Spr

engineering

[Krishna Agrawal @Krishnasagrawal]

Andrej Karpathy’s 1-hour Stanford lecture on AI engineering is one of the best explanations I’ve seen of how AI systems actually work.

The progression is simple:

10% → LLM
30% → Prompt
50% → Agent
70% → Loop
100% → Graph

The key takeaway:

AI engineering isn’t just about writing better prompts.

It’s about building systems around models — giving them context, memory, tools, feedback loops, and data flows.

“Delete everything, keep Graph.”

Definitely worth watching if you’re building with AI agents.

Watch → Bookmark →

[the lich @thelichhh]

An Indian institute of technology filmed a full course on how money actually moves through a banking system and released it for nothing.

The course is Economics of Banking and Finance Markets, produced at IIT Kanpur under NPTEL, India's national open courseware programme. It runs in sequence as one continuous argument rather than a set of clips.

A green chalkboard, a lapel microphone, a checked shirt, one fixed camera.

He opens with the only question that matters at the start. Where does surplus money sit, who needs it, and what stands between them.

Then he splits the answer in 2. Direct finance, where the saver reaches the borrower through a market. Indirect finance, where a bank stands in the middle and absorbs the risk instead.

Everything after that hangs off the split. Time value of money, risk, information asymmetry, why institutions exist at all, and what regulators are actually regulating.

The turn is what he does with information. Banks are not warehouses for money. They exist because one side of every transaction knows things the other side does not, and the whole industry is built on charging for that gap.

It matters more now. Every fintech product claims to cut out the middleman, and almost none of them remove the information problem that put him there.

Free, in order, subtitled, funded by a public programme.

The banking system explained from the first principle, for nothing.

Lecture 1. 2 channels. It is in the video.

[Saurabh Bharadwaj @Saurabh_MLAgk]

Welcome to the club

[Marketing Mind @MarketingMind_]

Wish we could do the same.

#MarketingMind #elonmusk #work

[Rahul Gandhi @RahulGandhi]

E20 - दाल में काला नहीं, पूरी दाल ही काली है।

[Nora @0xNoraa]

1 of the greatest mathematicians in the world found out his own famous proof was wrong. It had been accepted for years, only because everyone trusted his name.

His name was Vladimir Voevodsky. He had already won the Fields Medal, math's version of the Nobel Prize. He was a legend. When he proved something, people believed it.

So when another mathematician quietly pointed out a serious mistake in one of his big papers, almost nobody listened. Voevodsky himself brushed it off. He was too respected to be wrong.

But years later, he checked again. And the other man was right. His celebrated proof had a hole in it the whole time.

It broke something in him. He later called it the greatest failure of his life. And it forced him to face a scary truth about his entire field.

Math, he realized, runs on trust. Proofs get so long and complex that almost no one fully checks them. People just believe the famous names. And famous names are sometimes wrong.

So he stopped his old research and spent the rest of his life building a new kind of math, one where a computer checks every single step, so no proof can ever hide behind a reputation again.

He died suddenly in 2017, at just 51.

The lesson he left behind is unsettling. A lot of what we call "proven" was really only "trusted."

[Zyphor @Zyphorrl]

A PROFESSOR WHOSE STUDENTS CALL HIM THE MOST ENGAGING TEACHER THEY HAVE EVER HAD WALKED INTO A LECTURE HALL AND SHOWED WHY A CALCULATOR CAN DRAW ANY CURVE - BUT CANNOT TELL YOU WHY THE CURVE IS SHAPED THE WAY IT IS

He opens with one claim - a calculator will give you the picture, but it will never give you the understanding. To know why a curve rises here and falls there you need the first derivative test. That is what this lecture is about.

He starts with the setup. The first derivative tells you the slope. Positive slope means the function is increasing. Negative slope means it is decreasing. Zero slope means a horizontal tangent - and that is where something interesting might happen. Those are the only three cases. Everything else follows from them.

