Apple remained silent during the rapid rise of the ChatGPT robot, which significantly changed the tech sector shortly after its launch, as well as during the widespread adoption of generative artificial intelligence tools and features offered by major competing companies such as Google, Microsoft, and Samsung. However, it is expected that this will change soon, as Apple researchers have developed a new linguistic model called ReALM, which is intended to provide the voice assistant Siri with the upgrade that Apple device users are hoping for.
In response to this, the researchers said, “The new ReALM model addresses the issue of large language models (LLMs) that lie in the inability to understand context when it comes to contextual and ambiguous signals that appear during conversations and in the background. The model has shown a significant improvement compared to models with similar functions, with its smaller versions outperforming by up to 5% in processing screen content.” They also noted that the larger version of the ReALM model offers much better performance than GPT-4, the most advanced model offered by OpenAI.
Therefore, in this article, we will shed light on the new ReALM model Apple developed and explain how the assistant Siri can help outperform ChatGPT.
First of all, what is the ReALM model?
The Artificial Intelligence model (ReALM) developed by Apple aims to enhance the capabilities of its voice assistant (Siri) in understanding context and improving performance in processing visual content displayed on the screen.
This new model comes in four versions of different sizes: ReALM-80M, ReALM-250M, ReALM-1B, and ReALM-3B, all developed to be computationally efficient for integration into mobile devices. This efficiency enables consistent performance while reducing energy consumption and relieving processing resources, which is important for extending battery life and providing quick response times across a variety of devices.
Performance tests conducted by researchers have shown that the smallest versions of the ReALM model offer similar performance to the GPT-4 model, and even outperform it in processing screen content. The larger version of the model delivers significantly better performance than the GPT-4 model.
Secondly, how will the ReALM model help Siri outperform ChatGPT?
1- Screen Context Understanding:
Apple researchers developed the ReALM model using data that appears on the screen from web pages, including contact information, enabling the model to understand the text inside screen captures – for example, titles and bank account details – while the GPT-4 model is also capable of understanding images, but it has not been trained on screen capture images.
The researchers explained in the published research paper that this data makes the ReALM model better at understanding the information that appears on the screen, which Apple users ask Siri for help with.
2- Conversation Context and Background:
Dialogic inference refers to something relevant to the conversation, but it may not have been explicitly mentioned in the question, and by training the ReALM model on data such as business listings, it can understand requests like “call the nearby pharmacies” about the displayed list of nearby pharmacies on the screen, without the need for more specific requests.
Additionally, the ReALM model can understand background data entities, meaning something that works in the background of the device and may not necessarily be a direct part of what the user sees on their screen or during their interaction with the voice assistant, such as playing music, setting an alarm, or downloading files.
3- The advantages of having generative artificial intelligence directly on the device:
Apple is all about putting generative artificial intelligence features into the device to ensure top-notch user privacy and security. The goal is to give you some serious AI smarts without risking your data.
Therefore, Apple designed the ReALM model to operate entirely on the device, which is crucial as large language models require immense computational power and often rely on cloud computing. In contrast, ReALM is a small language model specifically tuned for the benchmarking task.
Apple values privacy, and they think it’s a major selling point for their gadgets. With this new Siri update using fancy AI tech right on the device, Apple’s sticking to its privacy-first image and taking a big step forward in AI-powered devices.
However, Apple has maintained the expected ambiguity regarding its plans related to generative artificial intelligence so far. Nonetheless, all reports and leaks indicate that the upcoming version of the iOS operating system will be the biggest update in Apple’s history thanks to artificial intelligence. All eyes are now on Apple’s annual Worldwide Developers Conference (WWDC 2024), scheduled for June 10th.