TDX’23 — LLMs or The Beauty of Large Language Models in Enterprise Software
Impact of Large Language Models (LLMs) such as Open AI’s GPTs on Enterprise Software: A look into the near and long-term future of conversational interfaces.
The Excitement of Large Language Models (LLMs) and How They are Changing NLP
If you’ve worked with natural language processing (NLP), you’re likely familiar with natural language understanding (NLU), named entity recognition (NER), and sentiment analysis. NLU predicts the intent behind customers’ interactions with your business, NER extracts specific data types, and sentiment analysis detects the sentiment in written text.
However, these systems come with their own set of challenges, such as training times, high-quality training data, and the need for relevant and large data sets in multiple languages. But this is changing with LLMs (Large Language Models) and there is something seen as new in the world of NLP that is causing a stir in the public sphere: Natural Language Generation (NLG).
Large language models (LLMs) represent a major development in AI, opening new possibilities for NLP. They have been dramatically increasing in size over recent years, providing pre-trained models for faster time-to-value. LLMs are global and pre-trained models that allow users to hit the ground running quickly, providing opportunities (but also threats) in this fascinating space.
In this article, we’ll overview LLMs, their breakthroughs, why they’re important, and the opportunities and challenges they bring to businesses.
Large Language Models or LLMs
LLMs are pre-trained models that can recognize relationships and dependencies between words, sentences, text passages, or documents. They are broad and pre-trained, allowing businesses to use them fast without requiring large data sets.
LLMs have been made possible by transformer models, which have gained traction because they allow language models to be trained in huge amounts of data faster. LLMs include astonishing developments such as BERT, GPTs, T5, DALL-Es, or Cohere.ai.
As a Product Manager in the realm of NLP, I couldn’t be more excited. And I believe the beauty of LLMs relies on the domain adaptation magic that can be baked in on top of these foundation models.
For example, at Salesforce’s Einstein Bots, we have launched our very own global pre-trained language model for NLU, which reduces data requirements to a one-shot training example per dialog intent class and can be trained in minutes. It also provides a range of supported languages out-of-the-box with true cross-linguality and transfer learning, making it possible to support over 100 languages for NLU.
Other successes worth underlining in my mind are of course generative pre-trained transformer models (aka GPTs) which have gone from using millions of parameters to hundreds of billions allowing these systems to deliver outstanding results.
Why Are LLMs Important?
LLMs fit nicely as general solutions that can be launched rapidly on day 0 or day 1 with limited or no customer-specific data. This is where Salesforce comes in: demystifying LLM usage, developing security from the ground up, and making LLM a successful vehicle for CRM-based customer experience.
Business Opportunities for LLMs
There are many business opportunities for LLMs, including automation, customer experience, and chatbot building. Automation can be achieved by creating NLP models that can extract specific data types or tasks from documents or data sources, reducing human effort and increasing efficiency. Customer experience can be improved by creating chatbots that can understand customers’ intent and respond with natural language.
My personal favorite application of LLMs is of course chatbot building. With LLMs, conversational interfaces will finally be put in their rightful place. They will punch UI right in the face. Traditional UIs have always been a bottleneck in the user experience, requiring users to navigate complex menus and interfaces to get what they want. Conversational interfaces, powered by LLMs, will provide a more natural and intuitive way for users to interact with technology.
Business Challenges and Risks with LLMs
Of course, there are also challenges and risks associated with LLMs, including the risk of AI bias, which could lead to misinformation and propaganda. There is also the challenge of hallucinations, which could be a problem for general audiences who would thus benefit from subject matter expert monitoring. That’s why we believe LLMs will (still) require human moderation and supervision: and that’s how we’re currently going to play this.
Despite these challenges, LLMs represent major improvements in AI and NLP, and they’re opening up new possibilities for businesses of all sizes.
What You Cannot Miss This Year at TDX’23
- Einstein Theater Session — Einstein GPT: Service Just Got A Whole Lot Smarter. Join this demo-led session to learn how Einstein GPT will change the game for customer service and help companies develop personalized responses to customer requests and resolve cases faster.
- Einstein Bots Platform breakout session: Learn how the new Einstein Bots Platform empowers developers to build innovative conversational experiences across any channel and use case, and unlock the power of automation.
- Einstein GPT Keynote: Demystifying Generative AI for CRM. Come learn how we are building and approaching generative AI at Salesforce with trust in mind. See our dev's demo on how to unlock productivity and better engage customers with Einstein GPT.
- TDX Main Keynote: Discover the latest product innovations across the Customer 360 Platform, including MuleSoft, Slack, and Tableau, and how you can use them to drive success now at your company.
I hope you’ll also find these sessions will provide you with valuable insights and learning opportunities. Don’t miss out on the chance to learn about the latest product innovations and network with other industry professionals. Register today at Trailblazer DX!
In a nutshell
Advancements in Large Language Models (LLMs) have opened up new possibilities for Natural Language Processing (NLP), enabling businesses to enhance existing capabilities and solve new tasks like Language Translation, Information Retrieval, Summarization, and Text Generation.
LLMs allow users to hit the ground running rapidly, providing opportunities in the space but also no negligent challenges.
The beauty of LLMs lies in domain adaptation which is the magic to be baked in on top of foundations. And benefits for LLMs are in essence that they lend themselves well to global solutions, which can be launched immediately with limited or no data.
LLMs also have business opportunities in Automation, Customer Experience, and my personal favorite Chatbot Building. However, LLMs do come with risks, such as AI bias and potential hallucinations. These risks make it necessary to keep a human loop. And with the proper use and human supervision, LLMs will become a successful vehicle for CRM-based customer experience.
Enjoy TDX!! Feel free to connect and reach out ⬇️
