The Rise of GenAI and LLMs

CHATGPT-3.0 AND BEYOND

Then, near the end of 2022, OpenAI released ChatGPT, an LLM-based chatbot capable of generating near-human conversation on various topics. ChatGPT became the most quickly adopted technology in history.

Microsoft has invested approximately $13 billion in Open- AI. It incorporated ChatGPT and OpenAI’s Dalle-2 image generation tool in the Windows operating system and on its Edge browser. Users may use these tools, called Copilots, for free. Microsoft also sells Copilots for use within its application suite.

Platforms like Hugging Face and Google’s Bard/Gemini further advanced the field by providing user-friendly frameworks and tools, empowering researchers and developers to create their own LLMs.

Since the release of ChatGPT, several organizations, including Google, Meta, and Anthropic, have released consumer and commercial models that continue to evolve. Tools such as LLM Studio allow end users with gaming or AI-capable computers to download models and run them locally, removing barriers that rely on cloud or other server-based resources.

In June 2024, Apple announced Apple Intelligence, which offers several variations on local LLM experiences while still offering access to larger models through secure interfaces.

Apple device owners will need to approve any query that leaves the confines of Apple’s walled AI garden. The company also announced servers that deliver to Apple ecosystem participants new guarantees about data privacy. Perhaps more importantly, Intelligence incorporates a knowledge index that ties content, activities, people, and other objects and events together, allowing the next version of the Apple ecosystem to act based on LLM inferences and recommendations.

LLMs evolved to handle text, images, and sounds, leading to multimodal LLMs. These models integrate text, images, audio, and video, enabling comprehensive data understanding and analysis. Applications include extracting text from digital images, deciphering ancient handwriting, and analyzing speech files for summarization and transcription. Multimodal LLMs represent a significant leap in AI’s ability to process and generate diverse data types.

WHAT’S NEXT?

My charge was to write about the rise of GenAI and LLMs. Condensing the history of AI into a single article requires a brevity that necessarily truncates the story and ignores side journeys. I endeavored to convey the essential developments built upon the work of Turing, Rosenblatt, and many others who innovated to deliver the GenAI systems that are beginning to permeate our lives and work. Some fear these tools. Some have overly exuberant expectations for them.

This brief history should assuage the fear that the AIs themselves have any conscious volition to harm. They are only complex mathematical algorithms. They do not encode intent or personality; they not-so-simply construct information based on their prediction of a good answer to a user query.

For those who only see AI’s unlimited potential, this history should illustrate the rich collaborative problem-solving reflected in LLM and chatbot algorithms. Not all the problems have been solved, and the higher ambitions for more sentient computing remain far from being achieved.

As a conclusion of the chapter on its history, the only closing remark that would be honest is this: The history of LLMs and GenAI continues to be written. Some of it will likely be written by the AIs themselves.

Download the Enterprise AI Sourcebook here