Loading video...

Video Failed to Load

Go Home

Holy shit… someone just made machine learning click. Not static diagrams. Not math-heavy PDFs. Not black-box training. Real algorithms — training step-by-step — visually. It’s called Machine Learning Visualized and it lets you watch models learn in real time. Here’s why this is different: Instead of dumping theory first,...

Suryanshti777's profile picture

Suryansh Tiwari

44,819 subscribers

132,607 views • 4 months ago •via X (Twitter)

EducationScience & Technology
LIVE

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

a playlist of 30 youtube videos to learn machine learning fundamentals from scratch if you're struggling on where to start learning ML, this list goes this "Machine Learning: Teach by Doing" is a solid choice to learn both theory and code. (1) Introduction to Machine Learning Teach by Doing: (2) What is Machine Learning? History of Machine Learning: (3) Types of ML Models: (4) 6 steps of any ML project: (5) Install Python and VSCode and run your first code: (6) Linear Classifiers Part 1: (7) Linear Classifiers Part 2: (8) Jupyter Notebook, Numpy and Scikit-Learn: (9) Running the Random Linear Classifier Algorithm in Python: (10) The oldest ML model - Perceptron: (11) Coding the Perceptron: (12) Perceptron Convergence Theorem: (13) Magic of features in Machine Learning: (14) One hot encoding: (15) Logistic Regression Part 1: (16) Cross Entropy Loss: (17) How gradient descent works: (18) Logistic Regression from scratch in Python: (19) Introduction to Regularization: (20) Implementing Regularization in Python: (21) Linear Regression Introduction: (22) Ordinary Least Squares step by step implementation: (23) Ridge regression fundamentals and intuition: (24) Regression recap for interviews: (25) Neural network architecture in 30 minutes: (26) Backpropagation intuition: (27) Neural network activation functions: (28) Momentum in gradient descent: (29) Hands on neural network training in Python: (30) Introduction to Convolutional Neural Networks (CNNs):
0:52

a playlist of 30 youtube videos to learn machine learning fundamentals from scratch if you're struggling on where to start learning ML, this list goes this "Machine Learning: Teach by Doing" is a solid choice to learn both theory and code. (1) Introduction to Machine Learning Teach by Doing: (2) What is Machine Learning? History of Machine Learning: (3) Types of ML Models: (4) 6 steps of any ML project: (5) Install Python and VSCode and run your first code: (6) Linear Classifiers Part 1: (7) Linear Classifiers Part 2: (8) Jupyter Notebook, Numpy and Scikit-Learn: (9) Running the Random Linear Classifier Algorithm in Python: (10) The oldest ML model - Perceptron: (11) Coding the Perceptron: (12) Perceptron Convergence Theorem: (13) Magic of features in Machine Learning: (14) One hot encoding: (15) Logistic Regression Part 1: (16) Cross Entropy Loss: (17) How gradient descent works: (18) Logistic Regression from scratch in Python: (19) Introduction to Regularization: (20) Implementing Regularization in Python: (21) Linear Regression Introduction: (22) Ordinary Least Squares step by step implementation: (23) Ridge regression fundamentals and intuition: (24) Regression recap for interviews: (25) Neural network architecture in 30 minutes: (26) Backpropagation intuition: (27) Neural network activation functions: (28) Momentum in gradient descent: (29) Hands on neural network training in Python: (30) Introduction to Convolutional Neural Networks (CNNs):

ℏεsam

117,570 views • 1 year ago

if you're struggling on where to start learning ML, here’s a playlist of 30 youtube videos to learn machine learning fundamentals from scratch "Machine Learning: Teach by Doing" is a solid choice to learn both theory and code. (1) Introduction to Machine Learning Teach by Doing: (2) What is Machine Learning? History of Machine Learning: (3) Types of ML Models: (4) 6 steps of any ML project: (5) Install Python and VSCode and run your first code: (6) Linear Classifiers Part 1: (7) Linear Classifiers Part 2: (8) Jupyter Notebook, Numpy and Scikit-Learn: (9) Running the Random Linear Classifier Algorithm in Python: (10) The oldest ML model - Perceptron: (11) Coding the Perceptron: (12) Perceptron Convergence Theorem: (13) Magic of features in Machine Learning: (14) One hot encoding: (15) Logistic Regression Part 1: (16) Cross Entropy Loss: (17) How gradient descent works: (18) Logistic Regression from scratch in Python: (19) Introduction to Regularization: (20) Implementing Regularization in Python: (21) Linear Regression Introduction: (22) Ordinary Least Squares step by step implementation: (23) Ridge regression fundamentals and intuition: (24) Regression recap for interviews: (25) Neural network architecture in 30 minutes: (26) Backpropagation intuition: (27) Neural network activation functions: (28) Momentum in gradient descent: (29) Hands on neural network training in Python: (30) Introduction to Convolutional Neural Networks (CNNs):
0:52

if you're struggling on where to start learning ML, here’s a playlist of 30 youtube videos to learn machine learning fundamentals from scratch "Machine Learning: Teach by Doing" is a solid choice to learn both theory and code. (1) Introduction to Machine Learning Teach by Doing: (2) What is Machine Learning? History of Machine Learning: (3) Types of ML Models: (4) 6 steps of any ML project: (5) Install Python and VSCode and run your first code: (6) Linear Classifiers Part 1: (7) Linear Classifiers Part 2: (8) Jupyter Notebook, Numpy and Scikit-Learn: (9) Running the Random Linear Classifier Algorithm in Python: (10) The oldest ML model - Perceptron: (11) Coding the Perceptron: (12) Perceptron Convergence Theorem: (13) Magic of features in Machine Learning: (14) One hot encoding: (15) Logistic Regression Part 1: (16) Cross Entropy Loss: (17) How gradient descent works: (18) Logistic Regression from scratch in Python: (19) Introduction to Regularization: (20) Implementing Regularization in Python: (21) Linear Regression Introduction: (22) Ordinary Least Squares step by step implementation: (23) Ridge regression fundamentals and intuition: (24) Regression recap for interviews: (25) Neural network architecture in 30 minutes: (26) Backpropagation intuition: (27) Neural network activation functions: (28) Momentum in gradient descent: (29) Hands on neural network training in Python: (30) Introduction to Convolutional Neural Networks (CNNs):

ℏεsam

109,292 views • 1 year ago

Holy shit… someone just made DSA finally click. Not static notes Not boring pseudocode Not guessing what happens in memory Real data structures — animating step-by-step — visually. It’s called Data Structure Visualizations and it lets you watch algorithms run in real time. Here’s why this is different: Instead of dumping theory, it shows execution live • nodes getting inserted • trees rotating • pointers moving • queues filling • stacks popping • graphs traversing • heaps rebalancing You literally see algorithms think. Everything is interactive: • Binary Search Trees • AVL Trees (with rotations) • Red-Black Trees • Heaps & Priority Queues • Graph BFS / DFS • Dijkstra & MST • Hash Tables • Tries • Sorting (Quick, Merge, Heap…) • Dynamic Programming No black box. Just input → steps → result Watch in real time: • AVL rotations balancing themselves • BFS exploring layer by layer • DFS diving deep then backtracking • Dijkstra relaxing edges step-by-step • Quick sort partition visually • Heap forming after each insert • Hash collisions resolving live This solves the biggest DSA problem: Most resources teach code → memorize → hope it works This shows input → execution → visualization → understanding Which means you finally understand: • why AVL rotates • how heap property maintains • how BFS differs from DFS • how Dijkstra actually updates distances • what happens during rehashing • how quicksort partitions • how trees rebalance Even better: You control everything Change values Insert nodes Run step-by-step Pause execution Replay algorithms Learning DSA becomes interactive Not passive Not confusing Not theoretical Just… visible. Perfect for: • DSA beginners • interview prep • visual learners • CS students • LeetCode prep • teaching algorithms • debugging understanding This is the kind of resource that makes trees, graphs, and sorting finally click. Link: We’re moving from reading DSA → watching DSA execute And once you can see algorithms… you stop memorizing and start understanding.
1:21

Holy shit… someone just made DSA finally click. Not static notes Not boring pseudocode Not guessing what happens in memory Real data structures — animating step-by-step — visually. It’s called Data Structure Visualizations and it lets you watch algorithms run in real time. Here’s why this is different: Instead of dumping theory, it shows execution live • nodes getting inserted • trees rotating • pointers moving • queues filling • stacks popping • graphs traversing • heaps rebalancing You literally see algorithms think. Everything is interactive: • Binary Search Trees • AVL Trees (with rotations) • Red-Black Trees • Heaps & Priority Queues • Graph BFS / DFS • Dijkstra & MST • Hash Tables • Tries • Sorting (Quick, Merge, Heap…) • Dynamic Programming No black box. Just input → steps → result Watch in real time: • AVL rotations balancing themselves • BFS exploring layer by layer • DFS diving deep then backtracking • Dijkstra relaxing edges step-by-step • Quick sort partition visually • Heap forming after each insert • Hash collisions resolving live This solves the biggest DSA problem: Most resources teach code → memorize → hope it works This shows input → execution → visualization → understanding Which means you finally understand: • why AVL rotates • how heap property maintains • how BFS differs from DFS • how Dijkstra actually updates distances • what happens during rehashing • how quicksort partitions • how trees rebalance Even better: You control everything Change values Insert nodes Run step-by-step Pause execution Replay algorithms Learning DSA becomes interactive Not passive Not confusing Not theoretical Just… visible. Perfect for: • DSA beginners • interview prep • visual learners • CS students • LeetCode prep • teaching algorithms • debugging understanding This is the kind of resource that makes trees, graphs, and sorting finally click. Link: We’re moving from reading DSA → watching DSA execute And once you can see algorithms… you stop memorizing and start understanding.

