Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

Can you solve these agentic AI math problems by hand ✍️? Getting a bit harder now. Download PDF: Problems 11 to 15: 11. Comparing three calls: same total tokens, three different bills 12. The average call: size a sample, then scale it to the day 13. The system prompt:... show more

Tom Yeh

64,941 subscribers

41,681 views • 20 days ago •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

New chapter! Context problems. Who can solve these by hand ✍️? Download PDF: The first five: 1. Mark the context: the window as cells you can count 2. What fills the window: the system prompt, the user, the model's reply 3. Turn by turn: every turn is the same block, and most of it is the reply 4. Will it fit? Number a new message into what's left, and see it run out 5. Turns until full: how many rounds before there is no room Previously, I shared Chapter 1: Token Problems 1 ~ 20 Why am I designing these math problems? AI is making people (including me) think less. It's just too easy to ask AI a few questions about a new concept and start to believe I get it. We used to practice on a coding problem, but now AI writes the code, and solving it technically is not the same as knowing anything. So I recast each concept as a simple math problem you have to work with pen and paper. Using our hands is one way to motivate ourselves to start thinking again. ✍️ If you solve them by hand, leave a comment to let me know. I will give you a shout out in my next post. Download, print, solve, and learn. ✍️ ~ Prof. Tom Yeh

New chapter! Context problems. Who can solve these by hand ✍️? Download PDF: The first five: 1. Mark the context: the window as cells you can count 2. What fills the window: the system prompt, the user, the model's reply 3. Turn by turn: every turn is the same block, and most of it is the reply 4. Will it fit? Number a new message into what's left, and see it run out 5. Turns until full: how many rounds before there is no room Previously, I shared Chapter 1: Token Problems 1 ~ 20 Why am I designing these math problems? AI is making people (including me) think less. It's just too easy to ask AI a few questions about a new concept and start to believe I get it. We used to practice on a coding problem, but now AI writes the code, and solving it technically is not the same as knowing anything. So I recast each concept as a simple math problem you have to work with pen and paper. Using our hands is one way to motivate ourselves to start thinking again. ✍️ If you solve them by hand, leave a comment to let me know. I will give you a shout out in my next post. Download, print, solve, and learn. ✍️ ~ Prof. Tom Yeh

Tom Yeh

11,641 views • 18 days ago

Grok is quietly turning into an agentic OS. Agents are processes you give custom system prompts to. Connectors are the system services that plug it into Gmail, GitHub, Notion, and whatever else you wire in. Skills are executables that fire from prompts, call connectors, and call other skills. Four built in the video. Grok Imagine Cinematic. Single video idea in, multi-scene prompt set out, ready to drop into Grok Imagine Agent mode. YouTube to cheat sheet. Lecture link in, polished LaTeX PDF out. Under the hood it calls a LaTeX specialist skill and a graph generator skill. Keep each one atomic. Treat Grok Skills like Unix tools.

Grok is quietly turning into an agentic OS. Agents are processes you give custom system prompts to. Connectors are the system services that plug it into Gmail, GitHub, Notion, and whatever else you wire in. Skills are executables that fire from prompts, call connectors, and call other skills. Four built in the video. Grok Imagine Cinematic. Single video idea in, multi-scene prompt set out, ready to drop into Grok Imagine Agent mode. YouTube to cheat sheet. Lecture link in, polished LaTeX PDF out. Under the hood it calls a LaTeX specialist skill and a graph generator skill. Keep each one atomic. Treat Grok Skills like Unix tools.

tetsuo

123,795 views • 3 months ago

ONE PROMPT AND ONE BUTTON NOW BUILDS A FULL SYSTEM, DATABASE, BACKEND, AND FRONTEND INCLUDED Most people still picture AI as something you ask questions to, then copy the answer into your own code by hand. A builder wired Claude into Obsidian through MCP instead, so a single written prompt becomes the actual spec for the system. Claude breaks that prompt into database schema, UI components, and API endpoints on its own, then writes, runs, and fixes the code until it actually works. MCP is what makes this possible, it lets the agent see the filesystem, installed packages, and running services before it starts building anything. The output isn't a code snippet sitting in a chat window, it's a live local app running on its own server, ready to open in a browser. What used to need a backend developer, a DevOps engineer, and two weeks now finishes in under fifteen minutes. See the full architecture and prompt structure below👇

