THE ERA OF artificial intelligence (AI) is well underway, with increasingly more AI features becoming an integral part of our daily lives.

“Generative AI is perhaps the most well-known, thanks to its wide-ranging capabilities to generate content in various forms—and among these, ChatGPT, which generates text based on a prompt, is most widely known. But generative AI can also include content in the form of images, simulations and computer codes,” says data scientist and consultant, Poo Kuan Hong, during a Google I/O Extended event.

“Generative AI is revolutionary in that you are now no longer limited to just search engines like Google. Instead of opening articles link by link, you can now expect complete answers from the likes of ChatGPT, Bard or even Bing Chat.”

Intelligence Personalised

The game changer is in how the technology is customisable to fit our needs. Enter Large Language Model (LLM), a subset of generative AI that is trained with massive data sets from a variety of sources to understand, summarise, generate and predict new content tailored to our specific requirements. While generative AI is already impressive on its own, it is only as good as the data used to train it. This is where LLM comes in—with billions of parameters and a multitude of natural-language tasks within its capabilities, LLM is designed to work with more exhaustive platforms.

“Let’s say you own a dog. You need to train your pet to obey commands, right? LLM follows a similar concept. It can be fine-tuned with more specific data relevant to your needs or requirements,” explains Dr. Poo.

LLM can be trained using various methods of prompting. For instance, generic LLM runs on limited predictive text capabilities based on patterns and general knowledge acquired from training, while instruction-tuned models require more targeted fine-tuning, and provide predetermined responses to input instructions. Lastly, there is the dialogue-tuned model, which engages users in conversation and allows for a more complex chain of thought (CoT) reasoning.

With increased personalisation and fine-tuning, LLM—and, by extension, generative AI—has the potential to revolutionise the world of machine learning and deep learning. Its lightning-fast processing and advanced contextual learning capabilities make it a great tool for businesses to prototype and build novel proof-of-concepts, while also being able to assess technical feasibility. “The applications and potential impact are tremendous. The key is ensuring better access to AI technology and digital literacy for all. Staying abreast of all these innovations is crucial, otherwise you risk lagging behind your competition,” says Dr. Poo.

The Expanding Reach Of AI

Internet browsers are now equipped with generative AI plug-ins, while phone manufacturers have also begun integrating AI into their operating softwares (OS).

At the forefront of digital technology, Google is constantly innovating its lineup of software tools; its Pathways Language Model (PALM) comprises 540 billion parameters—significantly larger than ChatGPT 3.5’s 175 billion. Google’s version of LLM leverages on the raw power of its proprietary Tensor Processing Unit (TPU) which enhances machine learning by accelerating training processes. Meanwhile, Android 14, the latest iteration of Google’s OS, now improves on user experience, security, privacy, accessibility, customisability and core functionality through the incorporation of AI features.

With all that being said, the ongoing evolution of AI has also led to greater digital risks. The growing use of smartphones for financial transactions increases the risks of cyber crime and financial fraud, for example. “AI is the brain within not just applications, but also increasingly in the phones themselves. It can be exploited to execute identity thefts, imitation software and data poisoning,” explains software engineer, Somkiat Khitwongwattana (Ake).

Just as AI can be misused, it can also be harnessed to create a safer online experience. “Different parties have their own thoughts on what constitutes responsible use of AI, even among financial institutions and software giants such as Google. But the bottom line is, utilising AI responsibly means respecting societal norms, safeguarding against privacy leaks and fraud, and ensuring ease of use without any element of prejudice,” asserts software entrepreneur, Guan Wang.

While AI is still at its nascent stage, its potential is already staggering. With derivative technologies permeating into our daily lives, it is only a matter of time before AI becomes indispensable to society.