In recent years, one of the most spoken and read words in the world of technology is artificial intelligence (AI). Almost every day we come across news that speak of this announced revolution, however, very rarely is it specified what type of artificial intelligence we are referring to exactly, since there are different types. What is causing such a sensation is generative AI . So let's try to better understand what generative AI is.
What is generative AI?
Generative Artificial Intelligence, or Generative AI, is a technology that rose to prominence with the Chat GPT software from OpenAI, a startup owned by Microsoft. Today, there are many other such software. Generative AI can improve the performance of various activities for both individuals and businesses, such as the production of texts, images, and standard software codes . Its use speeds up work and, to a certain extent, can be defined as creative, thanks to the combination of large amounts of sources and data used.
As defined by McKinsey, generative AI describes algorithms that can be used to create new content , including audio, code, images, text, simulations, and video.
Generative AI systems fall into the broad category of General armenia whatsapp data Artificial Intelligence (AGI) and Machine Learning ( ML). They have the potential to change the way we approach content creation for applications such as design, entertainment, e-commerce, marketing, scientific research, and human resources.
It is clear that we are in a time of great evolution, but we are still at the beginning and we will have to wait a little to better understand what the real developments of generative AI will be and what its consequences, opportunities and risks will be for our lives and jobs. What is clear is that training in this field, for example with a Master in Artificial Intelligence & Machine Learning for Business , is a choice that will open up interesting job prospects.
How does generative AI work?
Generative AI software starts from requests or descriptions ( prompts ) formulated in natural language by the user (human or software) and consequently generates texts from texts ( Text-to-Text ), images from texts ( Text-to-Image ) or even images of images ( Image-to-Image ). The results of these systems are combinations of the data used to train the algorithms .
Because of the huge amount of data used to 'feed' the software (the GPT-3 system on which Chat GPT was based was trained on 45 terabytes of text data), the results can appear 'creative'. In reality, what they generate is a compilation of a combination of sources , but given the huge amount of data processed, the result can be genuinely novel. After all, reworking can also be considered a form of creativity.
Understanding how generative AI works is not an easy task for non-experts: the idea is that, based on feedback and training, intelligence constantly improves. For example, Chat GPT technology could be defined as an example of a Generative Adversarial Network or GAN. However, the issue is debated, because according to some experts Chat GPT would be a Transformer (GPT is the acronym for Generative Pretrained Transformer ) and not a GAN. What does it mean?
Transformer is a deep learning model used in the field of NLP (natural language processing), where results are generated from a reworking of previously stored information. GANs, on the other hand, are a type of artificial intelligence algorithm that uses two competing neural networks to generate images, sounds, text, and other types of data. The first network, called the 'generator', tries to create fake images or data that look real; the second, called the 'discriminator', tries to identify whether the images or data are real or fake.
The two networks compete with each other: the generator tries to generate increasingly realistic data while the discriminator tries to better identify whether the data is real or fake. Over time, the generator gets better and better at generating realistic data that fools the discriminator, while the discriminator gets better and better at identifying fake data. The goal of a GAN model is to optimize deep learning and avoid shallow generalization errors due to data sparseness.