There are similarities between the cold war era and current times. In the former, there was a belief that alliances having stronger nuclear arms will wield larger global influence. Similarly, organizations (and nations) in the existing era believe that those controlling the AI narrative, will control the global narrative. Moreover, scale was, and is, correlated with superiority; there is a belief that “bigger is better”.
Global superpowers competed in the cold war on whose nuclear systems are largest (highest megaton weapons), while in the current era, large technology incumbents and countries are competing on who can build the largest model, with highest number of parameters. Open AI’s GPT 4 took global pole position last year, brandishing a model that is rumored to have over 1.5 trillion parameters. The race is not just about prestige; it is rooted in the assumption that larger models understand and generate human language with significant accuracy and nuance.
As the world is witnessing this massive paradigm shift, there is, however, a counter argument developing. The new narrative suggests that the future of AI may not lie in the gigantic frameworks of model, data and computation but rather in the nuanced and agile world of small language models.
Democratization of AI
One of the most compelling arguments for smaller language models lies in their efficiency. Unlike their “larger” counterparts, these models require significantly less computational power, making them accessible to a broader range of users. This democratization of AI technology could lead to a surge in innovation, as small businesses and individual developers gain the tools to implement sophisticated AI solutions without the prohibitive costs associated with large models. Furthermore, the operational speed and lower energy consumption of small models offer a solution to the growing concerns over the environmental impact of computing at scale.
Jack of all trades vs Master of One
Large language models’ popularity can be attributed to their ability to handle a vast array of tasks. Yet, this Jack-of-all-trades approach is not always necessary or optimal. Small language models can be fine-tuned for specific applications, providing targeted solutions that can outperform the generalist capabilities of larger models. This specialization can lead to more effective and efficient AI applications, from customer service bots tailored to a company’s product line to legal assistance tools tailored on a country’s legal system.
On-device Deployment
A significant advantage of small language models is their ability to run locally on devices, from smartphones to personal computers. This on-device deployment enhances privacy and security, as data does not need to be sent to remote servers for processing. It also reduces latency, making AI applications more responsive and practical for everyday use. The movement towards integrating small language models into consumer technology is already underway, with companies incorporating them into their latest devices.
The Environmental Imperative
The environmental impact of AI development is an issue that cannot be ignored. The massive energy requirements of training and running large language models pose a significant challenge in the search for sustainable technology development. Small language models offer a path forward that marries the incredible potential of AI with the urgent need to reduce our carbon footprint. By focusing on models that require less power and fewer resources, the AI community can contribute to a more sustainable future.
What’s next ?
As we stand on the cusp of technological breakthroughs, it’s important to question the assumption that bigger is always better. The future of AI may very well lie in the nuanced, efficient, and environmentally conscious realm of small language models. These models promise to make AI more accessible, specialized, and integrated into our daily lives, all while aligning with the ethical and environmental standards that our global community increasingly seeks to uphold.
Their partnerships with leading mobile OEMs globally which cover 63 per cent of the global Android market helps Fintech brands to feature their apps on alternative app platforms. They also offer guidance throughout the campaign lifecycle for expanded reach and new revenue opportunities. Furthermore, some new age app growth companies have also launched their proprietary tools which fine-tune campaigns in real-time across mobile OEM inventory, aligning them with performance goals for enhanced Return On Ad Spend (ROAS).