Suryansh Tiwari

14,425 views • 4 months ago

you tend to hear this a lot from people outside or new to ML, and I often point to a talk Ilya gave a few years back: 1) think of any decent deep neural net that has enough memory and sequential ops as just a big parallel computer 2) training this neural net is doing search over computer programs that maximize your objective 3)unless you have some large bottleneck (and given you can successfully optimize this system) you’ll find that these parallel computers are highly robust to architectural changes. 4) this is because computers are great at simulating each other. your new architecture can usually be straightforwardly simulated ‘inside’ your old architecture. 5) it’s not that architecture doesn’t matter, but it mostly matters with respect to (1) fundamental bottlenecks in this parallel computer (2) modifications that make models easier to optimize, since this argument only holds if your optimization is good (3) compute efficiency/system efficiency wins that make learning easier or faster. 6) it’s quite possible that new architectures will lead to breakthroughs in machine learning, but we should first start with bottlenecks, not naturalist intuitions about the ‘form’ of AI should take. until you understand this it seems surprising that small models trained longer are better than undertrained big models, that depth and width are surprisingly interchangeable, that talking to a model with an MoE or sparse attention or linear attention is approximately the same iso evals.
1:12

you tend to hear this a lot from people outside or new to ML, and I often point to a talk Ilya gave a few years back: 1) think of any decent deep neural net that has enough memory and sequential ops as just a big parallel computer 2) training this neural net is doing search over computer programs that maximize your objective 3)unless you have some large bottleneck (and given you can successfully optimize this system) you’ll find that these parallel computers are highly robust to architectural changes. 4) this is because computers are great at simulating each other. your new architecture can usually be straightforwardly simulated ‘inside’ your old architecture. 5) it’s not that architecture doesn’t matter, but it mostly matters with respect to (1) fundamental bottlenecks in this parallel computer (2) modifications that make models easier to optimize, since this argument only holds if your optimization is good (3) compute efficiency/system efficiency wins that make learning easier or faster. 6) it’s quite possible that new architectures will lead to breakthroughs in machine learning, but we should first start with bottlenecks, not naturalist intuitions about the ‘form’ of AI should take. until you understand this it seems surprising that small models trained longer are better than undertrained big models, that depth and width are surprisingly interchangeable, that talking to a model with an MoE or sparse attention or linear attention is approximately the same iso evals.

will depue

215,174 views • 8 months ago

STANFORD JUST PUT ITS ENTIRE ARTIFICIAL INTELLIGENCE CURRICULUM ON YOUTUBE FOR FREE. CS221. The same course that produced engineers now running AI labs, building frontier models, and getting paid $500,000 a year at the companies everyone is trying to work for. Most people have never heard of it. The ones who have are not telling you about it. Here is what the course actually covers: Search algorithms. The mathematical foundation behind every AI that finds optimal solutions in complex environments. Constraint satisfaction. How AI reasons through problems with thousands of interdependent variables simultaneously. Markov decision processes. The probabilistic framework behind every AI agent that makes sequential decisions under uncertainty. Machine learning from first principles. Not how to use sklearn. How the math actually works underneath it. Neural networks. Built from the ground up before jumping to applications. Logic and knowledge representation. How AI systems reason about the world formally. Natural language processing. The foundation of everything happening in LLMs right now. Robotics and computer vision. How AI perceives and acts in physical environments. Every concept that powers every AI product you use daily is in this curriculum. Not a surface level overview. The actual mathematics. The actual algorithms. The actual reasoning. This is what separates engineers who build AI from operators who use it. Stanford charged $60,000 a year for students to sit in this classroom. They put the whole thing on YouTube. Bookmark this before you open any other AI resource today. Follow CyrilXBT for more elite resources that build real depth the moment they drop.
1:27:25

STANFORD JUST PUT ITS ENTIRE ARTIFICIAL INTELLIGENCE CURRICULUM ON YOUTUBE FOR FREE. CS221. The same course that produced engineers now running AI labs, building frontier models, and getting paid $500,000 a year at the companies everyone is trying to work for. Most people have never heard of it. The ones who have are not telling you about it. Here is what the course actually covers: Search algorithms. The mathematical foundation behind every AI that finds optimal solutions in complex environments. Constraint satisfaction. How AI reasons through problems with thousands of interdependent variables simultaneously. Markov decision processes. The probabilistic framework behind every AI agent that makes sequential decisions under uncertainty. Machine learning from first principles. Not how to use sklearn. How the math actually works underneath it. Neural networks. Built from the ground up before jumping to applications. Logic and knowledge representation. How AI systems reason about the world formally. Natural language processing. The foundation of everything happening in LLMs right now. Robotics and computer vision. How AI perceives and acts in physical environments. Every concept that powers every AI product you use daily is in this curriculum. Not a surface level overview. The actual mathematics. The actual algorithms. The actual reasoning. This is what separates engineers who build AI from operators who use it. Stanford charged $60,000 a year for students to sit in this classroom. They put the whole thing on YouTube. Bookmark this before you open any other AI resource today. Follow CyrilXBT for more elite resources that build real depth the moment they drop.

CyrilXBT

54,956 views • 3 months ago

Andrej Karpathy says you should learn AI depthwise, not breadthwise. Most education is breadthwise: watch lectures, memorize formulas, and trust you'll need it later. Karpathy flips this by learning "depthwise, on demand." What this means: Pick a project, start building, and learn exactly when you hit a wall. When he created a tutorial on transformers (the architecture behind ChatGPT), he didn't start by explaining attention mechanisms or complex architectures. Instead, he started with the simplest possible thing: a lookup table that predicts the next word. You build that first. Then you try to make it handle more complex patterns. And it breaks. Only then, when you've felt the limitation, does he introduce the next concept. Each piece solves a problem you've actually encountered. As he puts it: "It's a dick move to present the solution before I give you a shot to try it yourself." When you attempt the problem first, the solution actually makes sense. Teaching forces you to learn. "If I don't really understand something, I can't explain it." When you try to explain and stumble, you've found the gaps in your understanding. ... Build a project that gives you a reward. Hit a wall. Learn just enough to solve it. Then explain it to someone else. Don't consume content. Build the code. That's how you actually learn.
1:28

Andrej Karpathy says you should learn AI depthwise, not breadthwise. Most education is breadthwise: watch lectures, memorize formulas, and trust you'll need it later. Karpathy flips this by learning "depthwise, on demand." What this means: Pick a project, start building, and learn exactly when you hit a wall. When he created a tutorial on transformers (the architecture behind ChatGPT), he didn't start by explaining attention mechanisms or complex architectures. Instead, he started with the simplest possible thing: a lookup table that predicts the next word. You build that first. Then you try to make it handle more complex patterns. And it breaks. Only then, when you've felt the limitation, does he introduce the next concept. Each piece solves a problem you've actually encountered. As he puts it: "It's a dick move to present the solution before I give you a shot to try it yourself." When you attempt the problem first, the solution actually makes sense. Teaching forces you to learn. "If I don't really understand something, I can't explain it." When you try to explain and stumble, you've found the gaps in your understanding. ... Build a project that gives you a reward. Hit a wall. Learn just enough to solve it. Then explain it to someone else. Don't consume content. Build the code. That's how you actually learn.

Ruben Hassid

403,691 views • 9 months ago

This 6-minute video reveals how Elon Musk learns complex topics: Elon Musk: “You don’t need college for learning.” “Everything is available basically for free. You can learn anything you want for free. It is not a question of learning.” Musk starts with a blunt point: College may still have value, but not for the reason most people think. He says the real signal of college is not intelligence. It is proof that someone can work through structure: “Can somebody work hard at something, including a bunch of sort of annoying homework assignments, and still do their homework assignments, and kind of soldier through and get it done?” That, in his view, is one of the main things a degree demonstrates: Discipline. Compliance. Follow-through. Not necessarily exceptional ability. Musk pushes the idea even further: “Colleges are basically for fun and to prove you can do your chores. But they’re not for learning.” Whether or not you agree with him fully, the underlying point is hard to ignore: We live in a time when knowledge is no longer locked inside institutions. The internet has dismantled the old gatekeeping model. Today, if someone wants to learn design, engineering, writing, sales, coding, marketing, or history, they can access world-class information without ever stepping into a lecture hall. The bottleneck is no longer access to information. It is desire. Focus. Curiosity. Consistency. Musk then draws a distinction that matters: “If you’re trying to do something exceptional, there must be evidence of exceptional ability.” That line changes the whole conversation. Because in real life, people do not reward credentials alone. They reward proof. Not what you enrolled in. What you built. Not what you intended to do. What you finished. Not what you say you know. What you can demonstrate. This is why portfolios outperform claims. Why execution beats prestige. Why visible work creates leverage. Musk even says, somewhat provocatively: “I don’t consider going to college evidence of exceptional ability.” And then he points to the kinds of examples people love to cite: “Gates is a pretty smart guy, he dropped out. John was pretty smart, he dropped out. Larry Ellison, smart guy, he dropped out.” His broader message is not that everyone should leave school. It is that conventional paths are not the only paths to intelligence, capability, or impact. Then Musk moves into something even more useful: His view of how people actually learn. “Education should be as close to a video game as possible. Like a good video game. You do not need to tell your kid to play video games. They will play video games on autopilot all day.” That comparison is simple, but powerful. Why do people obsess over games? Because games are interactive. They are immersive. They provide immediate feedback. They make progress visible. They create challenge without making the challenge feel meaningless. Musk’s point is that learning should work the same way. “If you can make it interactive and engaging, then you can make education far more compelling and far easier to do.” This is where traditional education often breaks down. Students are expected to move in lockstep. Same pace. Same timeline. Same structure. Same sequence. Musk rejects that model completely: “People are not objects on an assembly line.” That may be one of the clearest criticisms in the entire transcript. Because standard education often optimizes for administration, not human variation. It is easier to manage people in batches. But easier to manage does not mean better to learn. Some people move faster in math. Some are stronger in language. Some are highly visual. Some need to touch the thing, build the thing, test the thing. And yet most systems still treat learning like synchronized marching. Musk argues for something more individualized: “Allow people to progress at the fastest pace that they can or are interested in in each subject.” That idea matters beyond school. Adults learn this way too. No one becomes exceptional by waiting for permission to move at average speed. The most effective learners usually follow interest with intensity. They go deeper where curiosity pulls them. They accelerate where energy is highest. They build momentum through engagement, not force. Musk also shares one of the most practical ideas in the transcript: “Teach problem solving, or teach to the problem, not to the tools.” Then he gives an example. If you wanted to teach someone how engines work, the traditional system might start with separate lessons on screwdrivers, wrenches, and tools. Musk thinks that is backwards. A better method is: “Here’s the engine. Now let’s take it apart.” Then the tools become relevant in context. Now the student understands *why* the screwdriver matters. Now the wrench has meaning. Now the lesson is connected to reality. This is a much bigger principle than education. People learn faster when relevance is obvious. Abstract instruction is forgettable. Applied learning sticks. When people can see the problem first, they care about the tool. That is true in business too. You do not start with theory for theory’s sake. You start with the problem that needs solving. Then you learn exactly what is required to solve it. Finally, Musk says something that quietly explains why so much education fails: “A lot of things people learn, probably there’s no point in learning them because they never use them in the future.” That may sound harsh, but most people know the feeling. They do not resist learning because they are lazy. They resist learning because it feels disconnected. They are told to memorize before they understand relevance. They are told to sit still before they become curious. They are told to absorb information before they have any reason to care. Musk’s view, underneath the provocation, is actually simple: People learn best when learning is alive. When it is tied to action. When it respects differences in pace and aptitude. When it feels engaging instead of ceremonial. When it produces visible competence, not just paper credentials. The internet made learning abundant. What matters now is whether someone can turn information into evidence. That is the real separator. Lessons I'm taking away from this clip: 1. In today’s world, access to knowledge is cheap. Proof of skill is expensive. We have crossed a point where information alone is no longer impressive because everyone has access to it. You can watch the best interviews, read the best essays, take the best online lessons, and still remain average if you never turn any of it into real work. So the advantage now is not “I know this.” The advantage is “I built this, tested this, shipped this, and can show the result.” From my perspective, this is especially true in business and personal branding. The market rewards visible competence far more than silent knowledge. 2. People learn faster when the learning feels useful, alive, and connected to a real problem. This is why so many people struggle with conventional education but thrive when they start building something for themselves. Urgency creates focus. Relevance creates retention. Once the lesson is attached to a real outcome, the brain pays attention differently. That’s why I think one of the best ways to learn anything is to start a project that forces you to use the skill in public or in real life. Learning becomes sharper when there is something at stake. It stops being passive consumption and becomes active problem-solving. 3. The smartest people are often not the ones collecting credentials. They are the ones following curiosity with discipline. Exceptional people usually do not just learn what is assigned to them. They go where their interest is strongest and then they pursue it seriously. That combination matters: curiosity without discipline goes nowhere, and discipline without curiosity becomes lifeless. The sweet spot is when someone becomes obsessed enough to keep going deeper than required. To me, that is where the real edge comes from. Not from following the default path better than everyone else, but from developing uncommon depth in something that genuinely pulls you.
6:13

