Be A New Being

[Be A New Being @BeANewBeing]

Day 3 :

What is an LLM? How did these LLMs start how did they evolve as on today?

[Be A New Being @BeANewBeing]

Key Models:
Q-Learning, Deep Q-Networks (DQN), PPO, AlphaGo

Applications:
Robotics (e.g., robot arms)
Game AI (Go, Dota 2), autonomous driving
Dynamic pricing, recommendation systems

[Be A New Being @BeANewBeing]

Reinforcement Learning(RL) -learning through action

Goal: Train agents to learn by interacting with the environment
Built on: Trial-and-error learning + reward-based feedback

Evolution:
Complements deep learning by enabling decision making
Often combined with deep learning

[Be A New Being @BeANewBeing]

Key Models:
Transformer (2017), BERT, GPT, T5, RoBERTa

Applications:
Chatbots (ChatGPT), translation (Google Translate)
Summarisation, question answering

Limitation: Language-focused; doesn’t act in dynamic environments

[Be A New Being @BeANewBeing]

Natural Language Processing (NLP) with transformer-Language Intelligence
Goal: Understand and generate natural human language
Built on: Breakthrough in attention mechanism
Evolution:
Solves long dependency issues of RNN
Processes entire sequences in parallel using self-attention

[Be A New Being @BeANewBeing]

Key Models:
RNN, LSTM, GRU

Applications:
Text generation, sentiment analysis
Speech recognition, weather forecasting

Limitation: Struggles with long-term dependencies and parallelization

[Be A New Being @BeANewBeing]

Recurrent Neural Network (RNN) for understanding time and sequence

Goal: Process sequential data (time-series, language)
Built on: NN architecture + memory via feedback loops

Evolution:
Remembers past information via loops
Great for language, music, and time-sensitive data

[Be A New Being @BeANewBeing]

Key Models:
LeNet, AlexNet, VGG, ResNet

Applications:
Facial recognition, object detection
Medical image analysis, traffic sign detection

Limitation: Not suitable for sequence data (e.g., text, time series)

[Be A New Being @BeANewBeing]

Convolutional Neural Networks(CNN) for Vision Revolution

Goal: Improve accuracy on image and visual data
Built on: Neural networks + convolution operations

Evolution:
Specialised for spatial data (e.g., images)
Detects patterns like edges, shapes, textures

[Be A New Being @BeANewBeing]

Highlights:
First building block of modern AI
learns through layers of artificial neurons
Key Model:
Multi-Layer Perceptron (MLP)
Uses backpropagation for learning
Applications:
Handwriting digit recognition
Simple classification
Limitation: No effective images &sequence handling

[Be A New Being @BeANewBeing]

Day 2 :
Crunching the basics
Here is the evolution of AI from Neural networks to Reinforcement Learning improving the efficiency stage by stage
Neural Networks :

Goal: Enable machines to recognize patterns
📚 Foundation: Mathematical models inspired by the brain

[Be A New Being @BeANewBeing]

DL is the subset of ML that uses brain like structure called Neural networks to learn from Massive amounts of data.

Real life examples of AI:
Voice assistants of Siri/Alexa
Recommendations on Netflix/ YouTube
Google Maps traffic predictions
Self driving cars
Email spam filter

[Be A New Being @BeANewBeing]

These are 3 basic terms that are often used interchangeably when AI is the topic.. Artificial Intelligence(AI), Machine Learning(ML) and Deep Learning(DL).We now know what is AI. ML is the subset of AI that allows machines to learn from data without being explicitly programmed.

[Be A New Being @BeANewBeing]

Day 1:
What is Artificial Intelligence ?

AI refers to machines or computer programs that can perform tasks that required human intelligence. These tasks include understanding language, recognising images, solving problems, learning from experience and making decisions.

[Be A New Being @BeANewBeing]

Being accountable makes much difference in what we are learning and so is this blogging which I’ve never done before. Stepping out of my comfort zone and starting off with something I really wanted to learn and explore.. ‘Artificial Intelligence’.