Apple is preparing Apple GPT for the next iPhones, but with a feature that will make it unique

It seems that in 2024 too Apple will finally enter the artificial intelligence sector with its own chatbot called Apple GPT, but apparently the house of Cupertino doesn’t want to follow the path of other companiesbut it’s making sure the processing happens directly on the phone.

The news comes directly from a scientific articlewhich illustrates an innovative technology of use of flash memory (the one you use to store photos and apps) to allow you to implement a large language model (LLM) on devices with limited memory.

Due to the incredible data intensity used, LLM-based chatbots such as ChatGPT And Claude in fact, they require large amounts of memory to function, which means having to base all calculations on them server of companies.

Apple, however, wants to make sure that everything happens on the device and for this reason he tried to exploit a feature of telephones: the fact that the flash memory is more abundant than the RAM traditionally used in servers for running LLMs.

For this reason, in the article entitled “LLM in a flash: Efficient Large Language Model Inference with Limited Memory“, two techniques have been introduced which on the one hand reduce the data transfer and from above they maximize the surrender of flash memory.

The first is called Windowing, and allows you to “recycle” data already processed instead of reloading it every time: this reduces the need for constant memory retrieval, making the process faster and smoother.

The second was instead nominated Row-Column Bundlingand allows you to group data more efficiently so you can read it faster from flash memory. The basic concept is like reading a book in chunks rather than word by word, and this system accelerates AI’s ability to understand and generate language.

The result, at least on the article, is astonishing. These two methods combined allow AI models to work up to double of the iPhone’s available memory, which results in an increase of 4-5 times of speed on standard processors (CPU) and up to 20-25 times faster on graphics processors (GPU).

According to the authors of the article, which you can find at this address, this discovery is a real breakthrough capable of allowing the implementation of advanced LLMs in environments with limited resources.

And apparently that’s exactly what Apple is looking for, according to the most recent rumor he plans to launch his own Apple GPT on iPhones and iPads at the end of 2024, with features related to both cloud which – above all – elaborate directly on device.

Still little is known about this chatbot”made in Cupertino“, but apparently the apple house is aiming for Siri functionality more advanced, especially regarding integration with Messages and other proprietary apps, but also linguistic translation in real time and applications in photography and in the augmented reality (after all, we know how much Apple is aiming for the new visionOS).

And this seems to be the key point of the project. As with Apple’s operating systems, so is the new one Apple GPTwhose code name would be Ajax, would be developed to integrate AI more deeply into Apple’s entire ecosystem.