Apple is Investing in multimodal AI

There is growing evidence that Apple may be in the process of constructing a more powerful AI system by developing a new large language model (LLM). If this is the case, then it is likely that they are carrying out the groundwork for future AI applications to be run in Apple systems.

A new research paper entitled “MM1: Methods, Analysis & Insights from Multimodal LLM Pre-training” has turned up on the Cornell University arxiv.org site, and it details different types of data sets and model architectures for training AI.

What Are Large Language Models?

Humans have used language to communicate ideas for thousands of years, and it has become our main means of transmitting information, data and ideas to others. Language is a hugely flexible and convenient way of transmitting information. The same model that we use every day has been adapted as a basis for AI learning with the Eliza language model which debuted in 1966 and is one of the earliest examples of AI language. 

Based on the same constructional systems, a large language model is a class of artificial intelligence system that understands, summarises, generates, and forecasts new material using extremely huge data sets and deep learning methods. In truth, LLMs are a subset of generative AI that has been designed especially to assist in producing text-based content. The terms are very closely related.

Typically, LLM’s use a large number of parameters – usually many millions – to define the conditions in the language; the greater the number of parameters, the more intuitive the AI becomes. MM1, the system being developed by Apple, employs over 30 billion parameters, and includes specific features like dense models and mixture-of-experts variants. Mixture of experts (MoE) is a machine learning method that uses many expert networks (known as learners) to split a development problem space into areas that are all the same. MoE is different from ensemble techniques because their models are run on every input, but in MoE, only one or a few expert models are run for each input. All of these features make the MM1 a powerful tool.

Apple Gets Serious

Strong in-context learning capabilities enabled the MM1 model to use so called “chain-of-thought” prompting it to do multi-step reasoning over several input pictures in testing. Essentially, the AI element will examine an input a number of times to ensure that it fully understands the content and the context before moving on to the next string. This suggests that big multimodal models may take on difficult, open-ended challenges that call for the production and comprehension of grounded language. If that is the case, then Apple could be looking at an AI system that has huge capabilities in both learning and reasoning. In other words, another huge leap towards true AI.

Up until now, Apple has been lagging behind the likes of Google, Amazon, and Microsoft when it comes to AI, and are keen to catch up quickly. Apple is reportedly working on a huge language model framework that is being referred to as “Ajax,” as well as a chatbot that is being referred to internally as “Apple GPT”. These reports are becoming more frequent. 

The intention is to incorporate these technologies into a variety of applications and services, including Siri, Messages, Apple Music, and others. Artificial intelligence might be utilised to perform tasks such as automatically generating personalised playlists, assisting developers in the process of developing code, engaging in open-ended discourse, and completing tasks.

As AI becomes increasingly important to all areas of technology, to not be investing in such systems would place a major company at a major disadvantage, hence Apple’s huge thrust to get into the game. When it comes to big technological advances, Apple has a reputation for being a ‘Me-Too’ company rather than a pioneer. 

The Cupertino company is facing high-stakes competition as AI continues to revolutionise the digital world. According to MM1’s findings, Apple possesses the know-how and means to create innovative breakthroughs. How fast the infamously secretive corporation can respond to the intensifying AI weapons race is something that we are waiting to find out.

Apple is a very secretive company when it comes to its developments and tends to only drop the minimum amount of information. With the future of technology becoming increasingly entwined with AI, it is essential for major players in the field to carve out their niche in the field. We at Unity Developers will bring you all of the news as we get it.