Get live statistics and analysis of Vihan Singh's profile on X / Twitter

I revisit this Sam Altman post nearly every month:

Getting more and more tradable systems designed by LLMs. This is what I’ve learned actually works for me. 👇 Step-by-step breakdown of the process that turns AI ideas into real systems I can paper trade. 1/ I start in ChatGPT (o3 model). I use a custom bio where I describe myself as an experienced trader—this primes it into “expert mode.” Then I share a picture from a real trading session and describe the context. In this case: intraday breakout from a volatility “squeeze.” I explain the type of data I have + what kind of realistic performance I expect (after slippage & fees). Then I ask GPT to brainstorm and rank strategies by: - difficulty to implement - likelihood of working live 2/ That prompt usually delivers surprisingly good ideas. I pick the one that makes the most sense intuitively. Then I ask GPT to: - write a full, detailed trading plan - create step-by-step implementation instructions for a programmer to code it into a backtester 3/ These instructions go into Claude Code (my coding agent). Claude takes the logic and turns it into Python. I run the code, but only on strict In-Sample data. This first version is rarely perfect—but that’s okay. 4/ Then I ask Claude to: - show screenshots of the last 10 trades - plot all entries & exits I use prompts like “think deeply” or “audit the logic for bias.” After some iterations, I ask Claude to summarize the implemented logic in plain English. This is key for the next step. 5/ I paste the summarized logic back into ChatGPT. GPT often spots errors, edge cases, or logical holes. Its feedback then goes back to Claude for code fixes. 6/ Once we’re closer to a clean build, I run a full audit: - Entries - Exits - Stops - Position sizing - Indicators Claude checks each piece individually, not all at once. Why? LLMs are more accurate on small, focused tasks. 7/ When code is solid, I test Out-of-Sample and on different markets. The equity curve I shared is from QQQ on full history - The system was created only on a few years of SPY. That’s an important check against overfitting. 8/ The whole process took me nearly a day. Not “automagic,” but powerful. Yes, I might still find issues in paper trading—that’s part of the game. But every iteration trains me (and the LLM) what to look for next time. 9/ Is this system overfit? I don’t think so. The core logic is dead simple—but I probably wouldn’t have thought of it myself. Core idea: - Detects tight ranges (volatility “squeeze”) - Trades breakouts with stop, profit target, and EOD exit Clean, robust, and intuitive. 10/ Conclusion? This workflow is absolutely worth trying. But treat LLMs as junior analysts, not black-box gods. Use critical thinking. Stress-test every idea. The more experience you bring, the more value you’ll get from it.

I’ve saved hundreds of AI tool threads this year. Most are noise. But this one from a16z hits differently. Here's why: → It’s organized by use case, not just hype → It includes tools you’ve never heard of, but should → It goes beyond “chatbots and image gen” and into parenting, therapy, workflow agents, voice input, and learning tools This feels less like a thread and more like a snapshot of where AI actually fits into daily life. Share this with someone starting their first AI side project.

An old parable: "The most important part of the kingdom is the king, the most important part of the king is the king's head, the most important part of the king's head is the king's vision." "If the vision goes, the head goes. If the head goes, the king goes. If the king goes, the kingdom goes." "But if the vision remains, then nothing is lost." Therefore the vision is the kingdom.

Dear algo, please only show this tweet to people with $0 MRR.

