03-07-2025
IBM's new tool speeds up work with factual guardrails, stops AI errors
IBM's Generative AI platform uses an open-source approach that can help cut your business costs, adapt to changing situations and keep data secure
The AI-powered concierge robots at Bird Paradise at Mandai Wildlife Reserves use IBM's Granite small language learning models to provide information such as weather updates and customised itineraries to visitors.
'Would you prefer to see the orangutans or the white tigers first?' asks the female avatar, beaming out of a big screen mounted on wheels. If you only have a few hours to spare when visiting the Singapore Zoo, Mandy the concierge robot can help you design an instant zoo itinerary.
Mandy is one of the Autonomous Mobile Robots being trialled by the Mandai Wildlife Group. She can help you with directions, provide fun facts about the zoo's animals, or simply lead you to where you want to go. If dark clouds gather, she will pull live weather data to suggest a wet weather plan instead.
Her intelligence is powered by
IBM's data and artificial intelligence (AI) platform known as watsonx . Trained on Mandai Wildlife Group's data, Mandy can personalise responses to suit every visitor's needs and even react to facial expressions.
IBM's cutting edge Generative AI solution delivers the responsive and flexible customer experience that Mandai Wildlife Group needs. IBM data scientist Ang Ai Kiar says: 'We uniquely leveraged our Generative AI capabilities to set us apart, showcasing how Mandy can create novel responses on the spot, rather than being limited to pre-programmed answers.
A Gen AI-powered concierge may seem straightforward to implement, but turning AI ambitions into working solutions is a real challenge for most enterprises.
Large language models (LLMs) may have taken the world by storm, but they are not always a good fit for business. For instance, they might come with high costs, security concerns and limited customisation options for specific business needs.
Ms Catherine Lian, general manager and technology leader, IBM ASEAN, says that smaller, industry-tailored AI models can offer greater benefits for businesses: 'Many organisations are realising that large, general-purpose AI models require substantial computing resources, making it a challenge to scale their use affordably.'
Beyond the resource challenges, companies are also concerned about data security and the tendency of Gen AI to 'hallucinate' or make up false answers.
Despite these teething pains, companies see real productivity gains from Gen AI. According to an IBM survey,
57 per cent of Singapore organisations plan to increase AI investment in 2025 . The biggest investments will be in areas such as IT operations, data quality management and finance operations.
Open and nimble solutions
To address these concerns about resource requirements, data security and AI accuracy, IBM has taken a different path from other AI companies – one that emphasises openness, transparency and cost efficiency.
For starters, IBM's watsonx platform uses open-source AI models known as small language models (SLMs) that can save up to 98.5 per cent in costs. Unlike closed-source AI models that operate as 'black boxes' with proprietary code and training data that users cannot examine, open-source SLMs provide much greater transparency.
Another application of IBM's Granite small language models is the Ferrari Scuderia HP mobile app, which transforms complex race data into fan-friendly insights like overtaking analysis.
PHOTO: IBM
Organisations can inspect the code, understand how the model works, and even modify it to suit their business needs. This open-source approach allows business to maintain control and transparency over their AI models while leveraging the watsonx platform for enterprise-grade hosting, updates and support services.
Open-source models often become more accurate and reliable over time because developers from global communities can spot mistakes, fix technical problems and add new features. For businesses, this means getting new features more quickly – they are not stuck with one vendor's technology and are able to customise AI capabilities for their customers.
According to a survey of over 2,400 IT decision makers across 21 countries,
51 per cent of companies using open-source AI tools report positive returns on investment (ROI) compared to just 41 per cent of those using closed-source systems.
Expanding on its commitment to open-source AI,
IBM released the latest version of its Granite model family, Granite 3.2, in February 2025. Granite is IBM's collection of SLMs that power the watsonx platform.
The new Granite 3.2 models can match or exceed the performance of larger models while using less power. They include specialised models for different tasks for financial forecasting up to two years ahead, supply chain planning and inventory management. Businesses can access these Granite models or integrate other popular models such as Meta's Llama and Mistral, through IBM's platform.
Granite's strengths shine in environments like the Singapore Zoo and other wildlife parks. Robots powered by Granite SLMs don't require a constant internet connection – they can operate offline while using minimal power. This allows them to serve visitors for longer hours across the entire compound. Their flexibility also enables them to perform specific tasks such as tracking animal movements, identifying species, and monitoring environmental conditions.
Ms Ang says: 'Granite SLMs offer speed, cost efficiency, and adaptability, making them ideal for ongoing operations in resource-constrained environments like wildlife reserves.'
Granite SLMs are also powering the
new Scuderia Ferrari HP mobile app, bringing F1 fans closer than ever to their favourite team . Fans can now read race recaps within hours of a race's conclusion. Dynamic visuals, created using AI technologies on watsonx, let them interact with post-race data – including telemetry, weather, track conditions, session results and car and tyre strategies.
How IBM's open-source AI models are more cost-efficient
Cost comparison: 98.5 per cent lower cost than closed-source AI models
51 per cent of open-source users report positive ROI (vs. 41 per cent for closed-source models)
Technical benefits: Require less computing power while matching performance of larger models, reducing operational costs
Lower development costs through community-driven improvements rather than proprietary development
Business flexibility:
Eliminate vendor lock-in expenses, allowing more flexible and economical scaling
Leveraging innovation with trust and governance at its core
While IBM's open-source approach offers many technical and cost advantages, the company recognises that technology alone is not enough. As Gen AI becomes more powerful, businesses are increasingly concerned about trust, safety and control. While some AI companies have downsized their ethics teams, IBM has doubled down on responsible AI.
This approach to responsible AI isn't just a technical consideration for IBM, but central to their vision for how AI should be developed and deployed.
Ms Lian says: 'At IBM, we find purpose in bringing breakthrough innovations in AI. While AI is a profound opportunity, it comes with high stakes.
'Imagine if your organisation's Generative AI compromises the personal data of customers or defrauds a customer in an interaction. This simply cannot be happening.'
As Gen AI both disrupts and augments the enterprise, AI-powered assistants like Mandai Wildlife Group's Mandy show how businesses can already benefit from using IBM's agile and ethical framework.