ONE PROMPT AND ONE BUTTON NOW BUILDS A FULL SYSTEM, DATABASE, BACKEND, AND FRONTEND INCLUDED Most people still picture AI as something you ask questions to, then copy the answer into your own code by hand. A builder wired Claude into Obsidian through MCP instead, so a single written prompt becomes the actual spec for the system. Claude breaks that prompt into database schema, UI components, and API endpoints on its own, then writes, runs, and fixes the code until it actually works. MCP is what makes this possible, it lets the agent see the filesystem, installed packages, and running services before it starts building anything. The output isn't a code snippet sitting in a chat window, it's a live local app running on its own server, ready to open in a browser. What used to need a backend developer, a DevOps engineer, and two weeks now finishes in under fifteen minutes. See the full architecture and prompt structure below👇

wast3

20,348 views • 1 month ago

What are your thoughts on Chris Christie's words to Trump supporters in Florida? Does he have a point? “Now look, every one of those boos, everyone one of those cat calls, everyone one of those yells will not … solve one problem we face in this country. Your anger against the truth is reprehensible.” “When you think about the problems that our country and this world is facing, when you think about that, this type of pettiness … is beneath … the process of electing a president.” “The problem is … you fear the truth. You want to shout down any voice that says anything different than what you want to hear. And you can continue to do it. And believe me … it doesn’t bother me one bit.”

What are your thoughts on Chris Christie's words to Trump supporters in Florida? Does he have a point? “Now look, every one of those boos, everyone one of those cat calls, everyone one of those yells will not … solve one problem we face in this country. Your anger against the truth is reprehensible.” “When you think about the problems that our country and this world is facing, when you think about that, this type of pettiness … is beneath … the process of electing a president.” “The problem is … you fear the truth. You want to shout down any voice that says anything different than what you want to hear. And you can continue to do it. And believe me … it doesn’t bother me one bit.”

Ed Krassenstein

3,125,712 views • 2 years ago

A 19 year old gets on Zoom calls with business owners and shows them how they're about to throw away $600,000. He opens with one question. How much are you paying your receptionist? The dental office owner says $60,000. The kid says: over the next 10 years that's $600,000 going to someone who calls in sick, takes lunch breaks and goes home at 5. For a job a machine can do for $12,000 once. Then the second question. How many calls do you miss every day? Voicemails nobody calls back. People who hang up after three rings. New patients who try once and never try again. The owner stops. Pulls up his phone log. Counts. 8 missed calls a day. Sometimes more. Each one is a $200 patient walking past the front desk while it's empty for lunch. $1,600 a day. Almost $600,000 a year in revenue he never knew was leaving. That number sits on top of the $600,000 in salary. The owner is staring at a chart he never let himself draw before. That's when the kid says it. I can build you something for $12,000. Picks up every call. Doesn't sleep. Doesn't take vacation. Books patients while you're operating on someone else. Pays for itself in the first week. The $12,000 invoice doesn't feel like an expense. It feels like a refund. The kid closes 3 to 4 of these calls a month. $50,000 in revenue. He's 19. Behind the scenes he's not building anything from scratch. He's running Kimi K2.6 with 300 parallel sub agents, paying $0.60 per million tokens instead of $5 with Claude. The same reception agent a real developer would charge $40,000 to build, he ships in 30 minutes. 12 hour autonomous sessions. Zero human involvement. One prompt and the agent goes live. Answering calls. Booking patients. Routing emergencies. Before the owner finishes his coffee. His friends are working summer shifts at coffee shops for $14 an hour. He's making $600,000 a year by walking into a Zoom call and naming the most uncomfortable number a business owner has on his books. His secret isn't technology. He never argues with the owner about whether AI works. He hands the owner a calculator and asks the questions the owner has been avoiding for years. By minute 8 the owner has done the math himself. By minute 12 the kid quotes the price. By minute 15 the contract is signed. He told his dad about it once. His dad is an insurance salesman. 30 years selling the same product to every client for the same reason: fear of losing what they already have. His dad listened. Then said: you sell the same thing I sell. You just call it AI. The kid said: I don't even call it that. I let them call it whatever they want. His dad smiled. Walked out of the room. The kid had another call in 10 minutes.