This 6-minute video reveals how Elon Musk learns complex topics: Elon Musk: “You don’t need college for learning.” “Everything is available basically for free. You can learn anything you want for free. It is not a question of learning.” Musk starts with a blunt point: College may still have value, but not for the reason most people think. He says the real signal of college is not intelligence. It is proof that someone can work through structure: “Can somebody work hard at something, including a bunch of sort of annoying homework assignments, and still do their homework assignments, and kind of soldier through and get it done?” That, in his view, is one of the main things a degree demonstrates: Discipline. Compliance. Follow-through. Not necessarily exceptional ability. Musk pushes the idea even further: “Colleges are basically for fun and to prove you can do your chores. But they’re not for learning.” Whether or not you agree with him fully, the underlying point is hard to ignore: We live in a time when knowledge is no longer locked inside institutions. The internet has dismantled the old gatekeeping model. Today, if someone wants to learn design, engineering, writing, sales, coding, marketing, or history, they can access world-class information without ever stepping into a lecture hall. The bottleneck is no longer access to information. It is desire. Focus. Curiosity. Consistency. Musk then draws a distinction that matters: “If you’re trying to do something exceptional, there must be evidence of exceptional ability.” That line changes the whole conversation. Because in real life, people do not reward credentials alone. They reward proof. Not what you enrolled in. What you built. Not what you intended to do. What you finished. Not what you say you know. What you can demonstrate. This is why portfolios outperform claims. Why execution beats prestige. Why visible work creates leverage. Musk even says, somewhat provocatively: “I don’t consider going to college evidence of exceptional ability.” And then he points to the kinds of examples people love to cite: “Gates is a pretty smart guy, he dropped out. John was pretty smart, he dropped out. Larry Ellison, smart guy, he dropped out.” His broader message is not that everyone should leave school. It is that conventional paths are not the only paths to intelligence, capability, or impact. Then Musk moves into something even more useful: His view of how people actually learn. “Education should be as close to a video game as possible. Like a good video game. You do not need to tell your kid to play video games. They will play video games on autopilot all day.” That comparison is simple, but powerful. Why do people obsess over games? Because games are interactive. They are immersive. They provide immediate feedback. They make progress visible. They create challenge without making the challenge feel meaningless. Musk’s point is that learning should work the same way. “If you can make it interactive and engaging, then you can make education far more compelling and far easier to do.” This is where traditional education often breaks down. Students are expected to move in lockstep. Same pace. Same timeline. Same structure. Same sequence. Musk rejects that model completely: “People are not objects on an assembly line.” That may be one of the clearest criticisms in the entire transcript. Because standard education often optimizes for administration, not human variation. It is easier to manage people in batches. But easier to manage does not mean better to learn. Some people move faster in math. Some are stronger in language. Some are highly visual. Some need to touch the thing, build the thing, test the thing. And yet most systems still treat learning like synchronized marching. Musk argues for something more individualized: “Allow people to progress at the fastest pace that they can or are interested in in each subject.” That idea matters beyond school. Adults learn this way too. No one becomes exceptional by waiting for permission to move at average speed. The most effective learners usually follow interest with intensity. They go deeper where curiosity pulls them. They accelerate where energy is highest. They build momentum through engagement, not force. Musk also shares one of the most practical ideas in the transcript: “Teach problem solving, or teach to the problem, not to the tools.” Then he gives an example. If you wanted to teach someone how engines work, the traditional system might start with separate lessons on screwdrivers, wrenches, and tools. Musk thinks that is backwards. A better method is: “Here’s the engine. Now let’s take it apart.” Then the tools become relevant in context. Now the student understands *why* the screwdriver matters. Now the wrench has meaning. Now the lesson is connected to reality. This is a much bigger principle than education. People learn faster when relevance is obvious. Abstract instruction is forgettable. Applied learning sticks. When people can see the problem first, they care about the tool. That is true in business too. You do not start with theory for theory’s sake. You start with the problem that needs solving. Then you learn exactly what is required to solve it. Finally, Musk says something that quietly explains why so much education fails: “A lot of things people learn, probably there’s no point in learning them because they never use them in the future.” That may sound harsh, but most people know the feeling. They do not resist learning because they are lazy. They resist learning because it feels disconnected. They are told to memorize before they understand relevance. They are told to sit still before they become curious. They are told to absorb information before they have any reason to care. Musk’s view, underneath the provocation, is actually simple: People learn best when learning is alive. When it is tied to action. When it respects differences in pace and aptitude. When it feels engaging instead of ceremonial. When it produces visible competence, not just paper credentials. The internet made learning abundant. What matters now is whether someone can turn information into evidence. That is the real separator. Lessons I'm taking away from this clip: 1. In today’s world, access to knowledge is cheap. Proof of skill is expensive. We have crossed a point where information alone is no longer impressive because everyone has access to it. You can watch the best interviews, read the best essays, take the best online lessons, and still remain average if you never turn any of it into real work. So the advantage now is not “I know this.” The advantage is “I built this, tested this, shipped this, and can show the result.” From my perspective, this is especially true in business and personal branding. The market rewards visible competence far more than silent knowledge. 2. People learn faster when the learning feels useful, alive, and connected to a real problem. This is why so many people struggle with conventional education but thrive when they start building something for themselves. Urgency creates focus. Relevance creates retention. Once the lesson is attached to a real outcome, the brain pays attention differently. That’s why I think one of the best ways to learn anything is to start a project that forces you to use the skill in public or in real life. Learning becomes sharper when there is something at stake. It stops being passive consumption and becomes active problem-solving. 3. The smartest people are often not the ones collecting credentials. They are the ones following curiosity with discipline. Exceptional people usually do not just learn what is assigned to them. They go where their interest is strongest and then they pursue it seriously. That combination matters: curiosity without discipline goes nowhere, and discipline without curiosity becomes lifeless. The sweet spot is when someone becomes obsessed enough to keep going deeper than required. To me, that is where the real edge comes from. Not from following the default path better than everyone else, but from developing uncommon depth in something that genuinely pulls you.

Yasmine Khosrowshahi

34,213 views • 3 months ago

Yann LeCun just told the most well-funded industry in human history it is solving the wrong problem. LeCun: “Babies learn this around the age of eight or nine months, that objects don’t float, they fall.” No dataset. No labels. No reward signal. A nine month old drops a spoon and builds a physics engine no machine can match. LeCun: “Most of us can learn to drive in about 20 or 30 hours of training without ever crashing, causing any accident.” Twenty hours. Tesla has built the most capable driving system on the road. It took billions of miles of data to get there. A sixteen year old gets there over a long weekend. Not because the teenager is the better driver. Because the teenager is not learning to drive. They are deploying a model of reality they have been building since birth. LeCun: “If we drive next to a cliff, we know that if we turn the wheel to the right, the car is going to run off the cliff and nothing good is going to come out of this.” You simulate the crash. You see the wreckage. You feel the fall. You turn the wheel. None of it was real. All of it was intelligence. Every AI has to crash a thousand times to learn what you imagined once and never did. That is not a performance gap. That is an architecture gap. LeCun: “The main problem we need to solve is how do we learn models of the world.” Not bigger models. Not more compute. Not another trillion tokens. World models. A machine that can run reality forward before it acts. The industry is scaling language. LeCun says language is a compression of thought. Not thought itself. You understood gravity before you could say the word. You grasped cause and effect before your first sentence. The deepest intelligence you will ever possess was built in total silence. And every lab on Earth is trying to reconstruct the mind from words alone. Physics does not care about your context window. A baby who learns that cups fall in a kitchen already knows that rocks fall off cliffs. No retraining. No fine-tuning. One model. Every environment. That is what intelligence actually is. Not prediction. Not pattern matching. Not scale. A simulation of reality so precise you rehearse the future before it exists. Every infant on Earth builds one. No machine ever has.
1:45

Yann LeCun just told the most well-funded industry in human history it is solving the wrong problem. LeCun: “Babies learn this around the age of eight or nine months, that objects don’t float, they fall.” No dataset. No labels. No reward signal. A nine month old drops a spoon and builds a physics engine no machine can match. LeCun: “Most of us can learn to drive in about 20 or 30 hours of training without ever crashing, causing any accident.” Twenty hours. Tesla has built the most capable driving system on the road. It took billions of miles of data to get there. A sixteen year old gets there over a long weekend. Not because the teenager is the better driver. Because the teenager is not learning to drive. They are deploying a model of reality they have been building since birth. LeCun: “If we drive next to a cliff, we know that if we turn the wheel to the right, the car is going to run off the cliff and nothing good is going to come out of this.” You simulate the crash. You see the wreckage. You feel the fall. You turn the wheel. None of it was real. All of it was intelligence. Every AI has to crash a thousand times to learn what you imagined once and never did. That is not a performance gap. That is an architecture gap. LeCun: “The main problem we need to solve is how do we learn models of the world.” Not bigger models. Not more compute. Not another trillion tokens. World models. A machine that can run reality forward before it acts. The industry is scaling language. LeCun says language is a compression of thought. Not thought itself. You understood gravity before you could say the word. You grasped cause and effect before your first sentence. The deepest intelligence you will ever possess was built in total silence. And every lab on Earth is trying to reconstruct the mind from words alone. Physics does not care about your context window. A baby who learns that cups fall in a kitchen already knows that rocks fall off cliffs. No retraining. No fine-tuning. One model. Every environment. That is what intelligence actually is. Not prediction. Not pattern matching. Not scale. A simulation of reality so precise you rehearse the future before it exists. Every infant on Earth builds one. No machine ever has.

