BENGALURU: Small language models (SLMs) are more cost-effective and the inference cost is much less than large language models (LLMs), says Balakrishna D R, aka Bali, executive vice-president and global services head of AI and industry verticals at Infosys. In an interaction with TNIE, Bali spoke about DeepSeek, Agentic AI, and the need to upskill employees in AI. Excerpts:
What are small language models (SLMs) and what are their advantages?
Small is the new big. SLMs are more cost effective. The inference cost is much less than a large language model. Also, they can be fine-tuned for a specific context. Large language models usually are trained with all the data that is available in public forums. This can be fine-tuned for the specific data that you may have within your organisation as well. The other big advantage is that you can host it within your organisation firewall so that you can ensure secure handling of proprietary or sensitive data without having to go outside your enterprise boundaries.
Can you explain its use cases?
We see multiple use cases like monitoring systems where you can detect unusual log patterns and then you can resolve that content generation, summarization, translation tools, content moderation, and sentiment analysis. We are also seeing domain-specific areas where you are creating SLMs for a specific domain. In Infosys, we went about creating a small language model from scratch. We created what we call a made in Infosys Enterprise SLM. We further trained it with banking information on what is available, publicly available data. We trained it on the banking domain and then we further actually tuned it with data that we have in our Finacle product that we own. A lot of our clients also have an interest in using their proprietary data and creating small language models for their domains, whether it is banking or telco clients, etc., they are all interested to create that.
What kind of value can Agentic AI create for your clients?
Agentic AI is a big part of our conversations across the world right now with all our customers. The power of Agentic AI is its reasoning capabilities. It goes from a general-purpose tool to creating specialised agents, assistants, and collaborators. Through this, you can eliminate tasks and reengineer processes. Humans in the loop can have agents perform many tasks.
Everybody is now talking about DeepSeek and low-cost AI models. Does cost play an important role here?
Cost does come into play because it is all about ROI (return on investment). So, identifying the use cases that give you the big benefit contributes to get a better ROI and if you can reduce the cost then it also makes the ROI better and you can take up more use cases. The benefit and the cost both of them are actually important and now with the innovation that is happening, costs are only actually reducing year on year and now with DeepSeek, The expectation is that it will actually reduce even further. We are creating small language models and are looking at how we can use some of these innovations in the way that we create. The small language models bring down the cost. AI with a reduced cost is only good news for companies similar to us in the IT industry because now the business case becomes even better and you are able to take up much more.
Currently, if a fresher wants to join the IT industry, what kind of skill sets he/she should have?
Anybody who is coming into the workforce has to be AI literate. It doesn’t matter what specific technology they're working on. They should have skills like prompt engineering, should be able to use AI coding assistants and that is what we also train when we actually take people from the colleges. This is at the basic level. Then we have what we call AI builders and AI masters.
AI builders are the next level where they have higher skills where they can actually create AI solutions and can do things like semantic search, responsible AI, etc. AI masters are people with specialised skills. These experts can create things, optimise the model, fine-tune and deploy the model. Freshers joining the workforce need to be aware of AI. If you want to create a career in AI itself, then of course you have to have much deeper skills.