A 19 year old gets on Zoom calls with business owners and shows them how they're about to throw away $600,000. He opens with one question. How much are you paying your receptionist? The dental office owner says $60,000. The kid says: over the next 10 years that's $600,000 going to someone who calls in sick, takes lunch breaks and goes home at 5. For a job a machine can do for $12,000 once. Then the second question. How many calls do you miss every day? Voicemails nobody calls back. People who hang up after three rings. New patients who try once and never try again. The owner stops. Pulls up his phone log. Counts. 8 missed calls a day. Sometimes more. Each one is a $200 patient walking past the front desk while it's empty for lunch. $1,600 a day. Almost $600,000 a year in revenue he never knew was leaving. That number sits on top of the $600,000 in salary. The owner is staring at a chart he never let himself draw before. That's when the kid says it. I can build you something for $12,000. Picks up every call. Doesn't sleep. Doesn't take vacation. Books patients while you're operating on someone else. Pays for itself in the first week. The $12,000 invoice doesn't feel like an expense. It feels like a refund. The kid closes 3 to 4 of these calls a month. $50,000 in revenue. He's 19. Behind the scenes he's not building anything from scratch. He's running Kimi K2.6 with 300 parallel sub agents, paying $0.60 per million tokens instead of $5 with Claude. The same reception agent a real developer would charge $40,000 to build, he ships in 30 minutes. 12 hour autonomous sessions. Zero human involvement. One prompt and the agent goes live. Answering calls. Booking patients. Routing emergencies. Before the owner finishes his coffee. His friends are working summer shifts at coffee shops for $14 an hour. He's making $600,000 a year by walking into a Zoom call and naming the most uncomfortable number a business owner has on his books. His secret isn't technology. He never argues with the owner about whether AI works. He hands the owner a calculator and asks the questions the owner has been avoiding for years. By minute 8 the owner has done the math himself. By minute 12 the kid quotes the price. By minute 15 the contract is signed. He told his dad about it once. His dad is an insurance salesman. 30 years selling the same product to every client for the same reason: fear of losing what they already have. His dad listened. Then said: you sell the same thing I sell. You just call it AI. The kid said: I don't even call it that. I let them call it whatever they want. His dad smiled. Walked out of the room. The kid had another call in 10 minutes.

Marlow

295,813 views • 3 months ago

Fresh off the launch of Opus 5, Claude Code creator Boris Cherny (Boris Cherny) joins YC's Diana at Startup School 2026 to talk about what the newest models can do, how Claude Code came to be, and what it means to build products when the underlying capabilities keep accelerating. 00:07 — What Makes Opus 5 Different 02:06 — Solving Prompt Injection 03:21 — Why Claude Code Deleted 80% of Its System Prompt 06:37 — Press Delete on Your AI Product 07:20 — How to Rebuild Your System Prompt 10:30 — Product Overhang and “Unhobbling” AI 14:26 — Give Claude Harder Problems 19:32 — Prompt Engineering Is Changing 21:57 — The Two-Week Claude Code Prompt 24:42 — Running Thousands of AI Agents 30:15 — Coding Is (Almost) Solved 32:20 — What Every CS Student Should Still Learn Anthropic Claude

Fresh off the launch of Opus 5, Claude Code creator Boris Cherny (Boris Cherny) joins YC's Diana at Startup School 2026 to talk about what the newest models can do, how Claude Code came to be, and what it means to build products when the underlying capabilities keep accelerating. 00:07 — What Makes Opus 5 Different 02:06 — Solving Prompt Injection 03:21 — Why Claude Code Deleted 80% of Its System Prompt 06:37 — Press Delete on Your AI Product 07:20 — How to Rebuild Your System Prompt 10:30 — Product Overhang and “Unhobbling” AI 14:26 — Give Claude Harder Problems 19:32 — Prompt Engineering Is Changing 21:57 — The Two-Week Claude Code Prompt 24:42 — Running Thousands of AI Agents 30:15 — Coding Is (Almost) Solved 32:20 — What Every CS Student Should Still Learn Anthropic Claude

Y Combinator

85,554 views • 1 month ago

This is one of my favorite movie endings... It has a powerful lesson: "At some point, everything is going to go south on you...and you're gonna say 'this is it, this is how I end.' Now you can either accept that, or you can get to work. That's all it is, you just begin. You do the math. You solve one problem. Then you solve the next one. And the next. And if you solve enough problems, you get to come home." This is the beautiful thing about life: No matter where you are today—no matter how deep in the darkness—you are always just one good decision away from being in a better place tomorrow. You can always focus on that one decision—that one problem—in front of you. Don't worry about the hundreds or thousands of decisions that you still have to make to get to where you want to be—just focus on the next decision. Just begin. Solve one problem. Then solve the next one. And the next. "Life is really simple, but we insist on making it complicated." - Confucius