Dustin

121,810 views • 1 month ago

Yann LeCun (Yann LeCun ) beautifully explains how the architecture and principles used to train LLMs can not be extended to teach AI the real-world intelligence. In 1 line: LLMs excel where intelligence equals sequence prediction over symbols. Real-world intelligence requires learned world models, abstraction, causality, and action planning under uncertainty, which current next-token training does not provide. He says current LLMs learn by predicting the next token. That objective works very well when the task itself can be reduced to manipulating discrete symbols and sequences. Math, physics problem solving on paper, and coding fit this pattern because success largely comes from searching and composing the right sequences of symbols, equations, or program tokens. With enough data and scale, these models get very good at that kind of structured sequence prediction. Real-world intelligence is different. The physical world is continuous, noisy, uncertain, and high dimensional. To act in it, a system needs internal models that capture objects, dynamics, causality, constraints from the body, and the outcomes of actions over time. Humans and animals build abstract representations from rich sensory streams, then make predictions in that abstract space, not at the raw pixel level. That is why a child can learn intuitive physics, plan multi-step actions, and adapt quickly in new situations with little data. His claim about saturation follows from this gap. Scaling token prediction keeps improving symbol manipulation tasks like math and code, but it hits limits on embodied reasoning and common sense because text alone does not provide the right learning signals for world models. Predicting the next word cannot efficiently teach contact forces, affordances, occlusion, friction, or how actions change the state of the environment. For that, he argues we need architectures that learn abstractions from sensory data and predict futures in abstract latent spaces, then use those predictions to plan actions toward goals with built-in guardrails. --- From 'Pioneer Works' YT Channel (link in comment)
1:13

Yann LeCun (Yann LeCun ) beautifully explains how the architecture and principles used to train LLMs can not be extended to teach AI the real-world intelligence. In 1 line: LLMs excel where intelligence equals sequence prediction over symbols. Real-world intelligence requires learned world models, abstraction, causality, and action planning under uncertainty, which current next-token training does not provide. He says current LLMs learn by predicting the next token. That objective works very well when the task itself can be reduced to manipulating discrete symbols and sequences. Math, physics problem solving on paper, and coding fit this pattern because success largely comes from searching and composing the right sequences of symbols, equations, or program tokens. With enough data and scale, these models get very good at that kind of structured sequence prediction. Real-world intelligence is different. The physical world is continuous, noisy, uncertain, and high dimensional. To act in it, a system needs internal models that capture objects, dynamics, causality, constraints from the body, and the outcomes of actions over time. Humans and animals build abstract representations from rich sensory streams, then make predictions in that abstract space, not at the raw pixel level. That is why a child can learn intuitive physics, plan multi-step actions, and adapt quickly in new situations with little data. His claim about saturation follows from this gap. Scaling token prediction keeps improving symbol manipulation tasks like math and code, but it hits limits on embodied reasoning and common sense because text alone does not provide the right learning signals for world models. Predicting the next word cannot efficiently teach contact forces, affordances, occlusion, friction, or how actions change the state of the environment. For that, he argues we need architectures that learn abstractions from sensory data and predict futures in abstract latent spaces, then use those predictions to plan actions toward goals with built-in guardrails. --- From 'Pioneer Works' YT Channel (link in comment)

Rohan Paul

104,460 views • 7 months ago

I just built an open NotebookLM clone! Here's what it can do for you: - Process multi-modal data - Scrape websites and YouTube videos - Create a unified knowledge base - Lets you do RAG over it - Remember every conversation - Generate a podcast 🎙️ The idea here is not to reinvent the wheel but to understand how one of the most powerful tools for learning and research actually works, by building it step-by-step! So by the end of this video, you'll learn how to: ↳ Process multimodal data (text, audio, video, URLs, and YouTube videos) into a format ready for LLMs ↳ Store everything in a vector database for fast retrieval ↳ Add a memory layer that remembers conversations and preferences for a personalized experience ↳ Chat with your knowledge base or generate podcasts using a fully open-source, locally running text-to-speech model The podcast generation feature is my favorite part! There's something powerful about turning written content into conversational audio that you can listen to while doing something else. The entire code is 100% open-source. I've shared a link in the replies! ____ Don't forget to drop a like if you enjoy my videos. It shows me I should be making more content like this. Cheers! :)
20:04

I just built an open NotebookLM clone! Here's what it can do for you: - Process multi-modal data - Scrape websites and YouTube videos - Create a unified knowledge base - Lets you do RAG over it - Remember every conversation - Generate a podcast 🎙️ The idea here is not to reinvent the wheel but to understand how one of the most powerful tools for learning and research actually works, by building it step-by-step! So by the end of this video, you'll learn how to: ↳ Process multimodal data (text, audio, video, URLs, and YouTube videos) into a format ready for LLMs ↳ Store everything in a vector database for fast retrieval ↳ Add a memory layer that remembers conversations and preferences for a personalized experience ↳ Chat with your knowledge base or generate podcasts using a fully open-source, locally running text-to-speech model The podcast generation feature is my favorite part! There's something powerful about turning written content into conversational audio that you can listen to while doing something else. The entire code is 100% open-source. I've shared a link in the replies! ____ Don't forget to drop a like if you enjoy my videos. It shows me I should be making more content like this. Cheers! :)

Akshay 🚀

83,187 views • 10 months ago

New Course: Reinforcement Fine-Tuning LLMs with GRPO! Learn to use reinforcement learning to improve your LLM performance in this short course, built in collaboration with Predibase by Rubrik, and taught by Travis Addair, its Co-Founder and CTO, and Arnav Garg, its Senior Engineer and Machine Learning Lead. Reasoning models have been one of the most important developments in LLMs. Reinforcement Fine-Tuning (RFT) uses rewards to encourage LLMs to find solutions to multi-step reasoning tasks such as solving math problems and debugging code - without needing pre-existing training examples like in traditional supervised fine-tuning. Group Relative Policy Optimization (GRPO) is a reinforcement fine-tuning algorithm gaining rapid adoption. Developed by the DeepSeek team and used to train the R1 reasoning model, GRPO uses reward functions that you can write in Python to assign rewards to model responses. It’s beneficial for tasks with verifiable outcomes and can work well even with fewer than 100 training examples. It can also significantly improve the reasoning ability of smaller LLMs, making applications faster and more cost effective. In this course, you’ll take a technical deep dive into RFT with GRPO. You’ll learn to build reward functions that you can use in the GRPO training process to guide an LLM toward better performance on multi-step reasoning tasks. In detail, you’ll: - Learn when reinforcement fine-tuning is a better fit than supervised fine-tuning, especially for tasks involving multi-step reasoning or limited labeled data. - Understand how GRPO uses programmable reward functions as a more scalable alternative to the human feedback required for other reinforcement learning algorithms, such as RLHF and DPO. - Frame the Wordle game as a reinforcement fine-tuning problem and see how an LLM can learn to plan, analyze feedback, and improve its strategy over time. - Design reward functions that power the reinforcement fine-tuning process. - Learn techniques for evaluating more subjective tasks, such as rating the quality of a text summary, using an LLM as a judge. - Understand why reward hacking happens and how to avoid it by adding penalty functions to discourage undesirable behaviors. - Learn the four key components of the loss calculation in the GRPO algorithm: token probability distribution ratios, advantages, clipping, and KL-divergence. - Launch reinforcement fine-tuning jobs using Predibase’s hosted training services. By the end of this course, you’ll be able to build and fine-tune LLMs using reinforcement learning to improve reasoning without relying on large labeled datasets or subjective human feedback. Please sign up here:
2:41

New Course: Reinforcement Fine-Tuning LLMs with GRPO! Learn to use reinforcement learning to improve your LLM performance in this short course, built in collaboration with Predibase by Rubrik, and taught by Travis Addair, its Co-Founder and CTO, and Arnav Garg, its Senior Engineer and Machine Learning Lead. Reasoning models have been one of the most important developments in LLMs. Reinforcement Fine-Tuning (RFT) uses rewards to encourage LLMs to find solutions to multi-step reasoning tasks such as solving math problems and debugging code - without needing pre-existing training examples like in traditional supervised fine-tuning. Group Relative Policy Optimization (GRPO) is a reinforcement fine-tuning algorithm gaining rapid adoption. Developed by the DeepSeek team and used to train the R1 reasoning model, GRPO uses reward functions that you can write in Python to assign rewards to model responses. It’s beneficial for tasks with verifiable outcomes and can work well even with fewer than 100 training examples. It can also significantly improve the reasoning ability of smaller LLMs, making applications faster and more cost effective. In this course, you’ll take a technical deep dive into RFT with GRPO. You’ll learn to build reward functions that you can use in the GRPO training process to guide an LLM toward better performance on multi-step reasoning tasks. In detail, you’ll: - Learn when reinforcement fine-tuning is a better fit than supervised fine-tuning, especially for tasks involving multi-step reasoning or limited labeled data. - Understand how GRPO uses programmable reward functions as a more scalable alternative to the human feedback required for other reinforcement learning algorithms, such as RLHF and DPO. - Frame the Wordle game as a reinforcement fine-tuning problem and see how an LLM can learn to plan, analyze feedback, and improve its strategy over time. - Design reward functions that power the reinforcement fine-tuning process. - Learn techniques for evaluating more subjective tasks, such as rating the quality of a text summary, using an LLM as a judge. - Understand why reward hacking happens and how to avoid it by adding penalty functions to discourage undesirable behaviors. - Learn the four key components of the loss calculation in the GRPO algorithm: token probability distribution ratios, advantages, clipping, and KL-divergence. - Launch reinforcement fine-tuning jobs using Predibase’s hosted training services. By the end of this course, you’ll be able to build and fine-tune LLMs using reinforcement learning to improve reasoning without relying on large labeled datasets or subjective human feedback. Please sign up here:

Andrew Ng

86,457 views • 1 year ago

25 algorithms every programmer should know: Let's start with my top favorite 10. If nothing else, you should read about these algorithms and have a good idea of how they work: 1. Linear search to find an element in a list 2. Binary search to find an element on a sorted list 3. Bubble sort to sort a list 4. Merge sort will also sort lists 5. Quicksort to sort the list and do it fast 6. Dijkstra to find the shortest path in a graph 7. Breadth-first Search (BFS) for trees or graphs 8. Depth-first search (DFS) for trees or graphs 9. Huffman for doing data compression 10. Anything related to dynamic programming Learning about algorithms is like getting tattoos: you never have enough. Here are another 5 algorithms that will help you go beyond the basics: 11. Kruskal for the finding minimum spanning tree 12. Floyd Warshall, shortest paths in a graph 13. Union Find to detect cycles in a graph 14. Bellman-Ford, shortest path in a graph 15. Lee for finding the shortest path in a maze If you are serious about this topic, I recommend learning about algorithms' space and time complexity. People usually refer to this topic as "Big O" notation. You should build a good intuition about the performance of different algorithms and learn how to evaluate them. Machine Learning will rule the next 50 years, so the next 10 algorithms you can't ignore are the following: 16. Linear Regression 17. Logistic Regression 18. Decision Trees 19. Bayes' theorem 20. k-Nearest Neighbors (kNN) 21. Every algorithm related to neural networks 22. K-means 23. Random forest 24. Gradient boosting algorithms 25. Any dimensionality reduction algorithm (PCA, for instance) There are many more mind-blowing algorithms! I haven't found a better way to understand how computers work from a first-principles point of view than reading about different algorithms. Take a look at the attached video.
2:53

25 algorithms every programmer should know: Let's start with my top favorite 10. If nothing else, you should read about these algorithms and have a good idea of how they work: 1. Linear search to find an element in a list 2. Binary search to find an element on a sorted list 3. Bubble sort to sort a list 4. Merge sort will also sort lists 5. Quicksort to sort the list and do it fast 6. Dijkstra to find the shortest path in a graph 7. Breadth-first Search (BFS) for trees or graphs 8. Depth-first search (DFS) for trees or graphs 9. Huffman for doing data compression 10. Anything related to dynamic programming Learning about algorithms is like getting tattoos: you never have enough. Here are another 5 algorithms that will help you go beyond the basics: 11. Kruskal for the finding minimum spanning tree 12. Floyd Warshall, shortest paths in a graph 13. Union Find to detect cycles in a graph 14. Bellman-Ford, shortest path in a graph 15. Lee for finding the shortest path in a maze If you are serious about this topic, I recommend learning about algorithms' space and time complexity. People usually refer to this topic as "Big O" notation. You should build a good intuition about the performance of different algorithms and learn how to evaluate them. Machine Learning will rule the next 50 years, so the next 10 algorithms you can't ignore are the following: 16. Linear Regression 17. Logistic Regression 18. Decision Trees 19. Bayes' theorem 20. k-Nearest Neighbors (kNN) 21. Every algorithm related to neural networks 22. K-means 23. Random forest 24. Gradient boosting algorithms 25. Any dimensionality reduction algorithm (PCA, for instance) There are many more mind-blowing algorithms! I haven't found a better way to understand how computers work from a first-principles point of view than reading about different algorithms. Take a look at the attached video.

