2023-05-20 01:18:31
The artificial intelligence Generative AI (AI) capable of answering any question or creating images seemingly out of thin air have impacted society like no other technology before. The speed at which these systems are advancing has caused some countries to study setting limits, that personalities of the level of Elon Musk consider stopping their development, and that their own creators, such as Sam Altman, CEO of OpenAI, advise regulating it.
It all started with OpenAI and ChatGPT, but has been quickly followed by Microsoft with Bing and copilot, Google with Bard, and tools capable of generating images from text like Midjourney or Stable Diffusion. These tools can write creatively, render images like an illustrator would, create original music, and are going to take it to the next level at any moment, video.
There is already talk of applications that, from a handful of words provided by the user, will be able to create a complete movie for us. For example, if we want to see a movie about vampires at the time of the construction of the pyramids with Tom Hanks as the protagonist, we would only have to ask. If something like this happened, the entire film industry could suffer the effect. Last week, Google showed its search engine during its I/O event, giving results through this new artificial intelligence model, radically changing what an internet search has been in the last 20 years.
However, what the company has not clarified so far is that this new type of search will be up to ten times more expensive than the traditional, according to Cristiano Amon CEO of Qualcomm, one of the main global companies dedicated to the manufacture of chips. Despite this, the executive is convinced that artificial intelligence will invade every corner of our lives in the coming years.
The poor business of AI
Although the technology industry has fully immersed itself in the artificial intelligence business, and announcements and launches occur one after the other every day, the reality is different. Right now, not even with all the processing capacity installed in all the data centers in the world could we face a new era dominated by tools like ChatGPT. Although it seems that ChatGPT or Bing are cheap technologies because we have free access, they are not. It is estimated that OpenAI spends around 700.00 dollars a day in computing capacity, or what is the same, it uses 36 cents per day. every question we ask.
The strategy, taking this data into account, must be clear: give cheap access to collect more data with which to feed its artificial intelligence, and at the same time gain a greater market share. And that is not easy.
As Ignacio Contreras, senior director of Qualcomm, points out in conversation with ABC, generative AI not only requires greater use of computing, but more need for electricity and in turn requires more cooling, which also means more energy use. This means that, as this technology is conceived today, the business is not particularly profitable.
A possible solution
Qualcomm’s proposal is to create an architecture or a hybrid artificial intelligence. It consists of using the capacity of the processors, not only of the data centers, but also those that are in our phones, computers or cars, that is, using their computing capacity, to, in combination with the cloud, make the large language models are scalable and accessible.
The capacity of this new category of artificial intelligence is measured by the number of parameters it uses to give answers. ChatGPT uses a whopping zillion parameters every time we ask it a question. Right now a Qualcomm Snapdragon Gen 2 processor is capable of processing 1 billion parameters locally, inside the phone, and although it is far from the target, it is capable of generating an image in 15 seconds as Midjourney would. The next generation of Qualcomm processors, will reach 10,000 million parametersas Cristiano Amon has assured.
«ChatGPT is a foundational, generalist model, designed to solve practically any problem, but if we can adjust the model to your needs, the number of necessary parameters drops drastically, something that can be done when processing locally, on a mobile phone. Some of the more specific models that are coming out now use only 7 billion parameters,” says Zaid Asghar, Qualcomm’s vice president of product.
Another advantage of hybrid architecture is the private AI, that is, the one that does not leave our device and is trained with our personal data, but without reaching the cloud and, therefore, out of the reach of third parties. It also ensures the accessibility and democratization of AI, where anyone with a mobile phone should be able to use it without paying for third-party tools.
Those of us who have been in technology for a while know that most scientific advances do not end up permeating despite the novelty. But the big difference compared to other technologies that have generated great expectations but have since declined, as Ignacio Contreras tells us, is that the benefits of generative AI are going to be seen now.
#artificial #intelligence #change #life #solve #big #problem