This is one of my favorite movie endings... It has a powerful lesson: "At some point, everything is going to go south on you...and you're gonna say 'this is it, this is how I end.' Now you can either accept that, or you can get to work. That's all it is, you just begin. You do the math. You solve one problem. Then you solve the next one. And the next. And if you solve enough problems, you get to come home." This is the beautiful thing about life: No matter where you are today—no matter how deep in the darkness—you are always just one good decision away from being in a better place tomorrow. You can always focus on that one decision—that one problem—in front of you. Don't worry about the hundreds or thousands of decisions that you still have to make to get to where you want to be—just focus on the next decision. Just begin. Solve one problem. Then solve the next one. And the next. "Life is really simple, but we insist on making it complicated." - Confucius

Sahil Bloom

213,822 views • 2 years ago

A Berkeley professor just leaked Anthropic’s CCA exam that opens the door to $750,000 jobs. In 20 minutes he covers every single topic on the exam and explains how to actually solve it. This exam shows exactly what you need to know about AI right now and what’s not worth your time. Watch the presentation first, then read the guide below on how to build a system that improves itself.

A Berkeley professor just leaked Anthropic’s CCA exam that opens the door to $750,000 jobs. In 20 minutes he covers every single topic on the exam and explains how to actually solve it. This exam shows exactly what you need to know about AI right now and what’s not worth your time. Watch the presentation first, then read the guide below on how to build a system that improves itself.

Anatoli Kopadze

196,539 views • 20 days ago

69% of baseball fans say they'd rather have a computer vision AI system call balls and strikes than a human umpire. This season, the MLB gave them one. For the first time in league history, a human ump's ball-strike call is not final. A Sony computer vision system called Hawk-Eye makes the ruling. It can read the seam pattern on the ball, measure spin axis, and detect spin decay mid-flight. Hawk-Eye's Head of Computer Vision Engineering says the pipeline runs "various AI and machine learning models" from camera capture to data output. On Saturday, one umpire had 6 of 8 challenged calls overturned. Three of those missed by over 2 inches. The team ran out of challenges by the fourth inning trying to correct him, then the manager got thrown out for arguing a call they couldn't challenge. An AI computer vision system accurate to a sixth of an inch, sitting next to a human who misses by two. The crowd cheered for the machine.

69% of baseball fans say they'd rather have a computer vision AI system call balls and strikes than a human umpire. This season, the MLB gave them one. For the first time in league history, a human ump's ball-strike call is not final. A Sony computer vision system called Hawk-Eye makes the ruling. It can read the seam pattern on the ball, measure spin axis, and detect spin decay mid-flight. Hawk-Eye's Head of Computer Vision Engineering says the pipeline runs "various AI and machine learning models" from camera capture to data output. On Saturday, one umpire had 6 of 8 challenged calls overturned. Three of those missed by over 2 inches. The team ran out of challenges by the fourth inning trying to correct him, then the manager got thrown out for arguing a call they couldn't challenge. An AI computer vision system accurate to a sixth of an inch, sitting next to a human who misses by two. The crowd cheered for the machine.

The Rundown AI

198,387 views • 5 months ago

Anthropic and Andrew Ng built an agent that uses 90% fewer tokens from scratch: they dropped the entire book of Frankenstein into a prompt - 108,000 tokens asked one question the input dropped from 108,000 tokens to 11 here's how: step 1 → put everything that never changes at the top - tools, then system, then docs step 2 → mark where the static part ends. everything above it gets cached step 3 → one stray space breaks it - and you pay full price again step 4 → the cache dies in 5 min - every read resets the clock step 5 → cached tokens don't count against your rate limits. free headroom most people never touch this - it pays for itself on day one watch & bookmark - this 1-hour brilliant course ↓

Anthropic and Andrew Ng built an agent that uses 90% fewer tokens from scratch: they dropped the entire book of Frankenstein into a prompt - 108,000 tokens asked one question the input dropped from 108,000 tokens to 11 here's how: step 1 → put everything that never changes at the top - tools, then system, then docs step 2 → mark where the static part ends. everything above it gets cached step 3 → one stray space breaks it - and you pay full price again step 4 → the cache dies in 5 min - every read resets the clock step 5 → cached tokens don't count against your rate limits. free headroom most people never touch this - it pays for itself on day one watch & bookmark - this 1-hour brilliant course ↓

