AI decoded: what those buzzwords really mean

A large language model (LLM) is a specific type of AI system developed as part of natural language processing (NLP). While NLP encompasses all technologies that help computers understand human language, LLMs are one of its most advanced applications designed to understand and generate human-like text.

An LLM works by predicting what word should come next in a sentence. It then generates a response that sounds like a person could have written it. This capability powers AI tools such as ChatGPT and Claude, enabling them to hold conversations, answer questions and create content.

What makes these models “large” is the massive amount of text data they learn from and the billions of parameters that help them make predictions. Generally, models with more parameters can handle more complex language tasks and produce more nuanced responses.

Read more: Aradhita Parasrampuria’s mission to transform fashion: Bridging biotechnology and design for a sustainable future

Neural networks are the building blocks of modern AI and are inspired by how our brains work.

These computing systems consist of interconnected nodes organised in layers that work together to process information. Just as our brains learn from experience, neural networks learn from data, gradually improving their performance without being explicitly programmed for each task.

When information enters a neural network, it passes through layers of nodes, each analysing different aspects of the input before producing an output. This allows the networks to recognise patterns, make predictions and solve problems that traditional computing approaches struggle with—whether by identifying objects in images, translating languages or recommending movies you might enjoy.

Read more: Technology for good: Why former PR whiz Ellice Ng launched an app that empowers underprivileged kids in Malaysia

Deep learning represents the next evolution of neural networks. It uses many layers to process information with increasing levels of abstraction.

If a simple neural network is like learning to identify letters in the alphabet, deep learning is like understanding entire books—it can grasp complex concepts and nuances that simpler systems miss.

What makes deep learning powerful is its ability to automatically discover important features in raw data without human guidance. For example, when analysing images, early layers might detect edges and simple shapes, middle layers might recognise patterns like eyes or wheels, and deeper layers will identify complete objects like faces or cars.

This approach has revolutionised AI capabilities in speech recognition, image analysis and language understanding, enabling breakthroughs, such as the generating of realistic images, which seemed impossible just a decade ago.

Read more: From electricity and electrolysers, how tech leaders are transforming industries sustainably

AI hallucinations occur when AI systems generate information that sounds convincing but is actually incorrect, misleading or entirely fabricated. These aren’t perceptual errors but instances where the systems, particularly LLMs, confidently present false information as fact.

This could happen for several reasons, including gaps in the AI’s training data, misinterpreted patterns or attempts to provide answers when the system lacks sufficient knowledge.

These hallucinations highlight a limitation of current AI systems: they can be fluent without being factual, making it critical for humans to have oversight of AI-generated content.

Read more: Tatler House Dialogue: Doctor Anywhere founder Lim Wai Mun on his entrepreneurial journey and the role of technology in value creation

AI agents are digital assistants that can perform specific tasks with limited autonomy. Unlike basic chatbots that respond to prompts, agents can complete actions on your behalf.

They excel at defined workflows—think of a virtual assistant that can check your calendar, send emails or reserve a table at your favourite restaurant for you.

Agents can handle routine tasks without requiring constant human guidance. For example, it can compile information from different sources, create a summary report and email it to your team, saving you time spent on repetitive work. These tools are already enhancing productivity across many industries.

At the start of this year, OpenAI’s Sam Altman wrote in a blog post, “We believe that, in 2025, we may see the first AI agents ‘join the workforce’ and materially change the output of companies. We continue to believe that iteratively putting great tools in the hands of people leads to great, broadly-distributed outcomes.” 

Agentic AI represents a more advanced evolution where AI systems demonstrate more autonomy and reasoning.

While AI agents follow predetermined paths, agentic AI can set its own course to achieve broader goals. The key difference lies in its ability to understand context, make independent judgements and solve complex problems that arise along the way.

For instance, if you ask an agentic AI to plan your vacation, it wouldn’t just book your flights. It can also look up destinations based on your preferences, compare options, make reservations across multiple platforms and adjust plans if it encounters obstacles such as price changes or availability issues.

This represents a shift from AI that completes tasks to one that accomplishes missions, potentially transforming how we work and interact with technology.

Now, meet the Gen.T Leaders of Tomorrow from the Technology sector. 

NOW READ

AI or Human? GPT-4.5 proves it’s hard to tell the difference

From Nvidia’s droid to agile humanoids: Meet the next-gen robots shaping the future

Can AI become your closest confidant? That’s Jeanne Lim’s mission