Much AppleGPT has been rumored or the possibility that the Cupertino company is developing its own Artificial Intelligence system. This fact seems increasingly closer after the publication of a study by the company where it mentions a new way of generating data using mobile hardware.
So if you want to know how AppleGPT could work and why we believe it is an innovation in the market, we advise you to continue reading this article.
How do GPT models work?
Traditionally, GPT models use the cloud using an architecture called Transformer, developed by OpenIA, which is a type of neural network designed to process sequences of data, such as text, that uses attention mechanisms to process information in parallel and capture relationships to throughout the sequence.
These models are pre-trained on large amounts of unlabeled text data. During this pre-training phase, the model learns to predict the next word in a sequence given previous information, which helps the model develop a deep understanding of language and grammatical structures.
And through this analysis, artificial intelligence develops its ability to generate text in a generative way, trying to imitate the way that human beings express themselves.
Everything seemed to be going in these parts, until Apple came with its study āLLM in a flashā and send to all this for a walk with a new way of treating and understanding AI.
How would AppleGPT work? (in theory)
Mobile phones will be the hardware that AppleGPT will use, according to the report
In the study, Apple researchers open the door to two different processes being used to generate AI:
- Widowing: which allows the reuse of data already investigated by the AI āāitself, saving the cost of having to make new calculations again.
- Row-Column Bundling: a way to group data more efficiently, which would help accelerate the ability of AI to understand and generate new content.
With these two methods, Apple could afford to use AI without the need for the cloud and even without the use of the Internet in certain situations (but not in all of them, since the information we are looking for in GPT has to come from somewhere).
Furthermore, the use of mobile components would allow the AI āāprocess on the mobile CPU to be multiplied by about 4-5 times and the use of the mobile GPU by up to 25 times, which is not bad at all.
The probable evolution of AppleGPT: improving the use of Siri
But everything seems that Apple will not release a ChatGPT roll portal or similar, but rather will work on improving the home voice assistant, Siri, so that it is compatible with these Artificial Intelligence functionalities.
In this way, the assistant could answer more complex questions in a generative way, in a similar way to what Microsoft's Bing search engine would be doing, which uses ChatGPT among other engines.
There is also speculation that Siri can do simultaneous language translation, or that generative AI tools can reach fields such as photography or video, which will allow us to delve into the use of augmented reality from our iPhone.
Why is this way of understanding AI relevant?
This new way of managing AI that Apple proposes could be relevant for different reasons, which we will tell you about below.
Greater data protection
By running AI models locally, you can address privacy and security concerns with a novel approach, especially knowing that governments are now interested in putting some control over the use given to the data collected.
Apple GPT would not require data to be transferred to the cloud, being crucial for applications that handle sensitive or confidential information. This would meet the demands of governments such as that of the European Union, which are more protectionist in terms of preventing the use of personal data by companies.
Possibility of using AI in situations without mobile coverage
Furthermore, the offline processing of information will bring the advantage of use of AI technologies in environments where connectivity is low or absent, and in turn it would do so more quickly since by eliminating processing on servers, the latency associated with data transfer would be significantly reduced.
Efficiency of resources, which could be used for another purpose
By avoiding the need to send data over the network, resources and bandwidth are saved. This can be essential on devices with limited resources, such as IoT (Internet of Things) devices or embedded systems (which usually have low specifications) or it could leave part of those resources that are currently being allocated for AI to be used for other purposes.
It would reduce the environmental cost of AI
Also, we have the latent environmental problem which is entailing the development of artificial intelligence. Statements like those of Sergey Edunov, director of AI engineering at Meta, in which he indicates that to maintain all the AI āāthat is being developed, at least the power of two nuclear power plants would be needed, making clear to us the waste of resources that this represents for the world.
The useful life and recycling of the hardware used in the implementation of AI is an important factor to take into account, since current models the hardware "retires" too early, which is replaced by more efficient models. This hardware is discarded as technological waste, which ends up being a big environmental problem.
By using mostly native hardware, all of this would be solved since energy consumption and the associated hardware to develop AI would be reduced.
And this is where Apple has the power to hit the table and lay the foundations for more sustainable artificial intelligence.
Do you think AppleGPT will be a decent competitor to ChatGPT and other models? We want to know your opinion, don't forget to leave it in the comments!