by Andhie Wong
In the fast-evolving world of artificial intelligence, focus is shifting from large, resource-intensive language models (LLMs) toward smaller, more efficient language models (SLMs). That shift is set to democratise AI, putting advanced capability within reach of a much broader range of businesses and industries. OpenAI's recent launch of GPT-4o Mini is a good example of the trend — a more affordable, more efficient alternative to its larger models.
Why choose a smaller model?
Cost-effectiveness. Smaller models are cheaper to train and maintain, putting them within reach of startups and small businesses working with limited budgets — lower operational cost, and a smaller environmental footprint thanks to reduced power consumption.
Specialised precision. Unlike larger models, SLMs can be fine-tuned for specific tasks or industries — legal or medical applications, for instance — producing more relevant, more accurate output than a generalised model would.
Faster response times. Smaller models run with lower latency, making them well suited to real-time applications like chatbots and virtual assistants, and a better customer-service experience as a result.
Where SLMs are already a good fit
Healthcare: SLMs can be fine-tuned on medical datasets, letting providers generate accurate patient responses and manage records efficiently. Their grasp of specialised terminology makes them genuinely valuable here.
Finance: financial institutions use SLMs to analyse contracts and assess risk. Their targeted training helps ensure regulatory compliance while keeping costs well below what a larger model would demand.
Customer support: many companies are deploying SLMs in chatbots to handle customer queries — their fast response times and low latency lift the customer experience while letting businesses support customers more efficiently.
Education: educational institutions are using SLMs to build personalised learning experiences tailored to individual students, lifting both engagement and outcomes.
Self-hosting: the real differentiator
One of SLMs' biggest advantages is that they can be self-hosted — a critical capability for organisations concerned about data sovereignty and privacy. By running SLMs on their own servers, businesses can keep sensitive data fully within their own control, cutting the risk that comes with relying on third-party cloud services — especially important in sectors like healthcare and finance, where data-privacy regulation is strict.
Self-hosting also lets organisations customise their models further, tailoring them to specific operational needs without compromising security. That level of control over data doesn't just strengthen privacy — it builds real trust with customers increasingly concerned about how their information gets handled.
A natural fit for DTaaS
SLMs' self-hosting capability lines up neatly with the growing trend toward Digital Transformation as a Service (DTaaS). In Malaysia, companies like Agmo and SNS Network are among the first offering DTaaS solutions, letting local businesses integrate these models without major infrastructure overhauls — a smoother path into more advanced technology.
Closing
Used as part of a broader digital transformation strategy, SLMs can lift operational efficiency, improve customer interaction, and drive genuine innovation — fostering agility and helping businesses adapt faster to shifting market demand. Done well, this approach lets companies strengthen their operations while navigating the complexity of digital transformation, paving the way toward a more innovative, more secure AI future.
Part of the AK AI Corner column. Originally published in Oriental Daily (东方日报) on Jul 24, 2024.