Three Ways Your Business Can Get Started With LLMs
Generative AI is having its moment. No surprise then that the question I tend to get asked most these days is: what can GenAI do for my business, and how do I get started?
If you would like to learn what AI solutions could mean for your business, please do not hesitate to get in touch, or read our latest research report: "The GenAI Pulse Check: Are We Ready for the New AI-Enabled Age?"
Generative AI is having its moment. No surprise then that the question I tend to get asked most these days is: what can GenAI do for my business, and how do I get started? Usually there is a note of trepidation in people's voice and, to some degree, rightly so. As I wrote about before, large scale deployments of GenAI are not a trivial exercise, particularly when embedded in automated operations and processes. However, that does not preclude organisations getting started with some safe and simple use cases that can unlock a lot of value.
Area One: Human in the Loop Tasks
Human in the loop tasks, or activities where GenAI plays a supporting role to a human operator, have three major advantages. Firstly, from a technical perspective, there is no sophisticated integration or automation required. After all, every major Large Language Model (LLM) – whether from Google, Anthropic, or OpenAI – has chat interfaces available that can be used. Secondly, human supervision and scrutiny significantly reduce the risk of erroneous outputs or hallucinations. Thirdly, they allow for easy tweaking and customisation.
One specific use case is that of copy editing. LLMs can give detailed feedback to improve writing and, best of all, do so within a particular context. This is where prompting comes in. For example, compare the following two prompts and responses from Google's Bard when fed the first paragraph of this article:
PROMPT: You are a secondary school English teacher. Please critique the following paragraph through the lens of creative writing.
RESPONSE: The paragraph is well-written and informative, but it lacks the engaging and evocative language that one would expect from a piece of creative writing. The author's use of passive voice and technical jargon makes the text feel dry and impersonal. [continues...]
Ouch. Let us now try a different review, via the following prompt:
PROMPT: Please critique the same text, but now as a copy editor for the Harvard Business Review.
RESPONSE: The paragraph provides a concise overview of the growing interest in Generative AI (GenAI) and highlights the need for businesses to explore its potential benefits. However, the paragraph could be strengthened by providing more concrete examples and case studies to illustrate the practical applications of GenAI. [continues...]
Much better. While it might not replace professional copy editors at this point in time, the power of LLMs to increase the quality of human work should not be underestimated. Mind you, that also depends on the willingness of the human to accept and incorporate said feedback – not always easy!
Translating Languages and Formats
Given the sheer breadth of content, format, and languages modern LLMs have been trained on, they have a useful ability to translate content from one format or language to another. For example, not only can we use LLMs to translate a piece of code, but we can even translate to code using natural language. For example, here we are asking a model to create a website with a simple calculator.
PROMPT: Please generate the code for a webpage with a form consisting of two boxes and a button. The button, when pressed, should add the numeric values in the two boxes together and print the output below. Add a radio button to toggle between addition and subtraction. Oh, and make it look nice.
Once the resulting HTML is pasted into a text document using Notepad and given an .html extension, the following webpage is the result. Impressive!
Non-Factual Content Generation
While occasionally disparaged as probabilistic parrots, the ability of LLMs to generate copious volumes of text based on limited prompting is a boon in creative processes. While models can confidently generate false facts or contradictory outputs, in many cases this need not be an obstacle. Online creative content generally falls into three main categories: entertainment, inspiration, and education. The first two categories do not necessarily require accurate information. In other words, LLMs' outputs can do the job just fine!
PROMPT: You are a seasoned kindergarten teacher. Please tell a short story about a girl named Xiomara and an army of mice, suitable for a five year old.
RESPONSE: Once upon a time, in a cozy little cottage nestled amidst rolling hills and lush meadows, there lived a curious young girl named Xiomara. With her bright, inquisitive eyes and a heart brimming with kindness, Xiomara was known for her boundless imagination and her love for all creatures, great and small. [continues...]
While I do not want to spoil the ending, suffice to say that our heroine becomes fast friends with the mice. In the words of Bard: "Their friendship blossomed, a bond between two different species, united by their shared sense of curiosity and love for adventure." Who cares about facts when what we often crave is fiction. While there are ways of greatly improving the factual accuracy of responses, e.g. RAG or multi-agent LLMs, we will save these for a future edition.
While I could go on and on – and I am tempted to given how many fun examples there are! – it is enough to say that there is a lot of value here to be tapped. Many organisations have understandable concerns around the application and operationalisation of artificial intelligence and large language models in particular. However, as I have hopefully demonstrated, these concerns should not preclude them exploring and unlocking these powerful new capabilities.
– Ryan
If you would like to learn what AI solutions could mean for your business, please do not hesitate to get in touch, or read our latest research report: "The GenAI Pulse Check: Are We Ready for the New AI-Enabled Age?"
Cover image by Shubham Dhage.
Q* - Qstar.ai Newsletter
Join the newsletter to receive the latest updates in your inbox.