Large language models (LLMs) are computer programs that can generate text and ideas in ways that sound convincingly human and natural. They form the basis of many generative AI systems thanks to their ability to ‘interpret’ and then respond to user prompts. They are suited to text-based tasks such as answering questions, summarization and text generation.
Because of their impressive capabilities, LLMs are being used in a variety of applications, ranging from search to chatbots. They offer new ways of discovering information and of creating new (text-based) content.
A technology trained on vast amounts of text data that can offer human-like responses to prompts.
LLMs can not only make it easier to discover information and create new content, they can also produce novel ideas.
LLMs are prone to errors — that means it’s worth being cautious about moving from experimentation to production.
They’re being used to help users find information and to perform content generation tasks.
What are large language models (LLMs)?
Large language models (LLMs) are a type of generative AI — they’re like super-powered language learners. Trained on massive amounts of text data, they can seemingly understand and respond to questions and in ways that are not only very convincing and human-like, but also novel.
Imagine a program that reads millions of books and can chat, translate languages or even write different kinds of creative content — that’s essentially what a large language model is.
What’s in it for you?
Large Language Models offer many advantages for businesses. They can:
Automate routine tasks, giving people and teams more time to focus on complex and value-adding work
Aid creativity, helping teams generate content and ideas quickly
Improve information discovery and customer support experiences
What are the trade-offs of LLMs?
Despite the advantages, there are a few drawbacks to consider:
LLMs are trained on massive amounts of data, which can include biases or inaccuracies. Businesses need to be cautious about potential misinformation and ensure data quality.
While LLMs can mimic human-like text, they might struggle with true creative thinking or complex reasoning tasks.
It can be challenging to understand how LLMs arrive at their outputs. This lack of transparency can be a concern for tasks that require clear reasoning.
LLMs can hallucinate — they provide convincing outputs, not necessarily accurate or correct ones.
Training and running LLMs requires significant computing power, which can translate to higher costs.
In Technology Radar Vol.30 we warned against what we describe as overenthusiastic LLM use.
How are LLMs being used?
LLMs are being used in many different areas. These include improving customer experience — where they power chatbots that can better handle queries — and rapid and more personalized content creation. They can also be used in code generation, potentially speeding up the development of software, translation and even scientific research, where they can be used to analyze or summarize large amounts of data or information.
We used LLMs when working with recruitment organization Bolt.Works to augment their staff’s workflow by transforming vast amounts of unstructured text data into a more structured format. Within Thoughtworks, we’ve been experimenting with LLMs to help with strategy ideation.
