• Monday, July 15, 2024
businessday logo


Explainer: The role of large language models in creation of ChatGPT, others

Explainer: The role of large language models in creation of ChatGPT, others

Generative Artificial Intelligence (GenAI) tools like ChatGPT and Google’s Gemini rely heavily on Large Language Models (LLMs). These LLMs are AI programs designed to recognise and generate text, among other tasks.

They are termed ‘large’ due to their training on extensive datasets. Simply put, LLMs use a type of machine learning called deep learning to understand the structure and function of characters, words, and sentences.

According to a study by OpenAI, some LLMs used by GenAI are trained on datasets exceeding 1.5 trillion words. Although these datasets are gathered from the internet, the quality of the samples impacts how well LLMs learn natural language.

Read also: Artificial Intelligence in Nigeria: How ready are we?

Over the years, humans have developed spoken languages to communicate. Just as language is at the core of all forms of human and technological communication, language models serve a similar purpose, providing a basis for communication and generating new concepts in the AI world.

LLMs can be trained to perform several tasks. One of the most well-known uses is their application in GenAI. They can produce text replies when given a prompt or question, as seen with tools like Chat GPT and Google’s Gemini.

These platforms use LLMs for text and similar models for images, music, and other media. LLMs focus on processing and comprehending human language, enabling the creation of never-before-seen content, including images, audio, and text, which can enhance content quality and promote sales.

How LLMs work

LLMs begin by absorbing a large amount of data, often reaching petabytes in size. This data corpus forms the foundation for their learning. They identify patterns and connections between words and connections between words and concepts from vast amounts of unlabeled and unstructured data through unsupervised learning.

Some LLMs advance through supervised learning, where data is carefully labelled to provide clear examples. This structured approach allows the LLM to receive corrections and guidance, enhancing its understanding of various ideas more precisely.

Once foundational training is complete, LLMs undergo deep learning using a transformer neural network. This powerful tool enables the LLM to focus on specific parts of a sentence and analyse the relationships and connections between words.

Read also:Automation, artificial intelligence becoming affordable ways to achieve marketing goals Goodnews

After training, LLMs emerge as highly skilled language processors. Users can leverage the LLM’s capabilities by providing prompts. LLMs can answer questions, summarise long texts, and analyse the emotions in a write-up based on users’ instructions.

The development of LLMs is a significant milestone in AI development. As LLMs continue to be refined and trained on even larger datasets, their communication and creative content generation capabilities will grow. This technology holds immense potential to revolutionise various fields, from education and customer service to entertainment and scientific research.

However, addressing ethical considerations surrounding LLMs is crucial, especially around data privacy. Issues of bias and fairness arise because LLMs often reflect the biases in their training data. Recently, countries like Nigeria have launched LLMs to help AI models better understand their languages.