[ad_1]
There has been much discussion about Generative Artificial Intelligence (AI) and its implications. Clearly, its flexibility and usefulness in providing many of the tasks performed by humans is undeniable. However, there is also a lot of misinformation and dangers about it.
What is Generative AI? This word is made up of two things, AI and generative. AI is a term that describes how a computer program will perform tasks that would otherwise be performed by a human being. Generative is the fun part in that we’re creating new content, not necessarily having a computer look at it and be able to synthesize it and give us new things.
Generative involves creating new content (audio, code, images, text, video).
AI involves using computer programs to automate tasks.
Let’s look at text, an area called natural language processing and we’ll see how the technology works and hopefully uncover some myths and problems. Generative AI is not a new concept. Google Translator launched in 2006, so it’s been around for 17 years. Another example is Siri on the phone. It was launched in 2011. It was a sensation even then. This is another example of Generative AI.
In 2023, OpenAI, a San Francisco company, announced GPT-4. He claimed that he could get top scores in many exams like SAT, Law and Medical. Apart from examination, it can do many things. You can ask it to write a text for you or perform a task for you.
It is quite sophisticated and has created a sensation because it can do many things unlike the examples of Siri and Google Translator which only perform limited functions.
ChatGPT and its variants are based on the principle that I have some context, I will predict what will happen next. The function of the language model (LM) is that we have the context and we have a neural language model that will predict what the continuation is most likely to be. These are all realistic guesses and predictions of what is going to happen next. And that’s why sometimes they fail because they predict the most likely answer while you want the less likely answer.
So, you have this machinery that will learn for you and now the task is to predict the next word.
The question is how good can an LM become and what makes it great? Because when GPT came to GPT-1 and GPT-2, they weren’t surprising. So, the bigger, the better. I’m afraid size matters. There was a time when people did not believe in scale and now, we see that scale is very important. In fact, since 2018, we have seen a very significant increase in model size. Are big LMs always right or fair?
The simple answer is no. It is almost impossible to regulate the amount of material LLMs encounter during training. Because LLMs are trained on the web, they will always harbor historical biases and may reproduce harmful content. Generative AI has rightly generated buzz with its rapid growth in recent years. The point to be emphasized is that they give the most probable answer which may not be correct in some cases. However, concerns regarding the creation of deepfakes, the spread of misinformation, and potential job displacement are omnipresent. It is important to balance innovation with ethical considerations, implement safeguards against misuse, and address societal implications to maximize the benefits of generic AI while minimizing potential harms.
This article is written by Pankaj Jha, Consultant, Investment Banking and School Student Advisor.
[ad_2]