Then the first derivative test. Take the derivative. Set it equal to zero and solve for the critical numbers. Draw a number line, mark those critical numbers on it, and pick one test point in each interval between them. Plug each test point into the derivative - not the original function, the derivative. You do not care what number comes out. You only care whether it is positive or negative.

Then what the signs tell you. Positive on the left of a critical number, negative on the right - the function was climbing and then fell. That is a relative maximum. Negative on the left, positive on the right - the function fell and then climbed. That is a relative minimum. If the sign does not change, neither maximum nor minimum exists at that point.

Then the case with a denominator. When the derivative is a fraction, set the numerator equal to zero for critical numbers. But also set the denominator equal to zero - because the function can switch from increasing to decreasing at a point where the slope is undefined, not just where it is zero. Miss those points and the table is wrong.

Watch the moment he plugs test values into a derivative with a cube root in the denominator and shows that even though the slope does not exist at zero, the function still has a relative maximum there. Undefined slope does not mean nothing happens. It means you have to check.

A first year engineering student I know rewatched this lecture before a midterm on curve sketching. Said it was the first time the connection between the derivative sign chart and the actual shape of the graph felt obvious rather than memorized.

Free on YouTube.

bookmark this and watch later - after this lecture every curve you draw will feel like a story the first derivative is already telling you

[Mahax @Mahaximus_]

Google Brain founder, Andrew Ng:

"Prompting will die in 6 months.

Loops and Graphs are what's replacing it."

In 2 hours, he shows exactly what the best engineers already build instead, and how to start building it yourself.

The missing piece most people skip: how to connect those loops into a graph that compounds every time it runs.

Watch it, then read the full guide on loops and graphs below.

[Tukaram Mundhe @Tukaram_IndIAS]

The Best Leaders Build Systems That No Longer Need Them.
#TukaramMundhe #Leadership #Inpiration #Public #GenZ
https://t.co/bGFt8YliUt

[Zyphor @Zyphorrl]

HARVARD FILMED A PROFESSOR EXPLAIN IN 48 MINUTES WHAT TOOK MATHEMATICIANS 200 YEARS TO FIGURE OUT - AND SHOWS WHY THE SMARTEST MAN WHO EVER LIVED GOT THE INTUITION COMPLETELY WRONG

This is Joe Blitzstein, Harvard Statistics 110, lecture 4. He has won Harvard's Excellence in Teaching award multiple times, his textbook is used in over 200 universities worldwide, and his course has been taken by over 2 million people across 190 countries. He opens with one claim - conditioning is the soul of statistics. Everything else in the course follows from that.

He starts with De Montmort's matching problem. A deck of n cards labeled 1 through n, flipped one by one. What is the probability at least one card lands in its own position? The exact answer collapses into 1 minus 1 plus 1 over 2 factorial minus 1 over 3 factorial, continuing to n terms - which is exactly the Taylor series for e to the x at minus 1. The probability of no match converges to 1 over e, which is 0.37, no matter how large the deck gets.

Then the Newton-Pepys problem, 1693. At least one six from six dice, at least two sixes from twelve dice, or at least three sixes from eighteen dice - which is most likely? Pepys bet on the third. Wrote to Newton. Newton calculated correctly and showed it was the first, probability 0.665. Then the real punch - Newton's intuitive argument was wrong, and a statistician named Stigler proved it without even understanding what Newton wrote. Newton's argument never used the fact that the dice were fair. So it could not possibly be correct.

Then conditional probability. The definition is a single fraction - probability of A given B equals probability of A and B divided by probability of B. Blitzstein explains it two ways. First, pebble world - learning B occurred simply erases every outcome outside B and renormalizes what remains to sum to 1. Second, frequentist world - repeat the experiment many times, circle every run where B occurred, ask what fraction of those also had A.

Watch the moment he derives Bayes' rule in 10 seconds - divide both sides by P of B, end of proof. Then says controversies about this one line have raged for centuries and an entire field of statistics was built on top of it.

A data scientist I know rewatched this before switching careers into statistics. Said it was the first time probability felt like a system with rules rather than a collection of tricks.