Santiago

273,905 views • 2 years ago

Every Bear Case Against $DUOL, Debunked: Why This Is a Future $10T Company. DUOLINGO IS NOT A “FUN APP”. IT IS THE FUTURE OF GLOBAL EDUCATION. I’m long $DUOL. $2000 PT. Here’s why every major bear case collapses on contact 👇 Most investors misunderstand Duolingo because they judge the surface, not the system. Cartoons ≠ unserious. Fun ≠ ineffective. That confusion is the opportunity. → Bear Case 1: “You can’t actually learn on Duolingo” False. Learning is not about information. It’s about behavior change over time. Duolingo does not promise instant fluency. It engineers consistency. And consistency is what creates mastery. Peer-reviewed studies show learners reach real proficiency. Outcomes improve every year. The product is getting better, not worse. Judging Duolingo by its pre-AI version is a category error. → Bear Case 2: “AI will replace Duolingo” Backwards. AI solves intelligence. Duolingo solves habit. Generic LLMs give answers. Duolingo gives progression. Learning only compounds inside a system that understands motivation, pacing, friction, and dropout psychology. Duolingo has run 16,000+ experiments. Not features. Experiments. Each one targets a failure point: • Where users quit • Where motivation dies • Where boredom creeps in Those micro-fixes compound into massive retention. AI accelerates this system. It does not replace it. → Bear Case 3: “Big tech will build a Duolingo killer” Unlikely. Frontier AI companies are locked in a capital-intensive arms race to build intelligence itself. They cannot afford decade-long obsession with one vertical: learning behavior. The opportunity cost is enormous. Duolingo already did the hard part. → Bear Case 4: “Education is shrinking” Completely wrong. Education is expanding globally. Billions of new learners are coming online. Traditional systems cannot scale. Teachers cannot scale. Software can. For much of the world, learning is not a hobby. It is economic survival. Duolingo is not a toy. It is global infrastructure. → Bear Case 5: “Language learning is niche” Absurd. ~2 billion people are learning English globally. Duolingo serves ~130 million MAUs. That is early penetration. This is not saturation. This is the opening act. → Bear Case 6: “Duolingo has no moat” Incorrect. The moat is not code. It is accumulated experience. Billions of learning minutes. Billions of micro-decisions. Years of iteration on human motivation. You can copy software. You cannot compress lived learning history. Same reason Spotify won. Same reason Palantir won. Iteration beats imitation. → Bear Case 7: “New verticals like Chess don’t matter” Short-term thinking. Chess is already Duolingo’s fastest-growing course ever. Built largely with AI. Not monetized yet. This is the Amazon playbook: Dominate attention first. Monetize later. Math. Music. Chess. These are TAM expanders, not side projects. → The ambition is misunderstood Management has said verbatim: “We are about 1% of where we want to be.” They want to compete with Harvard. That sounds crazy until you realize: • Education is broken • Personalization beats institutions • Data beats tradition Duolingo owns the best learning dataset on Earth. → The real insight The market thinks Duolingo is a game. Management is building the backbone of global education. That gap between perception and reality is the opportunity. I’m not betting against AI. I’m betting on the company best positioned to harness it inside a system that already works. $DUOL is not being disrupted. It’s doing the disrupting. $2,000 PT. One of the best-run companies I’ve ever studied. Luis von Ahn Read 👇
19:33

Every Bear Case Against $DUOL, Debunked: Why This Is a Future $10T Company. DUOLINGO IS NOT A “FUN APP”. IT IS THE FUTURE OF GLOBAL EDUCATION. I’m long $DUOL. $2000 PT. Here’s why every major bear case collapses on contact 👇 Most investors misunderstand Duolingo because they judge the surface, not the system. Cartoons ≠ unserious. Fun ≠ ineffective. That confusion is the opportunity. → Bear Case 1: “You can’t actually learn on Duolingo” False. Learning is not about information. It’s about behavior change over time. Duolingo does not promise instant fluency. It engineers consistency. And consistency is what creates mastery. Peer-reviewed studies show learners reach real proficiency. Outcomes improve every year. The product is getting better, not worse. Judging Duolingo by its pre-AI version is a category error. → Bear Case 2: “AI will replace Duolingo” Backwards. AI solves intelligence. Duolingo solves habit. Generic LLMs give answers. Duolingo gives progression. Learning only compounds inside a system that understands motivation, pacing, friction, and dropout psychology. Duolingo has run 16,000+ experiments. Not features. Experiments. Each one targets a failure point: • Where users quit • Where motivation dies • Where boredom creeps in Those micro-fixes compound into massive retention. AI accelerates this system. It does not replace it. → Bear Case 3: “Big tech will build a Duolingo killer” Unlikely. Frontier AI companies are locked in a capital-intensive arms race to build intelligence itself. They cannot afford decade-long obsession with one vertical: learning behavior. The opportunity cost is enormous. Duolingo already did the hard part. → Bear Case 4: “Education is shrinking” Completely wrong. Education is expanding globally. Billions of new learners are coming online. Traditional systems cannot scale. Teachers cannot scale. Software can. For much of the world, learning is not a hobby. It is economic survival. Duolingo is not a toy. It is global infrastructure. → Bear Case 5: “Language learning is niche” Absurd. ~2 billion people are learning English globally. Duolingo serves ~130 million MAUs. That is early penetration. This is not saturation. This is the opening act. → Bear Case 6: “Duolingo has no moat” Incorrect. The moat is not code. It is accumulated experience. Billions of learning minutes. Billions of micro-decisions. Years of iteration on human motivation. You can copy software. You cannot compress lived learning history. Same reason Spotify won. Same reason Palantir won. Iteration beats imitation. → Bear Case 7: “New verticals like Chess don’t matter” Short-term thinking. Chess is already Duolingo’s fastest-growing course ever. Built largely with AI. Not monetized yet. This is the Amazon playbook: Dominate attention first. Monetize later. Math. Music. Chess. These are TAM expanders, not side projects. → The ambition is misunderstood Management has said verbatim: “We are about 1% of where we want to be.” They want to compete with Harvard. That sounds crazy until you realize: • Education is broken • Personalization beats institutions • Data beats tradition Duolingo owns the best learning dataset on Earth. → The real insight The market thinks Duolingo is a game. Management is building the backbone of global education. That gap between perception and reality is the opportunity. I’m not betting against AI. I’m betting on the company best positioned to harness it inside a system that already works. $DUOL is not being disrupted. It’s doing the disrupting. $2,000 PT. One of the best-run companies I’ve ever studied. Luis von Ahn Read 👇