codila

544,277 views • 1 month ago

A developer in Hangzhou runs an AI that remembers everything about him for $0.40 a year. No vector database. One file that never grows past 4,000 tokens. He published the whole schema. His version starts from the opposite idea. Memory is not storage. It's a write policy. Six fields. Rewritten every time, never appended: > IDENTITY - who you are, what you build. 300 tokens. Changes monthly at most > STATE - what you're on right now. 400 tokens. Rewritten daily > DECISIONS - what's already settled, so nothing gets re-argued. 800 tokens > CORRECTIONS - every time you said "no, not like that." 600 tokens > PEOPLE - names, roles, who's waiting on what. 500 tokens > DEAD - tried and abandoned, so it never comes back as a suggestion. 400 tokens Three thousand tokens. Ceiling of four. When a section fills, the model rewrites it shorter. Nothing is ever added. Only replaced. Kimi K2.5 bills $0.10 per million cached input tokens. Four thousand tokens a turn is $0.0004. That's 2,500 turns for a dollar. The free tier hands you 1.5 million tokens a day. 375 turns before you pay anything at all. CORRECTIONS is the field nobody builds, and it's the one that does the work. A model that remembers being wrong stops repeating it. Everyone else is paying to search their own history. He pays to keep it short. The bill stopped growing when the file did. Your memory system isn't defined by what it stores. It's defined by what it agrees to delete. The article below is the full build - schema, rewrite prompts, the compaction rule that keeps it under the cap. Save it. You'll want it open in the other tab.

A developer in Hangzhou runs an AI that remembers everything about him for $0.40 a year. No vector database. One file that never grows past 4,000 tokens. He published the whole schema. His version starts from the opposite idea. Memory is not storage. It's a write policy. Six fields. Rewritten every time, never appended: > IDENTITY - who you are, what you build. 300 tokens. Changes monthly at most > STATE - what you're on right now. 400 tokens. Rewritten daily > DECISIONS - what's already settled, so nothing gets re-argued. 800 tokens > CORRECTIONS - every time you said "no, not like that." 600 tokens > PEOPLE - names, roles, who's waiting on what. 500 tokens > DEAD - tried and abandoned, so it never comes back as a suggestion. 400 tokens Three thousand tokens. Ceiling of four. When a section fills, the model rewrites it shorter. Nothing is ever added. Only replaced. Kimi K2.5 bills $0.10 per million cached input tokens. Four thousand tokens a turn is $0.0004. That's 2,500 turns for a dollar. The free tier hands you 1.5 million tokens a day. 375 turns before you pay anything at all. CORRECTIONS is the field nobody builds, and it's the one that does the work. A model that remembers being wrong stops repeating it. Everyone else is paying to search their own history. He pays to keep it short. The bill stopped growing when the file did. Your memory system isn't defined by what it stores. It's defined by what it agrees to delete. The article below is the full build - schema, rewrite prompts, the compaction rule that keeps it under the cap. Save it. You'll want it open in the other tab.

wast3

15,862 views • 7 days ago

We just beat the best frontier models at a quarter of the price. Now you can too, in one API call. Today we're launching Merge Fusion. Send one prompt to a panel of models, and a judge writes one answer better than any of them alone. Fusion beats Fable 5 outright at 1/4th the cost. It also sets the highest score on the DRACO benchmark.

We just beat the best frontier models at a quarter of the price. Now you can too, in one API call. Today we're launching Merge Fusion. Send one prompt to a panel of models, and a judge writes one answer better than any of them alone. Fusion beats Fable 5 outright at 1/4th the cost. It also sets the highest score on the DRACO benchmark.

Shensi Ding

409,995 views • 1 month ago

Sam Altman shared a number that should change how you think about AI. Six and a half years ago, the heaviest AI user on the planet was an OpenAI employee. Using 100,000 tokens a month. That was considered insane at the time. The global average was basically zero. Today the global average is 100,000 tokens a month. What was insane is now normal. And the heaviest user at OpenAI right now? Hundreds of billions of tokens a month. Now here's where it gets wild. If the same pattern repeats, in another six and a half years the average person will use 500 billion tokens a month. That's a 5 million x increase from today. Think about what that means. → AI is not something you open once a day to ask a question anymore. It becomes always on. Running in the background across your work, education, personal life. → Agentic workflows, coding assistants, multi-step reasoning — that's what consumes billions of tokens. Not chatting. → OpenAI is already building for this. Hundreds of billions in infrastructure spend through 2030. Stargate. Partnerships with Microsoft, Oracle, Nvidia. And cost per token keeps falling roughly 10x every 12 months for the same capability level. So usage explodes because it keeps getting cheaper. The pattern is the same one we saw with computing, internet, and mobile. What the power user does today, the average person does in six years. Right now you are the power user. Act like it.