Free on YouTube, Harvard, over 2 million views.

bookmark this and watch later - after this lecture every time you learn something new you will feel yourself updating a probability rather than changing your mind

[Lunar @LunarResearcher]

Ex-Google engineer just released a free 3-hour course on building and monetizing AI agents.

How to go from one agent to a full system that finds leads and makes money:

00:00 - Design an AI agent system
07:38 - Add human handoffs
19:27 - Understand RAG and vector databases
52:38 - Deploy agents to Google Cloud
1:25:57 - Turn agents into a paid WhatsApp business
1:33:00 - Convert conversations into leads
2:24:01 - Learn loops vs graphs
2:33:08 - Build a multi-tool agent graph

Most people are still building AI demos.

This course shows the full stack:

Agents → RAG → Deployment → Leads → Revenue

Building agents is the old workflow.

Monetizing agent systems is the new one.

This 3-hour watch covers more than most $500 paid courses.

Save it before everyone starts selling the same thing.

[Anatoli Kopadze @AnatoliKopadze]

instead of watching 2 hours of Netflix tonight, watch this Stanford lecture given by Anthropic engineers

it's the clearest explanation I've seen of how AI agents actually work

useful whether you've never touched AI in your life or have been building with it every day

i took the key ideas and turned them into a practical guide, with ready-to-copy prompts

watch it, then read the guide below on how to build a system that improves itself

[Avid @Av1dlive]

Don't waste 2 years learning to build LLMs like Claude & ChatGPT.

Stanford just dropped a 2 hour 34 min course on how to build LLMs from scratch.

• 00:00 - LLM tokenization
• 25:44 - how LLMs decode user prompts
• 35:40 - training pipeline of LLMs
• 1:16:47 - LLM architecture from scratch

Anthropic pays $750,000/year to engineers who understand this exact knowledge of LLMs.

Bookmark this & give 2 hours today, no matter what. Then read the article below.

[Rohan Paul @rohanpaul_ai]

“If you hit recursive self-improvement, that curve will go to hyperexponential, and that is a key part of the investment thesis, the scientific thesis, and a key part of why society’s investing what it’s currently investing in.”

Google DeepMind Chief Strategy Officer Jasjeet Sekhon says AI infrastructure spending is financing a hoped-for self-improvement loop.

In its strongest form, recursive self-improvement (RSI) will let AI design, evaluate, and train more capable successors with progressively less human supervision.

The current evidence is still narrower: e.g. Google's AlphaEvolve proposes algorithms, then automated evaluators run and score them against human-defined objectives.

Google says its resulting kernel changes reduced Gemini training time by 1%, an example of bounded improvement rather than autonomous successor design.

----

From "Berkeley RDI" YouTube channel, (full video link in comment)

[Aravind @aravind]

34 mins of concentration on what Krishna says can improve your thinking, finances, life and make you a Jaganathan.

[Sagar @sagarcasm]

When Dharmendra Pradhan is made to resign

Nitin Gadkari:

[Abhishek @be_mewadi]

After resignation of Education minister
#DharmendraPradhan

[Hanako @hanakoxbt]

Anthropic Quant Andrej Karpathy to Andrew Ng:

"I wrote my own neural net library in JavaScript. Not to ship it - just to see backprop with my own eyes"

before any framework, he built convnetjs from scratch - every layer by hand.

his advice: skip TensorFlow at first. write it yourself at the lowest level, then pick up the framework once you know what's under the hood.

the line that lands: "if all you know is a stack of layers and that dropout helps, you can't debug or improve the model."

Watch the talk, then save the loop guide below.

[Movez @0xMovez]

Andrew Ng just dropped a 3-hour course on how to become an AI Engineer in 2026:

• 00:00 - How to build agentic AI systems
• 04:25 - Future of AI engineering
• 23:38 - AI Prompting full course
• 2:52:17 - Creating an app with AI in 30 minutes

This 3-hour watch could replace 10 AI engineering courses on the internet.

Watch it today, then read how to run a self-improving system in the article below.