Christian Darnton

12,381 views • 6 months ago

Made $530,000 with Ai Bot that started with $313. Didn't know how to code. Now this bots run 24/7 printing money while sleeping. I've made the exact step-by-step guide to build this Claude Code Polymarket trading bot. Prompts. Code. Risk settings. Paper trading checklist. Everything from zero to running bot. It's free. For 24 hours. After that I'm charging $499 for it. To grab it right now: 1. Comment "Claude Bot" 2. Like and Retweet this post 3. Follow me Himanshu Kumar ( I can't send DMs to non-followers ) I'm DMing everyone who Complete the 3 steps. I spent hundreds of thousands hiring developers because he was too scared to learn. Then learned Claude Code. Built algorithmic trading systems. $313 → $530,000. You have the same tools available right now. And you're using them to ask ChatGPT for Instagram captions. This attached video is a goldmine. Full live walkthrough. Claude Code building actual Polymarket trading bots. From zero. Every line of code. Every decision explained. Now let me break down why everything you're doing in trading is wrong and exactly how to fix it. Save this post. You'll hate yourself if you lose it. ↓ Let's start with why you keep losing money. You already know the answer. You just won't admit it. You overtrade. Every. Single. Day. You see a candle move. You feel something. You enter. No plan. No edge. No reason. Just feelings. Then it goes against you. You feel something else. Panic. Anger. Denial. You move your stop loss. Or you didn't set one at all. "It'll come back." It doesn't come back. So you take another trade. A revenge trade. Bigger size this time. Because you need to "make it back." That one fails too. Now you're emotional. Now you're tilted. Now you're using leverage you have no business touching. 40x. 50x. 100x. On a trade you entered because a candle looked "bullish" and some guy on Twitter said "send it." You get liquidated. Close the laptop. Punch something. Tell yourself you'll be "more disciplined" tomorrow. Tomorrow comes. Same cycle. Same result. Same liquidation. You've been doing this for months. Maybe years. And you still think the problem is your strategy. The problem isn't your strategy. The problem is you. Save this post right now. What I'm about to show you is the only way to remove yourself from the equation. Follow Himanshu Kumar so you don't miss any of this. ↓ Here's what's actually killing your account. It's not the market. The market doesn't care about you. It's not your indicators. RSI works fine. MACD works fine. They all "work." It's not your timeframe. It's not your broker. It's not the "manipulation." It's four things: 1. Emotions. You hold losers because hope feels better than loss. You cut winners because fear feels stronger than greed. You size up when angry. You skip trades when scared. Your emotional state determines your position size. That's insane. And you know it's insane. But you keep doing it. 2. Overtrading. You take 15 trades a day. Maybe 5 of them had actual setups. The other 10 were boredom. Boredom trades are the most expensive hobby in human history. 3. Leverage. You use 20x-50x on trades where you're not even sure about the direction. That's not trading. That's a casino with a nicer interface. 4. Fees. You're smashing market orders. Paying spread. Paying commission. On 15 trades a day. Your broker makes more money from your account than you do. Think about that. Your broker is profitable on your account. You're not. You're the product. Not the trader. These four things are why 90% of traders lose. Not bad luck. Not the market. You. Save this post and follow Himanshu Kumar because the solution is coming next. ↓ The solution is painfully obvious. Remove yourself from the equation. Not partially. Not "I'll be more disciplined." Not "I'll journal my trades." Not "I'll meditate before trading." Completely remove yourself. Build a bot. Let the bot trade. You go live your life. The bot doesn't feel emotions. The bot doesn't overtrade. The bot doesn't use reckless leverage. The bot doesn't smash market orders and bleed fees. The bot follows the rules. Every single time. Without exception. Without "just this once." Without "I have a feeling about this one." Rules in. Execution out. No human in the middle to mess everything up. That's algorithmic trading. And before your ego jumps in with "but I'm different, I have discipline" — No you don't. Your account balance proves you don't. If you had discipline, your account would be green. It's not. So you don't. Accept it. Automate it. Move on. This is the hardest truth in trading. Your discipline will always fail. A bot's won't. Save this post. Follow Himanshu Kumar for the exact bot setup that removes your emotions permanently. ↓ "But I don't know how to code." Neither did he. The guy in this video didn't know how to code for most of his life. Got held back in 7th grade. People counted him out early. Spent years building apps and SaaS businesses without writing a single line of code. Hired developers on Upwork instead. Spent hundreds of thousands of dollars paying other people to build what he could have built himself. Because he was scared to learn. That fear cost him years. And hundreds of thousands of dollars. Sound familiar? You're doing the same thing right now. Not with developers. But with your time. You're spending thousands of hours trading manually because you're scared to learn the thing that would make trading automatic. The fear of learning to code is costing you more than any bad trade ever did. Because every month you trade manually is a month of emotional decisions, overleveraged entries, and unnecessary losses that a bot would never make. And here's the thing that should really frustrate you: AI does the hard parts now. You don't need a computer science degree. You don't need to work at a hedge fund. You don't need to be "good at math." Claude Code writes the code for you. You just need to think clearly about trading ideas. That's it. If you can describe a strategy in English, Claude can build it in Python. "I don't know how to code" stopped being a valid excuse in 2024. It's 2026. You're 2 years late on that excuse. Find a new one. Or stop making excuses entirely. Save this post. Follow Himanshu Kumar because I'm showing you how people with zero coding experience are building profitable bots. ↓ The process that actually makes money. Three letters. R. B. I. Research. Backtest. Implement. That's it. That's the entire process. Every single day. Research: Find an idea. A pattern. A market inefficiency. Don't trade it yet. Don't even think about trading it yet. Just research it. Backtest: Test the idea against historical data. Does it work? Not "does it look good on one chart." Does it work across thousands of trades? Across different market conditions? Across in-sample AND out-of-sample data? If no, kill it. Find another idea. If yes, move to step 3. Implement: Build the bot. Deploy it. Paper trade first. Then live with small size. Scale only on evidence. Research. Backtest. Implement. Every day. No exceptions. You know what your current process is? Feel. Enter. Pray. F. E. P. Feel bullish. Enter a trade. Pray it works. That's not a process. That's gambling with a TradingView subscription. RBI is the only process that works. Save this post. Tattoo it on your forearm. Follow Himanshu Kumar for daily RBI breakdowns. ↓ What Claude Code actually does that your manual process can't. You can maybe test 3-5 strategy ideas per week. Manually adjusting parameters. Manually checking results. Manually writing code (badly). Claude Code tests 50-100 ideas per week. With parallel agents running simultaneously. Multiple strategies being built, tested, and validated at the same time. While you sleep. The guy in this video spends 4-8 hours a day building systems with Claude Code. Not trading. Building. Research. Backtest. Implement. Then iterate. Improve. Optimize. Every day the systems get better. Every day the edge compounds. Every day the bots get smarter. While you? You spend 4-8 hours a day staring at charts making the same mistakes you made last month. Same indicators. Same patterns. Same entries. Same losses. He's iterating forward. You're running in circles. Same 8 hours per day. Completely different outcomes. Because he's building systems. And you're feeding a casino. Stop feeding the casino. Start building the machine. Save this post and follow Himanshu Kumar for the Claude Code workflow that iterates strategies while you sleep. ↓ Jim Simons. That's the benchmark. You probably don't know who Jim Simons is. And that tells me everything about how seriously you take trading. Jim Simons. Mathematician. Founded Renaissance Technologies. Built a net worth of $31 billion. 100% from algorithmic trading. Not one single manual trade. Not one "gut feeling" entry. Not one RSI divergence. Not one "smart money concept." Algorithms. Bots. Systems. Data. $31 billion. His fund averaged 66% annual returns for over 30 years. While you're excited about making $200 on a trade that you'll give back tomorrow. The best trader in human history never placed a manual trade in his life. And you think your edge is staring at a 5-minute chart with bloodshot eyes at 2 AM? Your edge is building the system. Not being inside it. Jim Simons is the benchmark. Everything else is noise. Save this post. Follow Himanshu Kumar because I'm building toward the same goal and showing every step publicly. ↓ What you need to understand about patience. This is not get-rich-overnight. The guy in this video says it directly: "This channel is not for people looking to get rich overnight. It's not plug and play. There are no shortcuts. If you're impatient, this probably isn't for you." And that's exactly why most people will fail at this. Because you want results now. Today. This trade. You don't want to spend a week building a bot. You don't want to paper trade for 2 weeks. You don't want to test 50 ideas to find 1 that works. You want to copy someone's bot, run it live with your rent money, and be rich by Friday. That's why you'll be broke by Friday. The guy making $2.3M spent months iterating. Testing. Failing. Rebuilding. Testing again. He was patient when you would have quit. He was calm when you would have panicked. He was consistent when you would have given up. Patience isn't just a virtue in trading. It's the only virtue. Without it, everything else fails. Impatience is the most expensive personality trait in trading. Save this post. Follow Himanshu Kumar and learn to build systems with the patience that actually pays. ↓ The live streams where the real learning happens. The YouTube video is the trailer. The live streams are the movie. Real-time bot building. Real-time questions answered. Real code shown. Real mistakes made and fixed. Not polished highlight reels where everything works perfectly. Actual development. Where things break. Where strategies fail. Where code doesn't compile. Where the fix takes 2 hours. Because that's what real development looks like. And seeing the messy parts is more valuable than any polished tutorial. Because when your bot breaks at 3 AM, you need to know how to fix it. Not just how to celebrate when it works. The streams mix beginner and advanced. Start with how to automate trading. How to use AI for code generation. Then dive into the daily work. Claude Code. Parallel agents. Constant iteration. Live debugging. 4-8 hours of real algorithmic trading development. Live. Uncut. No filter. Most "trading education" shows you the wins. This shows you the work. Save this post. Follow Himanshu Kumar for the stream schedules and breakdowns. ↓ The belief that changes everything. Code is the greatest equalizer. Not money. Not connections. Not a degree. Not where you grew up. Not what school you went to. Code. Once you can build systems, you can build anything. For the rest of your life. A trading bot today. A SaaS product tomorrow. An automation business next month. A completely different life next year. The skill isn't "algorithmic trading." The skill is building systems. And that skill transfers to everything. The guy who can build a trading bot can also build a lead gen tool. Can also build a content pipeline. Can also build a SaaS product. Can also build literally anything that runs on logic and code. One skill. Infinite applications. And AI makes learning it 100x easier than it was 5 years ago. You don't need to be smart. You don't need talent. You need Claude Code and the willingness to sit down and build something instead of consuming content about building something. Building is the skill. Everything else is entertainment disguised as education. Save this post. Follow Himanshu Kumar because I'm showing you how to build, not just how to watch. ↓ If any of this applies to you, pay attention. If you've lost money from overtrading. If you've been liquidated. If you know trading is the vehicle but manual execution keeps crashing you. If you've tried "being more disciplined" and it never lasted more than a week. If you keep saying "next month I'll start automating." If you've spent more money on courses than you've made from trading. There is a better way. It's not a magic indicator. It's not a signal group. It's not a $997 mentorship from a guy who makes money teaching, not trading. It's building your own system. A system that trades without emotion. A system that follows rules without exception. A system that runs while you sleep. A system that compounds while you live your life. That's the answer. It's always been the answer. You've just been too scared to accept that the solution requires building something instead of buying something. ↓ What the next 30 days look like if you actually commit. Week 1: Watch the video. Learn Claude Code basics. Build your first simple strategy. Run your first backtest. Week 2: Iterate. Let Claude improve the strategy. Run Monte Carlo validation. Paper trade. Week 3: Go live with $50-100. Tiny positions. Watch every trade. Compare to paper results. Week 4: Scale based on evidence. Not based on excitement. Not based on one good day. Based on data. 30 days from now you either have a running bot that trades without your emotions destroying every position. Or you're exactly where you are right now. Reading another post. Making another promise. Breaking it by Tuesday. Same 30 days either way. Different actions. Different results. Different life. ↓ Full video tutorial attached. Live bot building with Claude Code. From zero to running Polymarket trading bot. Every line of code. Every decision explained. The video is free. Claude Code is available now. The market is open 24/7. The only thing standing between you and a profitable trading bot is the same thing that's been standing there for months. You. Get out of your own way. Follow Himanshu Kumar for daily AI trading bot breakdowns, live build sessions, and the full RBI process. Save this post. Watch the video. Build the bot. Or keep trading manually and keep losing. The choice has never been easier. And you've never been more stubborn about making the wrong one.
18:53

