Apple Intelligence - The Importance Of Small Language Models
Earlier this week, Apple unveiled its AI strategy at its 2024 World Wide Developer Conference in California.
The company has been facing mounting pressure since November 2022, when OpenAI made a significant breakthrough with the release of ChatGPT-3. Tech giants such as Google, Meta, and more, followed suit with their own propositions, intensifying the competitive landscape.
Since then, terms like artificial intelligence (AI), generative AI (GenAI) and large language models (LLMs) have become buzzwords in many organisations. Yet, as McKinsey reports, public and private sector entities are moving at pace to roll out their respective AI strategies in the hope that they can find value, growth and improved service or product adoption.
Only a few weeks ago I met up with IBM’s EMEA Business Leader for WatsonX and EmbeddableAI Hans-Peter Dalen. I raised the issues about reputation and cultural nuances and how GenAI tools can factor these in to give a more personal experience to users. His answer? Small Language Models, or SLMs.
While much attention has been placed on the value and importance of data and large language models, SLMs and smaller data sets are critical in delivering a unique, bespoke and personalised experience.
Tech companies that have released generative AI services have primarily focused on a cloud-based strategy and solution. Without a doubt, this has come with many challenges and issues, such as:
The level of processing power and shortage of chips,
Environmental issues based on high energy consumption needed
Privacy and datasets
Lack of cultural nuances by GenAI services, which can lead to reputational issues for companies delivering and organisations relying on AI
Small language models resolve many of these issues, and it looks like many companies, like Apple and Arm, are pivoting to offer a service that takes into account the individual user and this technology's energy and processing footprint.
Here is a great post and description from IBM’s Armand Ruiz on Small Specialised models:
In a recent blog, Arm announced that a new and ‘efficient arm computing enables a custom AI future’. In essence, generative AI is not just a cloud-based method, but, because ‘software advancements – coupled with more efficient and powerful Arm CPU technology – enable these smaller, more efficient language models to run directly on mobile devices, enhancing performance, privacy and the user experience.’
This strategy places the smartphone and the data within it, which is generally unique to each owner, at the centre of what information and insight it can deliver. The fact is that, as CNBC reports, over the last three decades, Arm has become the dominant company making this chip architecture, and it powers nearly every smartphone today. Apple bases its custom silicon for iPhones and MacBooks on Arm, and now Nvidia and AMD are reportedly making Arm-based PC chips, too.
Enter Apple, which this week revealed that its Apple Intelligence gen AI offer will be built into the core of its devices. It sees an environment where the datasets we have in our devices enable its offer to work on the device, yet for some complex queries, it can push them to either it's own private compute cloud or to OpenAI and its ChatGPT solution.
In securing scale and growth, organisations need trust and reputation, and this is something that Apple is investing in: the creation of an environment where privacy and security are at the centre of its offer, and personalisation makes the experience unique for users.
For a company that protects its reputation so much, Apple's move feels like the launch of the App Store. They have invested in delivering a personalised experience based on helping to deliver insight and solutions from the data that each device, through acknowledging SLMs, while connecting users with LLMs for more complex queries and prompts.
Wherever you are, whatever language you speak, for Apple, AI is about experience and privacy.
