or internal processes for which large language models would make a difference? What can be expected as a performance in these use cases? In the following workbook, we want to give answers and provide an overview of currently known applications for large language models for generic tasks and across industries.
Considering General Tasks of LLMs
We would like to warm you up before deep diving into the industrial use cases with several most important and common tasks of LLMs. They are the foundation and key success factors for most of the industrial use cases. We collected several open source examples from OpenAI, there are of course many more successful and interesting research companies and institutes also doing researches on LLMs, we selected OpenAI because it's bride covered research topics in LLMs and also their playground provides rich examples you can get a very direct impression with easy access, thus we strongly recommend you to try it out ! Also worth to mention is the collaboration between OpenAI and Microsoft with a strategy based on Azure Cloud (i.e. AI models as a platform) also made them stay in one of the best positions of industrialization.
Click on the below tabs to learn more about the each task.
Text Generation
Text generation, formally referred to as natural language generation, aims to produce reasonable and readable text in human language based on input data, e.g. a sequence and keywords. It’s also the foundation of the question and answering system (i.e. response generation, answer generation).
The automated generation and completion of text on sentence or paragraph level is an important Natural Language Processing (NLP) task and can be widely used in many industries with shared use cases for generating articles, news, marketing slogans, social posts, protocol, and all kinds of product descriptions. This list can be extended for many more text types, therefore, text generation is an important task across industries and use cases. We recommend you to get an impression of the state of the art technology by trying it out yourself, the Generation API , it is one of the most powerful and popular software offerings of OpenAI , based on GPT-3. It can generate all kinds of content, the popular example which you can test out below is to generate new ideas , business plans, marketing slogans or even support your interview questions based on your prompt (see explanation in workbook 1). Below is an example to get you motivated.
Taking text generation as a core NLP task, many startups and software have been initialized and generate a lot of usages. For example, CopyAI is well-known as a GPT-3 powered tool to automate creativity tools and generate marketing copies within a short period of time efficiently. Businesses can use this AI model for digital ad copy, social media content, website copy, eCommerce copy, blog content, and sales copy. And Snazzy AI is known for providing the simplest way to create content with GPT-3. Multiple services include creating landing pages, copywriting, Google Ads, and many more with just three clicks.
Considering Selected Industries: Success Story & Use Cases
One of the big challenges of applying LLMs is domain adoption, for each industry there’s so many special requirements and processes and vocabularies. In the remaining part of this workbook let’s deep dive into some success stories from different industries and some real use cases where LLM has already been used and generated an impact or even become a game changing factor.
Click on the White Boxes to learn more about each use case.
Conclusion
This workbook should have helped you to answer the question we start with, about what kind of potential use cases are feasible and worth implementing using LLMs in AI use cases in enterprises across industries.
We predict that LLMs will have a game-changing impact among many products and processes in all kinds of industries. In particular, it will be a future core technology in the interaction between humans and machines or systems. However, we also believe that the rise of LLMs has just started and industrialization is still challenging in manifold directions. Much more needs to be explored better today than tomorrow in order to exploit this significant AI technology and generate added value in industry.
With the tremendous speed LLMs are growing, it might result in decreasing returns, increased expense, increased complexity, and rising risks. That is what we are going to deep dive into with the workbook 3.