GenLens: Introduction to Large Language Models

Welcome to Language Lens, where Edditter.io will try to demystify and break down famous significant language model concepts, history, theory, and practical implementations. The series is intended to introduce the vast field of Generative AI through niche topics and concepts you may not see well explained online.

As part of this segment, let us start with “Introduction to Large Language Models.” Before diving in, though, let us first take a look at what Generative AI is and what is the weight 

Generative AI

Generative AI, in all its simplicity, means that artificial intelligence has aided generation, be it text, image, or sound. The exciting part of Generative AI, however, is that this generation, which is done by AI, is modeled to make it indistinguishable from something a human would have generated.

Generative AI serves as a stepping stone for lines of code to, somewhere along the line, resulting in a singularity of generation of creative opinions in the form of Artificial General Intelligence. Unlike modern, narrow, purpose-driven models, AGI aims to exhibit human-like cognitive abilities.

Introduction to Large Language Models

Large Language Models (LLMs) are a section of Generative AI that particularly deals with the Generation of anything Text-Related. Tasks include Text Generation, Translation, Text Classification, and Document Indexing. LLMs have recently shown prowess by being deployed in chatbots and various other applications such as Grammarly, Duolingo, etc.

In the modern application of the concept, LLMs basically boil down to two types of technical buckets: GPT-based and LLaMA-based. GPT, coming from OpenAI, is the older and more battle-hardened model that works on concepts such as a Mixture of Experts and RAGs. It has inspired a series of enterprise, purpose-driven models that are only powered by their black box GPT-3 and GPT-4 APIs.

On the other hand, Llama takes a more Open-Source approach to the model, allowing developers and enthusiasts to play with the massive pre-trained model and create famous variations such as Vicuna, Stable Beluga, and Vigogne. With the wonders of LLMs comes the socio-economic impact of these models, so let us look at some of those.

Impact of LLMs

Before getting into the social impacts of Large Language Models, let us first try to comprehend how much money LLM startups such as OpenAI, Anthropic, and Stability.ai have been able to record in the year 2023.

Referring to this study by Reuters, we see that OpenAI increased its users from 57 Million in January 2023 to 150 Million by September. Talking about their pure financial front, all the product launches targetted Enterprises and Personal GPTs, the company’s revenue has jumped from 100 Million a year in 2022 to 1600 Million in the year 2023. Similarly, Anthropic saw an annual recurrent revenue of somewhere in the ballpark of 100 Million, which they project to reach up to 850 Million in 2024.

Regarding the social impact of Large Language Models, we have seen a noticeable uptick in technology usage from Fortune 100 to Startups. These models have revolutionized Content Generation and other aspects of textual information, whether sentiment analysis or classification.