Himanshu Ramchandani
Posts
𝟱 𝗔𝗹𝗮𝗿𝗺𝗶𝗻𝗴 𝘁𝗵𝗶𝗻𝗴𝘀 𝘆𝗼𝘂 𝗺𝘂𝘀𝘁 𝗸𝗻𝗼𝘄 𝗮𝗯𝗼𝘂𝘁 𝗚𝗲𝗻e𝗿𝗮𝘁𝗶𝘃𝗲𝗔𝗜 𝗮𝘀 𝗮 𝗟𝗲𝗮𝗱𝗲𝗿🚀

𝟱 𝗔𝗹𝗮𝗿𝗺𝗶𝗻𝗴 𝘁𝗵𝗶𝗻𝗴𝘀 𝘆𝗼𝘂 𝗺𝘂𝘀𝘁 𝗸𝗻𝗼𝘄 𝗮𝗯𝗼𝘂𝘁 𝗚𝗲𝗻e𝗿𝗮𝘁𝗶𝘃𝗲𝗔𝗜 𝗮𝘀 𝗮 𝗟𝗲𝗮𝗱𝗲𝗿🚀

Microsoft Copilot, GenAI for Leaders, ML Lead Jobs

Himanshu Ramchandani
September 24, 2023 • Estimated Reading Time: 4 minutes

Welcome Back, Hero!

Hope you are doing well.

Upgraded some things in the newsletter, hope you like it!

Today’s Content →

How to Leverage Data, Products & AI for Your Business 🏢
𝟱 𝗔𝗹𝗮𝗿𝗺𝗶𝗻𝗴 𝘁𝗵𝗶𝗻𝗴𝘀 𝘆𝗼𝘂 𝗺𝘂𝘀𝘁 𝗸𝗻𝗼𝘄 𝗮𝗯𝗼𝘂𝘁 𝗚𝗲𝗻e𝗿𝗮𝘁𝗶𝘃𝗲𝗔𝗜 𝗮𝘀 𝗮 𝗟𝗲𝗮𝗱𝗲𝗿🚀
1 Action Tip from Data Experts for Leaders 🎬→ Start Small
For Developers 🧑‍💻 → Microsoft Copilot
Career & Job in the AI field 🚀 → ML Lead

How to Leverage Data, Products & AI for Your Business 🏢

𝟱 𝗔𝗹𝗮𝗿𝗺𝗶𝗻𝗴 𝘁𝗵𝗶𝗻𝗴𝘀 𝘆𝗼𝘂 𝗺𝘂𝘀𝘁 𝗸𝗻𝗼𝘄 𝗮𝗯𝗼𝘂𝘁 𝗚𝗲𝗻e𝗿𝗮𝘁𝗶𝘃𝗲𝗔𝗜 𝗮𝘀 𝗮 𝗟𝗲𝗮𝗱𝗲𝗿🚀

GenerativeAI talk by Himanshu Ramchandani

GenerativeAI Revolution

1 — Behind NLP

→ The objective of NLP is to make computers understand what we are trying to day.

→ We have to make sure it understands 2 things - context and intent.

2 — Context and Intent

Context is if I tell you - “Your T-shirt is killer”

We as humans take this example as a compliment.

The “killer” word for the computer is a negative.

But the context is positive.

Suppose you tell a vegetable seller that you got the money from your mother, and she wants me to buy tomatoes from the market, otherwise, she will be angry.

In the whole part, the vegetable seller will understand that you need tomatoes.

The whole story behind it is not important to the seller.

Thats Intent!

Real-world examples of NLP —

- spam detection
- social media sentiment analysis
- machine translation
- text summarization
- semantic search

3 — GenerativeAI had 2 biggest milestones

1 - Artificial Neurons

In deep learning, millions of these are used in dense neural networks.

2 - Pre-trained models

We can reuse the model weights so that we don't have to train it again and again.

4 — GPT-3 - let’s break it down

✅Generative

→ Generative means generating text (GPT3).
→ Generative Modeling is the branch of statistical modeling.
→ In order for the model to generate text similar to ours, we need a lot of data.

✅Pre-Trained

→ Training a model takes a lot of time as well as resources along with cost.
→ Using our trained model will allow us to fine-tune it for downstream tasks.
→ Understanding chess rules will help us compete with others in a match.

GPT3 is trained on 5 datasets
- Common Crawl
- WebText2
- Books1
- Books2
- Wikipedia.

✅Transformers

→ Transformers work on attention mechanisms.

Example - “Lakshya is sitting on a chair. He will take the lunch now”

In this statement, we know that the word “He” is used for Lakshya. That's attention.

→ Transformer works on the sequence-to-sequence architecture.

→ A sequence of words in one language is translated into a sequence of words in another language.

That’s all, GPT

5 — The versions

The version 3 has 175 billion text parameters.

GPT-2 had 1.5 billion parameters
GPT-1 had 117 million parameters

PS → Reach out to me if you need help with GenAI for product development, Data Science, Data Engineering, and Infra on Cloud → https://topmate.io/himanshuramchandani/447594