Madhan @madhan_code

@AdityaThakurGG IITians wont coming to you with less than 1lakh stipend

@karanveer04 Can you refer me sir?

is anyone taking flipkart’s grid oa today?

@ai_for_success Comet better

@only1wilsn Deep learning

still useless. nothing has changed

Rhys @RhysSullivan

a year ago

incredible how pretty much the entire github homepage is useless

471 858 20K 2.9M 1K

@NoahEpstein_ SKILLS

The most frequently used Mathematics concepts in AI & ML: - Probability Distribution Functions (PDFs) - Singular Value Decomposition(SVD) - KL Divergence Every beginner needs to know these concepts to understand many Machine Learning, Deep Learning, and Reinforcement Learning concepts or models. Plus, these are very important concepts for job interviews as well. To give you a quick idea of what, how, and when of these concepts: 1. Probability Distribution Functions (PDFs): PDFs are foundational in ML and DL. Understanding them helps you reason about uncertainty, model assumptions, generative processes, and even design loss functions. A PDF describes how probability mass or density is distributed over the possible values of a random variable. There are two kinds: i. Discrete distributions: Use a Probability Mass Function (PMF) ii. Continuous distributions: Use a Probability Density Function (PDF) We mostly use or deal with Continuous distributions, especially the Normal(Gaussian) Distribution. Use cases: - Maximum Likelihood Estimation (MLE), - Bayesian Inference, Generative Models (VAEs, GANs, GMMs), - Information Theory(Cross-entropy in Decision Tree and KL Divergence are based on PDFs) - Anomaly detection, - Loss functions 2. Singular Value Decomposition(SVD): It is a fundamental matrix factorization technique in linear algebra, widely used in machine learning for purposes like Matrix Factorization and Dimensionality Reduction. Given any real (or complex) matrix A of size m×n, SVD decomposes it into three matrices: A=UΣV^T Where: A: Original matrix (size m×n) U: Left singular vectors (orthogonal matrix, size m×m) Σ: Diagonal matrix with singular values (size m×n); non-negative and sorted in decreasing order V^T: Transpose of right singular vectors (orthogonal matrix, size n×n) Intuition: SVD expresses any matrix A as: "Rotate(V) → Stretch(Σ) → Rotate(U)" Use cases: - Dimensionality reduction, - Noise reduction, - Recommender systems, - Semantic Analysis in NLP, - Image compression, etc. 3. Kullback–Leibler(KL) Divergence KL Divergence is extremely important in machine learning and deep learning. It's a core mathematical concept used in probability distributions, information theory, and optimization. KL Divergence is a measure of how one probability distribution differs from another. For two distributions, P (true distribution) and Q (approximate or predicted distribution), the KL divergence D(P∥Q)=∑​P(x)log(P(x)/P(x)​) KL Divergence acts as a loss function, regularizer, or objective to minimize when training models that deal with probability distributions. Use cases: - Variational Inference (VI), to approximate intractable posteriors. - Variational Auto Encoders(VAEs), GANs, - Policy Optimization in Reinforcement Learning, - Language Modeling & NLP, - Knowledge Distillation in LLM etc., (you must have heard "Distilled version of Llama3 or Qwen2.5", etc. - Bayesian Deep Learning Will share the best learning resources for these topics in upcoming posts if you need. Let me know. PS. There are other frequently used maths concepts too, but these three are the most used ones.

@sadhna6389 Send

@shydev69 People still learning github?

@shydev69 @adiautomates Aunty-national

@shivangitwt You forgot to hide your face in linkedin lol

@ajonit @lostinpadania @PerplexityComet @perplexity_ai @AravSrinivas Can you invite me?

@chrisuniverse @PerplexityComet Mee

This. #WWDC25

As predicted, unifying version numbers

I’m 18 I don’t party I don’t fool around I don’t waste time with friends I read I exercise I meditate I run my company That’s just me.

Did my posts show in your timeline?

Starting AI and ML, any piece of advice or good resources to learn?

How AI "Speaks" Our Language: Unlocking the Secrets of Large Language Models Have you ever wondered how artificial intelligence (AI) can chat with us, write stories, or answer questions like a human? The answer lies in Large Language Models (LLMs); incredible systems that power tools like chatbots and virtual assistants. At their core, LLMs predict the next word in a sentence, kind of like a supercharged autocorrect. But while your phone might guess "the" after "I went to," LLMs take it to another level by analyzing the whole conversation to make smarter, more natural predictions. Let’s dive into how they work in a way anyone can understand! Inside an LLM is something called a neural network, which you can picture as a giant web of tiny "brain cells" connected together. These networks are trained on massive piles of text; think millions of books, websites, and social media posts. By studying all this data, the AI learns patterns in language, like how "rain" often pairs with "clouds" or how "How are you?" usually gets a "Fine, thanks!" It’s like the AI builds a mental map of how words and ideas link up, so it can string together sentences that actually make sense. Here’s where it gets cool: context is everything. When you talk to an LLM, it doesn’t just look at the last word you said; it considers everything you’ve written so far. Say you ask, "What’s the best way to cook a turkey?" The model zeroes in on "cook" and "turkey" and predicts something like "roast" or "fry" based on what fits. It’s like handing the AI a treasure map, and your words are the clues guiding it to the right answer. Without context, it’d be lost, but with it, the possibilities start to narrow into something useful. Now, imagine every word the LLM picks is a fork in the road. After "I love to," it could go "eat," "travel," or "dance"; each choice leading to a different path. As it predicts word after word, it’s like exploring a maze of possibilities, building sentences that could turn into a recipe, a poem, or a history lesson. This is why LLMs can be so creative; they’re not stuck on one track but can wander through language in endless ways, adapting to whatever you throw at them. But it’s not all magic; there’s a catch. LLMs learn from human-written text, and that text isn’t perfect. If the data has biases (like stereotypes or outdated ideas), the AI might pick them up too, leading to mistakes or unfair answers. Plus, they don’t truly understand the world; they’re just really good at guessing patterns. Researchers are working hard to fix these flaws, making LLMs more accurate and ethical every day. So next time you chat with an AI, think of it as a brilliant word-guessing wizard, weaving responses from a web of learned patterns, guided by your input. #AI #TechForAll #LanguageModels #MachineLearning #ExplainItSimple