Dear readers, have you ever dreamed of having a little iPhone in your pocket, being able to have Siri smarter than it is now, even being able to translate alien languages in real time, or giving professional photographer-level advice when you take photos?Well, it may not be so exaggerated, but Apple's latest research results have indeed brought these "sci-fi" scenarios closer and closer to us!
In this era of AI, where data and memory are eating more than the big stomach, Apple's researchers have come up with a new trick. Instead of buying more memory modules, they're looking at the flash memory that stores us and our cats. Through an innovative technology, they were able to deploy large language models (LLMs) on memory-constrained devices like iPhones, making Siri smarter and making real-time language translation and augmented reality more powerful.
Windowing: It's like when we eat a buffet, instead of going to get a new dish every time, we finish it on the plate first. In this way, the AI model can process data and reuse some of the data that has already been loaded, reducing frequent access to memory.
Row-Column**: Imagine that you are reading a wonderful **, not reading word by word, but sweeping a whole paragraph at a glance. This approach makes data reads more efficient, speeding up AI's understanding and responsiveness.
Not only does this technology make Siri more flexible, but it also brings real-time language translation and more sophisticated AI capabilities to iPhone. Apple's goal is for Siri to be more than just a little assistant asking for directions, but an intelligent companion that can understand and complete complex tasks.
The rumored "Apple GPT", codenamed "Ajax", is Apple's secret** with 200 billion parameters and is expected to give us an unprecedented AI experience when iOS 18 is released. This means that the iPhone and iPad of the future will not only be our communication tools, but also our smart companions.
Large language models (LLMs).: An AI model trained on large amounts of data that understands and generates natural language.
Flash: An electronic storage medium used to store data, such as apps and apps in a mobile phone.
Windowing: A data processing technique that reduces memory access by reusing parts of data.
Row-Column**: An optimization method for reading data, which improves the reading efficiency by grouping data.
Generative AI: An artificial intelligence system capable of creating new content such as text, images, etc.