Sam Altman shared a number that should change how you think about AI. Six and a half years ago, the heaviest AI user on the planet was an OpenAI employee. Using 100,000 tokens a month. That was considered insane at the time. The global average was basically zero. Today the global average is 100,000 tokens a month. What was insane is now normal. And the heaviest user at OpenAI right now? Hundreds of billions of tokens a month. Now here's where it gets wild. If the same pattern repeats, in another six and a half years the average person will use 500 billion tokens a month. That's a 5 million x increase from today. Think about what that means. → AI is not something you open once a day to ask a question anymore. It becomes always on. Running in the background across your work, education, personal life. → Agentic workflows, coding assistants, multi-step reasoning — that's what consumes billions of tokens. Not chatting. → OpenAI is already building for this. Hundreds of billions in infrastructure spend through 2030. Stargate. Partnerships with Microsoft, Oracle, Nvidia. And cost per token keeps falling roughly 10x every 12 months for the same capability level. So usage explodes because it keeps getting cheaper. The pattern is the same one we saw with computing, internet, and mobile. What the power user does today, the average person does in six years. Right now you are the power user. Act like it.

Vaibhav Sisinty

24,323 views • 25 days ago

I'm making over $1,000 an hour with one AI offer. The entire thing runs on Claude Opus 4.8. I call it the AI Concierge. Clients pay me $1,500+ a month for two 45-minute calls where we build their AI systems live, on their screen. I have 4 clients. I'm capping at 6. Here's the entire model: 1) The intake form is the audit. A 10-minute JotForm (built by Claude) surfaces their time sinks and hands me 1-3 AI opportunities before call one. 2) Done-with-you, not done-for-you. They share their screen. We build skills, set up Cowork, and write context files together. They learn to drive. (Done-for-you is the upsell.) 3) Every session runs through AOA: Audit, Optimize, Automate. Fix the process first, then turn it into a skill. Automating chaos just gives you faster chaos. 4) Day one has to move the needle. We ship at least one skill or automation on call one. No first-call win, dead engagement. 5) Unlimited Voxer between calls. They send a voice message, I reply in under 12 business hours. A 24/7 partner, not a guy they see twice a month. 6) The Notion hub is the renewal mechanism. Every call logs a quantified list of what we built. "Call one: 2 skills, 3 context files, Cowork live" makes $1.5K a month a no-brainer. 7) I never fill Notion out by hand. Two Claude skills log the call, pull the action items, and draft the recap email. 30 seconds. 8) Pricing ladder: $1,000/month, then $1,500 at 2 clients, then $1,800. At $1,500 you're already at $1,000/hour. If everyone says yes then you're priced too low. Two things that make this work: 1) Build the fulfillment infrastructure once. An afternoon. Then it runs itself outside the calls. 2) The value must be visible. People renew what they can measure. Full breakdown below. Go watch.

I'm making over $1,000 an hour with one AI offer. The entire thing runs on Claude Opus 4.8. I call it the AI Concierge. Clients pay me $1,500+ a month for two 45-minute calls where we build their AI systems live, on their screen. I have 4 clients. I'm capping at 6. Here's the entire model: 1) The intake form is the audit. A 10-minute JotForm (built by Claude) surfaces their time sinks and hands me 1-3 AI opportunities before call one. 2) Done-with-you, not done-for-you. They share their screen. We build skills, set up Cowork, and write context files together. They learn to drive. (Done-for-you is the upsell.) 3) Every session runs through AOA: Audit, Optimize, Automate. Fix the process first, then turn it into a skill. Automating chaos just gives you faster chaos. 4) Day one has to move the needle. We ship at least one skill or automation on call one. No first-call win, dead engagement. 5) Unlimited Voxer between calls. They send a voice message, I reply in under 12 business hours. A 24/7 partner, not a guy they see twice a month. 6) The Notion hub is the renewal mechanism. Every call logs a quantified list of what we built. "Call one: 2 skills, 3 context files, Cowork live" makes $1.5K a month a no-brainer. 7) I never fill Notion out by hand. Two Claude skills log the call, pull the action items, and draft the recap email. 30 seconds. 8) Pricing ladder: $1,000/month, then $1,500 at 2 clients, then $1,800. At $1,500 you're already at $1,000/hour. If everyone says yes then you're priced too low. Two things that make this work: 1) Build the fulfillment infrastructure once. An afternoon. Then it runs itself outside the calls. 2) The value must be visible. People renew what they can measure. Full breakdown below. Go watch.

