Large Language Models are on every blue-chip corporation’s roadmap. Meta, Microsoft, Amazon, X, you name them. And it doesn’t seem like the hype is stopping anytime soon. For Tim Cook, the train has only started and if his company’s newly released paper is anything to go by, they are not here for child’s play.
AI is what most investors are really excited about. Almost all momentum in the market in general is being fueled by AI.
Brian Mulberry, a portfolio manager at the Apple shareholder Zacks Investment Management. Source: The Wall Street Journal
Related: OpenAI’s Diverse Projects: Advancing AI Across Domains
Apple has tossed its hat into the artificial intelligence arena after releasing a paper that argues ReALM, its new language model outperforms OpenAI’s GPT-4. One of the benchmarks that the language model shines at is the ability to understand different contexts through Reference resolution. For example, a user could query ReALM, which stands for “Reference Resolution As Language Modeling, about a podcast playing in the background and it should be able to understand.
ReALM’s Performance Evaluation
Reference resolution is a term in linguistic processing where a word in a conversation ends up referring to something else. For example, it might be easy for you to understand references like ‘that’, and ‘them’ but wouldn’t be obvious for a language model. As per the paper, Apple has designed its AI model’s reference resolution to perfection so that chatbot users can have a “truly hands-free experience”
For it to achieve this seamlessly, the language model will harness its understanding of three major types of entities, background, onscreen, and conversational entities. Onscreen entities as the name suggests, are anything that appears on a user’s screen. For example, a user could point at something displayed on the screen and ReALM will be able to identify and provide a plausible response about it.
Conversational entities are anything of significance and relevance to the conversation between the chat and users. For instance, the language model will derive context from previous conversations when asked things like “What should I post on my Linkedin today?” or “What Should I add to my diet this week”, so that if you had previously asked about a 6-month diet, the model can work out the previous conversation and provide a relevant response.
Like we mentioned earlier, ReALM has the capability to make sense of let’s say a podcast or audio file playing in the background. If the user makes a reference to it, the language model can put the background entity into context and respond accordingly to a specific query about that entity.
According to the paper, the Artificial Intelligence model demonstrated absolute gains of more than 5% for on-screen entities reference resolution when they compared its smallest version with GPT-3.5 and GPT-4. When they made comparison between OpenAI’s GPTs against their larger version, ReALM significantly outperformed them, the researchers wrote.
Implications of Apple’s AI Model
The paper demonstrated how the company’s advancements in AI will disrupt large language models and natural language processing when performing reference resolution. Most of the existing models are good at processing text inputs but with ReALM, Apple aims to pass both text and non-text inputs into an LLM using its specialized features.
More Info:
While Apple has not made it clear whether it intends to incorporate the language model on its line of products, there are speculations that ReALM could be integrated in the company’s software offerings on macOS and iOS. Or perhaps even integrate it with Siri to bolster the virtual assistant. Meanwhile, this is the company’s third research paper in recent months to report on Apple’s quantum leap into artificial intelligence.
Disclaimer: All materials on this site are for informational purposes only. None of the material should be interpreted as investment advice. Please note that despite the nature of much of the material created and hosted on this website, HODL.FM is not a financial reference resource and the opinions of authors and other contributors are their own and should not be taken as financial advice. If you require advice of this sort, HODL.FM strongly recommends contacting a qualified industry professional.
