The Evolution of Generative AI
Generative AIs are artificial intelligences (AI) that are trained on language, images, video and other content. When prompted to do so, they use the data they’ve been trained on to generate new content, in the form of text, images, sounds, and even code. Code is, after all, a form of language.
The language-based generative AIs are based on large language models (LLMs). They do exactly what it sounds like they do – they accept queries or instructions posed in formal languages, be it a human language or code, and generate a response in the appropriate format.
Avnet Silica has a wealth of experience using machine learning (ML) tools to create LLMs for use in generative AIs.
The evolution of generative AI
Generative AI is not new. It has, however, recently taken an enormous leap in what it can do, inspiring no little amazement, as well as a bit of alarm.
A popular candidate as the first generative AI is the first chatbot, ELIZA, introduced in the 1960s, which was trained to respond like a psychologist. The subsequent history of AI research was marked by AI “winters” alternating with periods of resurgence typically coincident with some applicable technological advance, such as the introduction of microprocessors or the formulation of AI techniques such as back-propagation.
From the public’s standpoint, AI’s history is one of intermittent highlights: Zojirushi introducing a rice cooker based on fuzzy logic in 1983; IBM’s Deep Blue defeating chess master Garry Kasparov in 1997; Apple’s introducing its Siri voice assistant in 2011. But research was always bubbling along.
Generative AI in the last few years
The most recent chapter in the history of generative AIs is predicated on the formulation in 2014 of a particular AI model called a generative adversarial network (GAN) that has proven to be particularly useful for the purpose.
Starting in 2016, OpenAI began publishing a series of research papers that claimed significant progress with generative models trained on text and images.
Right about that time, developers started using rudimentary generative AIs to create a series of apps, most of them designed to be fun and enticing, some of which became social media phenomena. One popular example was the Faceapp ageing app, which made use of GANs. People were asked to upload photos of themselves. The AI, trained on photos of other individuals growing old, would modify the submitted snapshots, generating new images of what each person might look like after decades of ageing.
Galvanised by the introduction of GANs and by the popularity (and results) of such apps, and especially by the progress that OpenAI was reporting, other companies jumped in. But even competitors who thought they were pacing OpenAI on LLM development – including enormous multinationals such as Amazon, Baidu, and Google – were taken by surprise at the end of 2022 when OpenAI released the first commercial generative AI tool, ChatGPT (the G stands for generative), an instance of an LLM.
General Generative AI
ChapGPT and other chat tools that followed are adept at responding to requests for data about almost any subject (sumo, semiotics, ska, etc.) with a report that summarises the available knowledge on that subject. Generative AI tools and products are being offered by a growing number of companies. Examples include Microsoft making chat capabilities available through its Bing portal to Rabbit offering a personal assistant based on generative AI in its standalone R1 device.
If trained on fiction or on music, generative AIs can output poems, stories, songs, or symphonies that are new, albeit based on the training examples. Trained on visual input, a generative AI can take elements of the imagery it has learned and combine them to generate new photographs, graphics, or movies.
![Two people looking at screen as part of Generative AI for business](/wcm/connect/dae730b1-950c-4ae9-ad44-2fd184823970/Generative_AI_for_Business_16x9-min.png?MOD=AJPERES&CACHEID=ROOTWORKSPACE-dae730b1-950c-4ae9-ad44-2fd184823970-p45yQII)
Figure 1: Generative AI can be the basis for online tools that companies can build for their customers to query to identify and select the most appropriate products & services for their needs. Companies can also build tools based on generative AIs, trained entirely on internal data, to do a wide variety of tasks, including writing code or generating test routines.
silica content library/home/solutions/technologies/artificial-intelligence-machine-learning/generative-ai/evolution-of-generative-ai/evolution of gai main body (lc)
Evolution of GAI Main Body (LC)