Corey Ganim

104,244 views • 2 months ago

HERMES AGENT WITHOUT THESE 3 FILES IS A CHATBOT. WITH THEM IT KNOWS WHO IT IS, WHO YOU ARE, AND WHAT IT LEARNED. SOUL.md — who the agent is. first thing in the system prompt. defines personality, voice, values, how it operates, what it can and can't do. structure yours like this: → identity (name, role, relationship to you) → values (what matters, what principles guide decisions) → voice (how it communicates, tone, style) → operations (autonomy level, ground rules) → restrictions (what it never does) → failure protocol (how to operate when things break) lives in ~/.hermes/SOUL.md. auto-seeded on first install. edit anytime. scanned for prompt injection on every load. keep it concise. SOUL.md injects into every turn. a 200-line soul burns tokens on every message. aim for 50-80 lines. one paragraph per section. MEMORY.md — what the agent remembers. persistent facts, insights, preferences. survives across sessions and restarts. capped at ~800 tokens by default: memory: memory_char_limit: 2200 the agent writes to this automatically as it learns about your work. USER.md — who you are. your profile, preferences, context. capped at ~500 tokens by default: memory: user_char_limit: 1375 injected every turn so the agent always knows who it's working for. bonus: AGENTS.md — project-specific instructions. drop one in any project folder. subdirectory AGENTS.md files load lazily during tool calls, not at startup. keeps your system prompt light. prompt stack order on every turn: SOUL.md → tool guidance → MEMORY.md + USER.md → skills index → AGENTS.md → platform hints skills come preloaded. 60+ built-in tools. the agent creates more skills from completed work. you focus on these 3 files. each profile gets its own copy: ~/.hermes/SOUL.md (default profile) ~/.hermes/profiles/researcher/SOUL.md ~/.hermes/profiles/ops/SOUL.md different agent, different soul, different memory, same machine. full breakdown of Hermes as a Personal AI OS in the article 👇

YanXbt

28,924 views • 2 months ago

andrej karpathy spent two hours teaching one thing: tokens are the atom of llms. tokenization is at the heart of every llm weirdness you've ever debugged. [watch the 15-min clip below. then run the 7-day playbook] ↓ save this before everyone copies it learn how the tokenizer works. understand how your llm actually consumes input. then run the engineering roadmap that took one production agent from $4,800/mo to $620/mo in 7 days. 87% reduction. no model swap. no framework migration. no quality drop on the eval set. token cost in 2026 is an engineering discipline. every line of your system prompt is rent you pay forever. what was eating the budget: → a single forgotten cron job ate 47% of one team's bill. they turned it off on a tuesday and the bill dropped before they wrote any optimization code. → anthropic ships a 90% discount on cache reads. one config line, cache_control ephemeral, break-even after one hit. most teams cache the volatile parts of the prompt and watch their hit rate sit at 12%. → one production agent went from 14,500 tokens of context overhead per turn to 850. a 94% drop. output quality held within 2% of the uncompressed baseline. → 60% of agent calls are haiku-tier work running on opus rates. classify the task first. pick the model second. → retry loops are the silent killer. no MAX_STEPS bound, one bad search query, $14 burned in a single session. one team traced 38% of their bill to this single pattern. karpathy gave you the atom. the playbook below gives you the harness. watch the lecture. read the playbook ↓

andrej karpathy spent two hours teaching one thing: tokens are the atom of llms. tokenization is at the heart of every llm weirdness you've ever debugged. [watch the 15-min clip below. then run the 7-day playbook] ↓ save this before everyone copies it learn how the tokenizer works. understand how your llm actually consumes input. then run the engineering roadmap that took one production agent from $4,800/mo to $620/mo in 7 days. 87% reduction. no model swap. no framework migration. no quality drop on the eval set. token cost in 2026 is an engineering discipline. every line of your system prompt is rent you pay forever. what was eating the budget: → a single forgotten cron job ate 47% of one team's bill. they turned it off on a tuesday and the bill dropped before they wrote any optimization code. → anthropic ships a 90% discount on cache reads. one config line, cache_control ephemeral, break-even after one hit. most teams cache the volatile parts of the prompt and watch their hit rate sit at 12%. → one production agent went from 14,500 tokens of context overhead per turn to 850. a 94% drop. output quality held within 2% of the uncompressed baseline. → 60% of agent calls are haiku-tier work running on opus rates. classify the task first. pick the model second. → retry loops are the silent killer. no MAX_STEPS bound, one bad search query, $14 burned in a single session. one team traced 38% of their bill to this single pattern. karpathy gave you the atom. the playbook below gives you the harness. watch the lecture. read the playbook ↓