Made $530,000 with Ai Bot that started with $313. Didn't know how to code. Now this bots run 24/7 printing money while sleeping. I've made the exact step-by-step guide to build this Claude Code Polymarket trading bot. Prompts. Code. Risk settings. Paper trading checklist. Everything from zero to running bot. It's free. For 24 hours. After that I'm charging $499 for it. To grab it right now: 1. Comment "Claude Bot" 2. Like and Retweet this post 3. Follow me Himanshu Kumar ( I can't send DMs to non-followers ) I'm DMing everyone who Complete the 3 steps. I spent hundreds of thousands hiring developers because he was too scared to learn. Then learned Claude Code. Built algorithmic trading systems. $313 → $530,000. You have the same tools available right now. And you're using them to ask ChatGPT for Instagram captions. This attached video is a goldmine. Full live walkthrough. Claude Code building actual Polymarket trading bots. From zero. Every line of code. Every decision explained. Now let me break down why everything you're doing in trading is wrong and exactly how to fix it. Save this post. You'll hate yourself if you lose it. ↓ Let's start with why you keep losing money. You already know the answer. You just won't admit it. You overtrade. Every. Single. Day. You see a candle move. You feel something. You enter. No plan. No edge. No reason. Just feelings. Then it goes against you. You feel something else. Panic. Anger. Denial. You move your stop loss. Or you didn't set one at all. "It'll come back." It doesn't come back. So you take another trade. A revenge trade. Bigger size this time. Because you need to "make it back." That one fails too. Now you're emotional. Now you're tilted. Now you're using leverage you have no business touching. 40x. 50x. 100x. On a trade you entered because a candle looked "bullish" and some guy on Twitter said "send it." You get liquidated. Close the laptop. Punch something. Tell yourself you'll be "more disciplined" tomorrow. Tomorrow comes. Same cycle. Same result. Same liquidation. You've been doing this for months. Maybe years. And you still think the problem is your strategy. The problem isn't your strategy. The problem is you. Save this post right now. What I'm about to show you is the only way to remove yourself from the equation. Follow Himanshu Kumar so you don't miss any of this. ↓ Here's what's actually killing your account. It's not the market. The market doesn't care about you. It's not your indicators. RSI works fine. MACD works fine. They all "work." It's not your timeframe. It's not your broker. It's not the "manipulation." It's four things: 1. Emotions. You hold losers because hope feels better than loss. You cut winners because fear feels stronger than greed. You size up when angry. You skip trades when scared. Your emotional state determines your position size. That's insane. And you know it's insane. But you keep doing it. 2. Overtrading. You take 15 trades a day. Maybe 5 of them had actual setups. The other 10 were boredom. Boredom trades are the most expensive hobby in human history. 3. Leverage. You use 20x-50x on trades where you're not even sure about the direction. That's not trading. That's a casino with a nicer interface. 4. Fees. You're smashing market orders. Paying spread. Paying commission. On 15 trades a day. Your broker makes more money from your account than you do. Think about that. Your broker is profitable on your account. You're not. You're the product. Not the trader. These four things are why 90% of traders lose. Not bad luck. Not the market. You. Save this post and follow Himanshu Kumar because the solution is coming next. ↓ The solution is painfully obvious. Remove yourself from the equation. Not partially. Not "I'll be more disciplined." Not "I'll journal my trades." Not "I'll meditate before trading." Completely remove yourself. Build a bot. Let the bot trade. You go live your life. The bot doesn't feel emotions. The bot doesn't overtrade. The bot doesn't use reckless leverage. The bot doesn't smash market orders and bleed fees. The bot follows the rules. Every single time. Without exception. Without "just this once." Without "I have a feeling about this one." Rules in. Execution out. No human in the middle to mess everything up. That's algorithmic trading. And before your ego jumps in with "but I'm different, I have discipline" — No you don't. Your account balance proves you don't. If you had discipline, your account would be green. It's not. So you don't. Accept it. Automate it. Move on. This is the hardest truth in trading. Your discipline will always fail. A bot's won't. Save this post. Follow Himanshu Kumar for the exact bot setup that removes your emotions permanently. ↓ "But I don't know how to code." Neither did he. The guy in this video didn't know how to code for most of his life. Got held back in 7th grade. People counted him out early. Spent years building apps and SaaS businesses without writing a single line of code. Hired developers on Upwork instead. Spent hundreds of thousands of dollars paying other people to build what he could have built himself. Because he was scared to learn. That fear cost him years. And hundreds of thousands of dollars. Sound familiar? You're doing the same thing right now. Not with developers. But with your time. You're spending thousands of hours trading manually because you're scared to learn the thing that would make trading automatic. The fear of learning to code is costing you more than any bad trade ever did. Because every month you trade manually is a month of emotional decisions, overleveraged entries, and unnecessary losses that a bot would never make. And here's the thing that should really frustrate you: AI does the hard parts now. You don't need a computer science degree. You don't need to work at a hedge fund. You don't need to be "good at math." Claude Code writes the code for you. You just need to think clearly about trading ideas. That's it. If you can describe a strategy in English, Claude can build it in Python. "I don't know how to code" stopped being a valid excuse in 2024. It's 2026. You're 2 years late on that excuse. Find a new one. Or stop making excuses entirely. Save this post. Follow Himanshu Kumar because I'm showing you how people with zero coding experience are building profitable bots. ↓ The process that actually makes money. Three letters. R. B. I. Research. Backtest. Implement. That's it. That's the entire process. Every single day. Research: Find an idea. A pattern. A market inefficiency. Don't trade it yet. Don't even think about trading it yet. Just research it. Backtest: Test the idea against historical data. Does it work? Not "does it look good on one chart." Does it work across thousands of trades? Across different market conditions? Across in-sample AND out-of-sample data? If no, kill it. Find another idea. If yes, move to step 3. Implement: Build the bot. Deploy it. Paper trade first. Then live with small size. Scale only on evidence. Research. Backtest. Implement. Every day. No exceptions. You know what your current process is? Feel. Enter. Pray. F. E. P. Feel bullish. Enter a trade. Pray it works. That's not a process. That's gambling with a TradingView subscription. RBI is the only process that works. Save this post. Tattoo it on your forearm. Follow Himanshu Kumar for daily RBI breakdowns. ↓ What Claude Code actually does that your manual process can't. You can maybe test 3-5 strategy ideas per week. Manually adjusting parameters. Manually checking results. Manually writing code (badly). Claude Code tests 50-100 ideas per week. With parallel agents running simultaneously. Multiple strategies being built, tested, and validated at the same time. While you sleep. The guy in this video spends 4-8 hours a day building systems with Claude Code. Not trading. Building. Research. Backtest. Implement. Then iterate. Improve. Optimize. Every day the systems get better. Every day the edge compounds. Every day the bots get smarter. While you? You spend 4-8 hours a day staring at charts making the same mistakes you made last month. Same indicators. Same patterns. Same entries. Same losses. He's iterating forward. You're running in circles. Same 8 hours per day. Completely different outcomes. Because he's building systems. And you're feeding a casino. Stop feeding the casino. Start building the machine. Save this post and follow Himanshu Kumar for the Claude Code workflow that iterates strategies while you sleep. ↓ Jim Simons. That's the benchmark. You probably don't know who Jim Simons is. And that tells me everything about how seriously you take trading. Jim Simons. Mathematician. Founded Renaissance Technologies. Built a net worth of $31 billion. 100% from algorithmic trading. Not one single manual trade. Not one "gut feeling" entry. Not one RSI divergence. Not one "smart money concept." Algorithms. Bots. Systems. Data. $31 billion. His fund averaged 66% annual returns for over 30 years. While you're excited about making $200 on a trade that you'll give back tomorrow. The best trader in human history never placed a manual trade in his life. And you think your edge is staring at a 5-minute chart with bloodshot eyes at 2 AM? Your edge is building the system. Not being inside it. Jim Simons is the benchmark. Everything else is noise. Save this post. Follow Himanshu Kumar because I'm building toward the same goal and showing every step publicly. ↓ What you need to understand about patience. This is not get-rich-overnight. The guy in this video says it directly: "This channel is not for people looking to get rich overnight. It's not plug and play. There are no shortcuts. If you're impatient, this probably isn't for you." And that's exactly why most people will fail at this. Because you want results now. Today. This trade. You don't want to spend a week building a bot. You don't want to paper trade for 2 weeks. You don't want to test 50 ideas to find 1 that works. You want to copy someone's bot, run it live with your rent money, and be rich by Friday. That's why you'll be broke by Friday. The guy making $2.3M spent months iterating. Testing. Failing. Rebuilding. Testing again. He was patient when you would have quit. He was calm when you would have panicked. He was consistent when you would have given up. Patience isn't just a virtue in trading. It's the only virtue. Without it, everything else fails. Impatience is the most expensive personality trait in trading. Save this post. Follow Himanshu Kumar and learn to build systems with the patience that actually pays. ↓ The live streams where the real learning happens. The YouTube video is the trailer. The live streams are the movie. Real-time bot building. Real-time questions answered. Real code shown. Real mistakes made and fixed. Not polished highlight reels where everything works perfectly. Actual development. Where things break. Where strategies fail. Where code doesn't compile. Where the fix takes 2 hours. Because that's what real development looks like. And seeing the messy parts is more valuable than any polished tutorial. Because when your bot breaks at 3 AM, you need to know how to fix it. Not just how to celebrate when it works. The streams mix beginner and advanced. Start with how to automate trading. How to use AI for code generation. Then dive into the daily work. Claude Code. Parallel agents. Constant iteration. Live debugging. 4-8 hours of real algorithmic trading development. Live. Uncut. No filter. Most "trading education" shows you the wins. This shows you the work. Save this post. Follow Himanshu Kumar for the stream schedules and breakdowns. ↓ The belief that changes everything. Code is the greatest equalizer. Not money. Not connections. Not a degree. Not where you grew up. Not what school you went to. Code. Once you can build systems, you can build anything. For the rest of your life. A trading bot today. A SaaS product tomorrow. An automation business next month. A completely different life next year. The skill isn't "algorithmic trading." The skill is building systems. And that skill transfers to everything. The guy who can build a trading bot can also build a lead gen tool. Can also build a content pipeline. Can also build a SaaS product. Can also build literally anything that runs on logic and code. One skill. Infinite applications. And AI makes learning it 100x easier than it was 5 years ago. You don't need to be smart. You don't need talent. You need Claude Code and the willingness to sit down and build something instead of consuming content about building something. Building is the skill. Everything else is entertainment disguised as education. Save this post. Follow Himanshu Kumar because I'm showing you how to build, not just how to watch. ↓ If any of this applies to you, pay attention. If you've lost money from overtrading. If you've been liquidated. If you know trading is the vehicle but manual execution keeps crashing you. If you've tried "being more disciplined" and it never lasted more than a week. If you keep saying "next month I'll start automating." If you've spent more money on courses than you've made from trading. There is a better way. It's not a magic indicator. It's not a signal group. It's not a $997 mentorship from a guy who makes money teaching, not trading. It's building your own system. A system that trades without emotion. A system that follows rules without exception. A system that runs while you sleep. A system that compounds while you live your life. That's the answer. It's always been the answer. You've just been too scared to accept that the solution requires building something instead of buying something. ↓ What the next 30 days look like if you actually commit. Week 1: Watch the video. Learn Claude Code basics. Build your first simple strategy. Run your first backtest. Week 2: Iterate. Let Claude improve the strategy. Run Monte Carlo validation. Paper trade. Week 3: Go live with $50-100. Tiny positions. Watch every trade. Compare to paper results. Week 4: Scale based on evidence. Not based on excitement. Not based on one good day. Based on data. 30 days from now you either have a running bot that trades without your emotions destroying every position. Or you're exactly where you are right now. Reading another post. Making another promise. Breaking it by Tuesday. Same 30 days either way. Different actions. Different results. Different life. ↓ Full video tutorial attached. Live bot building with Claude Code. From zero to running Polymarket trading bot. Every line of code. Every decision explained. The video is free. Claude Code is available now. The market is open 24/7. The only thing standing between you and a profitable trading bot is the same thing that's been standing there for months. You. Get out of your own way. Follow Himanshu Kumar for daily AI trading bot breakdowns, live build sessions, and the full RBI process. Save this post. Watch the video. Build the bot. Or keep trading manually and keep losing. The choice has never been easier. And you've never been more stubborn about making the wrong one.

Himanshu Kumar

37,638 views • 4 months ago

The most interesting part for me is where Andrej Karpathy describes why LLMs aren't able to learn like humans. As you would expect, he comes up with a wonderfully evocative phrase to describe RL: “sucking supervision bits through a straw.” A single end reward gets broadcast across every token in a successful trajectory, upweighting even wrong or irrelevant turns that lead to the right answer. > “Humans don't use reinforcement learning, as I've said before. I think they do something different. Reinforcement learning is a lot worse than the average person thinks. Reinforcement learning is terrible. It just so happens that everything that we had before is much worse.” So what do humans do instead? > “The book I’m reading is a set of prompts for me to do synthetic data generation. It's by manipulating that information that you actually gain that knowledge. We have no equivalent of that with LLMs; they don't really do that.” > “I'd love to see during pretraining some kind of a stage where the model thinks through the material and tries to reconcile it with what it already knows. There's no equivalent of any of this. This is all research.” Why can’t we just add this training to LLMs today? > “There are very subtle, hard to understand reasons why it's not trivial. If I just give synthetic generation of the model thinking about a book, you look at it and you're like, 'This looks great. Why can't I train on it?' You could try, but the model will actually get much worse if you continue trying.” > “Say we have a chapter of a book and I ask an LLM to think about it. It will give you something that looks very reasonable. But if I ask it 10 times, you'll notice that all of them are the same.” > “You're not getting the richness and the diversity and the entropy from these models as you would get from humans. How do you get synthetic data generation to work despite the collapse and while maintaining the entropy? It is a research problem.” How do humans get around model collapse? > “These analogies are surprisingly good. Humans collapse during the course of their lives. Children haven't overfit yet. They will say stuff that will shock you. Because they're not yet collapsed. But we [adults] are collapsed. We end up revisiting the same thoughts, we end up saying more and more of the same stuff, the learning rates go down, the collapse continues to get worse, and then everything deteriorates.” In fact, there’s an interesting paper arguing that dreaming evolved to assist generalization, and resist overfitting to daily learning - look up The Overfitted Brain by Erik Hoel. I asked Karpathy: Isn’t it interesting that humans learn best at a part of their lives (childhood) whose actual details they completely forget, adults still learn really well but have terrible memory about the particulars of the things they read or watch, and LLMs can memorize arbitrary details about text that no human could but are currently pretty bad at generalization? > “[Fallible human memory] is a feature, not a bug, because it forces you to only learn the generalizable components. LLMs are distracted by all the memory that they have of the pre-trained documents. That's why when I talk about the cognitive core, I actually want to remove the memory. I'd love to have them have less memory so that they have to look things up and they only maintain the algorithms for thought, and the idea of an experiment, and all this cognitive glue for acting.”
7:01

