Decoding AI Jargons with Garam Chai!

So we all have come across these “Jargons” in our lives right? Super annoying , seem meaningless and at times boring too!! But, there’s a side of us that wishes to know their meaning, understand the reason behind convention. Cause the truth is, once we become friends with such terms, our confidence on the subject grows massively . Suddenly we get the feeling like we can potentially own this territory!

Lets dive into some of these terms!!

1. Transformer: This is one of those terms that we might not be too familiar with at first glance. But definitely a term that we interact with on an EVERYDAY basis- Don’t believe me? Then what do you think ChatGPT stands for— Chat Generative Pre-Trained Transformer!! A Transformer is simple a network that takes an input text sequence and produces a text sequence as output. To dive a little deeper, these networks are Neural Networks. For eg - “Hello” in English becomes “Bonjour” in French , thanks to Transformer.

2. Tokenization - Okay so this one is plain and simple. Tokenization i basically just converting a given phrase or a sentence to a series of TOKENS! These Tokens are numbers that hold value with respect to the word they represent. Why do we do this? Cause the computer doesn’t understand English. But it loves dealing with numbers!

3. Vector Embeddings - Wow, seems like that’s 2 complex words, eh? NOT AT ALL ! It is one of the most interesting terms. Okay I’ll tell you how it works using an example, I’ll give you 3 words and you gotta guess the 4th word, okay? So, a) India b) New Delhi c) China …. and what would you say? And yes you’re right it is d) Beijing! So these patterns that your brain understands and processes to give the most probable next word - is what Vector Embeddings do! They create such patterns on a 3D scale to network big big sentences and paragraphs and help in producing the output!

4. Temperature - No it’s not the weather that AI is talking about here ;) Very simply put, it is just the creative parameter of the chatbot with respect to its response! When it is set at a higher number, then chatbot is elaborative and is more into your subject , while setting a lower value makes it kind of a less interactive, boring model.

5. Feed-Forward Neural Network - This one seems like a PROPER JARGON- with a very simple meaning. FNN is basically just a Neural Network where the information is only passed in the forward direction- from the input layer through the hidden layer and to the output layer! Okay so what is not happening here? BACKPROPAGATION! The model takes an input, lets say an image and processes to give an output, the caption of the image.

6. RAG Application - Retrieval Augmented Generation Application, I made it more difficult for ya didnt I ! Haha! Another very fun term to know about. So what we know about GPTs is that they are Pre-Trained! And that they cant generate real-time data. Now that is exactly what RAG Applications can do ! They generate real-time data by accessing external databases in real time to produce the output, that’s all.

So now you understand how interesting AI Jargons are !! Undoubtedly , AI holds one of the most Fun and Appealing concepts to learn and implement . Cant wait to get more and more deep into the world of Artificial Intelligence!

#chaicode