DailyPost 3027
KHANMIGO – THE SLM WAY!
Whatever may be the pace of technology, the leaders who decide to remain relevant remain so. General purpose technology is put to proper use. Khan Academy showed the way for democratization of quality education, globally, which was unimaginable to that day. While we neck deep in understanding AI’s large language model, LLMs as we call them, the innovators have way around the complexity of it, and gone ahead with the task of using AI the way it could suit them best. One of the shining examples is Khanmigo, an AI driven educational assistant developed by Khan Academy.
What is fascinating is that it does run on a LLM. Unlike the traditional AI tools, Khanmigo uses SLMs to offer a more efficient and accessible learning experience. While LLMs have already found a place as our second brain or gone under our skins, SLMs are still an unknown commodity. What are SLMs? SLMs are small language models. Small and big is the first simple difference for a layman. It is a lightweight generative AI model. What does small refer to in technical terms? It refers to the model’s neural network, the number of parameters used for decision making and volume of data required for training.
Without getting into competition and unnecessary comparisons with GPTs of the day, the SLMs are quietly proving their immense value. What can the SLMs do? They can handle specific tasks with far less computational power compared to LLMs, suits industries and businesses with limited resources. SLMs are making waves in the area of providing efficient and effective AI solutions which have been tailor made for niche applications. It has made cutting edge AI affordable leading to the democratization of AI in the true sense. The smaller size allows for deployment on edge devices.
While there are different models of SLMs like the Distilled, Pruned and the Quantized ones, the models which are trained from scratch may turn out to be most effective. This allows them to be especially built for a particular task or domain from the ground up. It can thus be optimized for specific tasks or industries, such as legal, finance and healthcare. Given their requirements, SLMs can do wonders there. Another element that goes in their favour in these sectors, is that they are more controllable and have very few dependencies. With regards to privacy and security, they are the clear cut winners. Deployed on-premise, they are ideally suited for operations where data privacy is critical.
SLMs ARE DEMOCRATISED VARIANT OF LLMs IN EVERY WAY.
Sanjay Sahay