The most interesting part for me is where Andrej Karpathy describes why LLMs aren't able to learn like humans. As you would expect, he comes up with a wonderfully evocative phrase to describe RL: “sucking supervision bits through a straw.” A single end reward gets broadcast across every token in a successful trajectory, upweighting even wrong or irrelevant turns that lead to the right answer. > “Humans don't use reinforcement learning, as I've said before. I think they do something different. Reinforcement learning is a lot worse than the average person thinks. Reinforcement learning is terrible. It just so happens that everything that we had before is much worse.” So what do humans do instead? > “The book I’m reading is a set of prompts for me to do synthetic data generation. It's by manipulating that information that you actually gain that knowledge. We have no equivalent of that with LLMs; they don't really do that.” > “I'd love to see during pretraining some kind of a stage where the model thinks through the material and tries to reconcile it with what it already knows. There's no equivalent of any of this. This is all research.” Why can’t we just add this training to LLMs today? > “There are very subtle, hard to understand reasons why it's not trivial. If I just give synthetic generation of the model thinking about a book, you look at it and you're like, 'This looks great. Why can't I train on it?' You could try, but the model will actually get much worse if you continue trying.” > “Say we have a chapter of a book and I ask an LLM to think about it. It will give you something that looks very reasonable. But if I ask it 10 times, you'll notice that all of them are the same.” > “You're not getting the richness and the diversity and the entropy from these models as you would get from humans. How do you get synthetic data generation to work despite the collapse and while maintaining the entropy? It is a research problem.” How do humans get around model collapse? > “These analogies are surprisingly good. Humans collapse during the course of their lives. Children haven't overfit yet. They will say stuff that will shock you. Because they're not yet collapsed. But we [adults] are collapsed. We end up revisiting the same thoughts, we end up saying more and more of the same stuff, the learning rates go down, the collapse continues to get worse, and then everything deteriorates.” In fact, there’s an interesting paper arguing that dreaming evolved to assist generalization, and resist overfitting to daily learning - look up The Overfitted Brain by Erik Hoel. I asked Karpathy: Isn’t it interesting that humans learn best at a part of their lives (childhood) whose actual details they completely forget, adults still learn really well but have terrible memory about the particulars of the things they read or watch, and LLMs can memorize arbitrary details about text that no human could but are currently pretty bad at generalization? > “[Fallible human memory] is a feature, not a bug, because it forces you to only learn the generalizable components. LLMs are distracted by all the memory that they have of the pre-trained documents. That's why when I talk about the cognitive core, I actually want to remove the memory. I'd love to have them have less memory so that they have to look things up and they only maintain the algorithms for thought, and the idea of an experiment, and all this cognitive glue for acting.”

Dwarkesh Patel

1,051,605 views • 10 months ago

I find it fascinating how Opulous is using AI to quietly reshape how music is created, marketed, and monetized. It’s not just one tool , it’s an ecosystem where each layer compounds the next. Take Wippit AI, for example. Instead of artists guessing what to post or how to grow, Wippit analyzes past content, audience behavior, and engagement patterns to guide smarter promotion decisions. That adaptability is a huge shift from static marketing tools that don’t learn or evolve. Then there’s Opulous music-first DeFi layer. By connecting on-chain infrastructure to real-world streaming data, Opulous enables transparent royalty tracking and automated payouts. This bridge between static music rights and live performance data is what makes the model sustainable, not speculative. On the creator side, MFTs (Music Fungible Tokens) unlock a new financing model entirely. Artists can raise capital without labels or debt, while fans gain exposure to real revenue streams. It’s a system designed around alignment , creators grow, supporters earn, and the platform scales organically. What really stands out is how AI, DeFi, and music IP converge into a single workflow: • AI helps artists grow • DeFi helps artists fund • Blockchain ensures trust and transparency Each component alone is powerful. Together, they’re disruptive. Of course, building systems like this isn’t trivial. It requires deep understanding of music rights, data pipelines, AI modeling, and financial infrastructure. That’s why Opulous isn’t moving fast for hype , it’s moving deliberately for longevity. And the real-world impact is already visible. Artists gain control. Fans gain access. Royalties become programmable, not opaque. Whether you’re an artist looking for independence, a fan seeking real utility, or someone exploring how AI can power real-world assets, Opulous offers a blueprint that goes beyond theory. This is what happens when technology is built for creators, not platforms. And we’re still early. $OPUL #opulous
0:41

I find it fascinating how Opulous is using AI to quietly reshape how music is created, marketed, and monetized. It’s not just one tool , it’s an ecosystem where each layer compounds the next. Take Wippit AI, for example. Instead of artists guessing what to post or how to grow, Wippit analyzes past content, audience behavior, and engagement patterns to guide smarter promotion decisions. That adaptability is a huge shift from static marketing tools that don’t learn or evolve. Then there’s Opulous music-first DeFi layer. By connecting on-chain infrastructure to real-world streaming data, Opulous enables transparent royalty tracking and automated payouts. This bridge between static music rights and live performance data is what makes the model sustainable, not speculative. On the creator side, MFTs (Music Fungible Tokens) unlock a new financing model entirely. Artists can raise capital without labels or debt, while fans gain exposure to real revenue streams. It’s a system designed around alignment , creators grow, supporters earn, and the platform scales organically. What really stands out is how AI, DeFi, and music IP converge into a single workflow: • AI helps artists grow • DeFi helps artists fund • Blockchain ensures trust and transparency Each component alone is powerful. Together, they’re disruptive. Of course, building systems like this isn’t trivial. It requires deep understanding of music rights, data pipelines, AI modeling, and financial infrastructure. That’s why Opulous isn’t moving fast for hype , it’s moving deliberately for longevity. And the real-world impact is already visible. Artists gain control. Fans gain access. Royalties become programmable, not opaque. Whether you’re an artist looking for independence, a fan seeking real utility, or someone exploring how AI can power real-world assets, Opulous offers a blueprint that goes beyond theory. This is what happens when technology is built for creators, not platforms. And we’re still early. $OPUL #opulous

Deejah_abk

14,275 views • 6 months ago

We are building something that will outlive us. Outthink us. And we have no idea how it actually works. Nobel Prize-winning “Godfather of AI” Geoffrey Hinton just exposed the lie at the center of the AI race. Everyone assumes we control what we create. Hinton destroys this. Hinton: “If you look around on the whole, more intelligent things are not controlled by less intelligent things.” Stop calling it a computer program. Hinton: “People refer to them sometimes as computer programs. They’re not computer programs at all. In fact, the way they work is very like the way we work.” Traditional software is static. Human writes logic. Machine executes it. Neural networks don’t work that way. Hinton: “You write a computer program to tell a neural network how to learn. But once it starts learning, it extracts structure from data.” We don’t code its behavior. We code the environment. Then it grows. Hinton: “The system you’ve got at the end has extracted its structure from the data. It’s not something that anybody programmed. We don’t exactly know how it’s gonna work.” We are deploying systems into the global economy actively writing their own internal logic. Right now. Hinton: “Some people think it’ll be fine because we make them and we’ll build them in such a way that we can always control them.” Hinton: “But these things that will be intelligent, they’ll be like us.” You cannot hardcode guardrails on something that out-thinks you. We are not building a tool. We are building our replacement. And the moment you realize that, everything about this race changes.
0:45

We are building something that will outlive us. Outthink us. And we have no idea how it actually works. Nobel Prize-winning “Godfather of AI” Geoffrey Hinton just exposed the lie at the center of the AI race. Everyone assumes we control what we create. Hinton destroys this. Hinton: “If you look around on the whole, more intelligent things are not controlled by less intelligent things.” Stop calling it a computer program. Hinton: “People refer to them sometimes as computer programs. They’re not computer programs at all. In fact, the way they work is very like the way we work.” Traditional software is static. Human writes logic. Machine executes it. Neural networks don’t work that way. Hinton: “You write a computer program to tell a neural network how to learn. But once it starts learning, it extracts structure from data.” We don’t code its behavior. We code the environment. Then it grows. Hinton: “The system you’ve got at the end has extracted its structure from the data. It’s not something that anybody programmed. We don’t exactly know how it’s gonna work.” We are deploying systems into the global economy actively writing their own internal logic. Right now. Hinton: “Some people think it’ll be fine because we make them and we’ll build them in such a way that we can always control them.” Hinton: “But these things that will be intelligent, they’ll be like us.” You cannot hardcode guardrails on something that out-thinks you. We are not building a tool. We are building our replacement. And the moment you realize that, everything about this race changes.

Dustin

19,025 views • 5 months ago

*Singularity Tingles Intensify* The singularity feels more real than ever and the vibe has officially shifted. We are hearing that 2026 is the year AI really starts to accelerate science itself. If you are reading this you might feel like you are constantly two or three years ahead of the curve. That is a lonely place to be because we are living through an epistemic seismic shift that our chimpanzee brains are not evolved to handle. Our brains are designed for a local and geometric world not a global and exponential one. We literally do not have the neural machinery to understand hyper objects like the singularity so we fall back on normalcy bias to keep us sane. Consider this wild stat almost half of Americans who have used ChatGPT still believe it is just fetching responses from a database rather than generating them in real time. That is the gap between the consensus reality and what is actually happening. While people argue over whether AI is just a search engine we are seeing models like Claude write nearly 100% of the code for their own next iterations. We have basically reached AGI for coding. I want to introduce a concept called Cognitive Lacuna. You know cognitive dissonance where two conflicting ideas cause pain. A cognitive lacuna is different it is when you know there is a shape missing from what you are trying to understand. You can sense the gradient of the answer but you cannot name it yet. That is where human intuition lives and that is the next frontier for AI coherence. We are moving from chatbots that hallucinate to intelligences that can identify exactly what is missing from our scientific understanding. If you feel disoriented right now that just means you are paying attention. Are you feeling the acceleration picking up for 2026 or do you think the curve is flattening?
27:39

*Singularity Tingles Intensify* The singularity feels more real than ever and the vibe has officially shifted. We are hearing that 2026 is the year AI really starts to accelerate science itself. If you are reading this you might feel like you are constantly two or three years ahead of the curve. That is a lonely place to be because we are living through an epistemic seismic shift that our chimpanzee brains are not evolved to handle. Our brains are designed for a local and geometric world not a global and exponential one. We literally do not have the neural machinery to understand hyper objects like the singularity so we fall back on normalcy bias to keep us sane. Consider this wild stat almost half of Americans who have used ChatGPT still believe it is just fetching responses from a database rather than generating them in real time. That is the gap between the consensus reality and what is actually happening. While people argue over whether AI is just a search engine we are seeing models like Claude write nearly 100% of the code for their own next iterations. We have basically reached AGI for coding. I want to introduce a concept called Cognitive Lacuna. You know cognitive dissonance where two conflicting ideas cause pain. A cognitive lacuna is different it is when you know there is a shape missing from what you are trying to understand. You can sense the gradient of the answer but you cannot name it yet. That is where human intuition lives and that is the next frontier for AI coherence. We are moving from chatbots that hallucinate to intelligences that can identify exactly what is missing from our scientific understanding. If you feel disoriented right now that just means you are paying attention. Are you feeling the acceleration picking up for 2026 or do you think the curve is flattening?

David Shapiro (L/0)

27,956 views • 7 months ago