Rohit

73,447 views • 3 months ago

MIT PhD student Alex Zhang reveals the scaling result where a model trained on short tasks generalizes to problems 100x longer for free: "If you're very clever about the design of your harness or how you use the language model, you can almost get scaling gains for free." "If you train a model naively, there's no tricks. It's just the same way you train a model on these RL environments. You just roll it out, and then you just get some reward." "If you train it on only short tasks, like only tasks that are 10,000 tokens long, and then you were to run it on a similar domain, but at a million tokens, or 10 million tokens, or 100,000 tokens, it generalizes really, really well. If you look at it compared to even the base transformer, you get way better generalization properties." "When the model uses an RLM (Recursive Language Model) after it's trained on these short tasks, it will see some kind of trajectory of actions that it does. Between these two problems of different lengths, the RLM learns to see them as almost the same problem." "Token for token, they're almost the same. You can describe it in code. In one code setting, maybe the for loop is a little bigger, but it's the same kind of code and it derives the constants from the data. There's no hard coding, so they literally look the same." alex zhang

MIT PhD student Alex Zhang reveals the scaling result where a model trained on short tasks generalizes to problems 100x longer for free: "If you're very clever about the design of your harness or how you use the language model, you can almost get scaling gains for free." "If you train a model naively, there's no tricks. It's just the same way you train a model on these RL environments. You just roll it out, and then you just get some reward." "If you train it on only short tasks, like only tasks that are 10,000 tokens long, and then you were to run it on a similar domain, but at a million tokens, or 10 million tokens, or 100,000 tokens, it generalizes really, really well. If you look at it compared to even the base transformer, you get way better generalization properties." "When the model uses an RLM (Recursive Language Model) after it's trained on these short tasks, it will see some kind of trajectory of actions that it does. Between these two problems of different lengths, the RLM learns to see them as almost the same problem." "Token for token, they're almost the same. You can describe it in code. In one code setting, maybe the for loop is a little bigger, but it's the same kind of code and it derives the constants from the data. There's no hard coding, so they literally look the same." alex zhang

MTS

100,519 views • 1 month ago

People think Matt faked it when he said one prompt built a whole cod game. So i read the prompt properly and ran it myself. Spent over 30% of my weekly tokens on the 20x plan for 12 hours. It's real. but here's the part nobody's talking about. The prompt says don't stop until you beat call of duty in a blind side by side. It never beat it. It kept making the scores, and judges the game every single round. it failed its own test the entire time. That's why the game got better and better . An impossible high bar means it never stops improving.

People think Matt faked it when he said one prompt built a whole cod game. So i read the prompt properly and ran it myself. Spent over 30% of my weekly tokens on the 20x plan for 12 hours. It's real. but here's the part nobody's talking about. The prompt says don't stop until you beat call of duty in a blind side by side. It never beat it. It kept making the scores, and judges the game every single round. it failed its own test the entire time. That's why the game got better and better . An impossible high bar means it never stops improving.

Ziwen

86,753 views • 1 month ago

Chris Camillo reacts to Grant Cardone saying you can make $1,000,000/year as an AI consultant “I said the same thing on your show and got 4,000,000 views in two days on X.” “He stole it from me. I think he stole it from me. Because what I said was that you can make $500,000 a year basically doing exactly that. I said you go into a company and you say, 'Give me one problem that you have.' And the problem could be as simple as, 'We're missing phone calls every night and because we're missing those phone calls because we close at five or six, they call one of our competitors.' Create an AI widget to answer those calls, figure out what the issue is, immediately send responses out, inform the entire company, so the next morning you already have a lead sheet, you're already engaging with that customer, and you never lose that customer. You could potentially increase a company's revenue by 10 or 15%. Small businesses, HVAC companies, plumbing companies, doesn't matter what it is. You can just go company to company and say, 'I'm going to fix simple problems like that for you.' Put them on a monthly retainer and make half a million dollars a year just having"

Chris Camillo reacts to Grant Cardone saying you can make $1,000,000/year as an AI consultant “I said the same thing on your show and got 4,000,000 views in two days on X.” “He stole it from me. I think he stole it from me. Because what I said was that you can make $500,000 a year basically doing exactly that. I said you go into a company and you say, 'Give me one problem that you have.' And the problem could be as simple as, 'We're missing phone calls every night and because we're missing those phone calls because we close at five or six, they call one of our competitors.' Create an AI widget to answer those calls, figure out what the issue is, immediately send responses out, inform the entire company, so the next morning you already have a lead sheet, you're already engaging with that customer, and you never lose that customer. You could potentially increase a company's revenue by 10 or 15%. Small businesses, HVAC companies, plumbing companies, doesn't matter what it is. You can just go company to company and say, 'I'm going to fix simple problems like that for you.' Put them on a monthly retainer and make half a million dollars a year just having"

The Iced Coffee Hour

58,940 views • 3 months ago