Profiles like this are getting $100M offers from Meta. Let me show you the EXACT career path that’s worth more than most IPOs: The Journey: • 2010-2012: Intern at Microsoft Research Asia (2 years) • 2014: 4-month research intern at Microsoft • 2017: PhD from MIT in Computer Science • 2017-2023: Staff Research Scientist at Google DeepMind (6+ years) • 2023-Present: Member of Technical Staff at OpenAI Translation: This person has been building the AI systems that are reshaping civilization. What makes someone worth $100M? 🔥 Deep expertise in foundational AI research 🔥 Experience at the 3 most important AI labs on Earth 🔥 PhD from the world’s top CS program 🔥 6+ years at DeepMind during the breakthrough era 🔥 Currently at OpenAI during the ChatGPT revolution The brutal reality: While most people argue about whether AI will take jobs, this person IS the AI revolution. Career lessons hidden in plain sight: ❌ “Follow your passion” → This person followed the hardest problems ❌ “Work-life balance” → This person worked at the epicenter of history ❌ “Job security” → This person built skills so rare they’re worth $100M ❌ “Climb the ladder”* → This person jumped between the world’s best labs The $100M reality: - Total compensation packages: Up to $100M over 4 years - Annual value: $25M+ per year for top talent - Base salaries: $2M+ confirmed - Stock/equity: The majority of the package For context: This is more than most Fortune 500 CEOs make. More than most hedge fund managers. More than most professional athletes. The uncomfortable truth: AI researchers are becoming the new athletes. The top 50 people in AI are worth more than entire companies. What this person probably knows: • How large language models actually work • The secret sauce behind ChatGPT/GPT-4 • Google’s next-generation AI strategies • The technical details that determine AI’s future Market reality: There are maybe 500 people on Earth with this specific combination of skills. Meta is offering compensation packages worth up to $100M to poach OpenAI’s top researchers. The new career math: Traditional path: 40 years × $200K = $8M lifetime earnings AI research path: 4 years × $25M = $100M Questions for your career: • Are you building skills that can’t be replicated? • Are you working on problems that matter? • Are you positioning yourself at the center of major shifts? The bigger picture: We’re watching the birth of a new aristocracy. The people who understand AI will own the future. Everyone else will rent it from them. My prediction: In 10 years, this salary will look cheap. Because AI researchers aren’t just employees. They’re the architects of the next economy.

Trading easy. Making money from trading is the hard part.

This was the email the Ghost of Silicon Valley (Soham) used to land dozens of engineering jobs & work them simultaneously to earn 7 figures. I was curious about this story so I had to dig further. Let me preface... I don't condone this behavior.. ..but ethics aside, this email slaps. Here's why: - quick tldr that gets to the point, pulling you in to read more - startups want cracked engineers, this delivers an 'obsessive focus' tone - establishes credible authority through notable past startups - details the outcomes of his work, not buzzword skills - conveys a generalist vs. a specialist range of experience - no pushy ask, just communicates desire to join the team Thoughts on this story? I think it’s wild.

From what I can see, there are 3 key dogmas around AGI. 1) the belief that continuously scaling up the models will lead to AGI without need for new architecture (dario/sam). 2) the belief that we need spatial intelligence for AGI and thus new architecture (fei fei). 3) the belief that we need data efficiency and learning on the fly for AGI and thus new architecture. ML + discrete program search measured by ARC benchmarks. (Francois Choulet).

Some random team in San Francisco has probably already built what you’re working on. That’s why now more than ever: Distribution > Product (exceptions: biotech, hardware, enterprise SaaS) I’ve seen startups with average products do ridiculous numbers—just because they show up online. They post. They tell stories. The truth is that most people scrolling don’t even care about the product. They care about the people behind it. If you can get them interested in you, you can sell. Some version of your product has probably already been done—or is being worked on right now by another team halfway across the country. Your product might be gold, but in 2025, attention is the currency. If you don’t have it, you’re broke.

Given what we know about the JS trade now, you can understand why most short gamma option traders, especially those that had trend following bias, were struggling last 1-2 years. With JS's market manipulation being a mean reverting trade (lift prices up then suddenly push them down), most if not all trend / breakout trades would be slaughtered. Take the simple 920 straddle with X% SL as an example. This trade is a short vol bet with a directional breakout trigger embedded (as the SL). In the first half of the market when underlying prices rise, the CE leg would get sqd off, creating a long delta position (short PE). This leg would then get blazed in the second half when underlying prices are hammered. Moreover, given the size concentration in such trades across retail traders, prop desks, etc., a liquidation cascade would further exacerbate the down move. tldr; midfreq short gamma + trend trades do fine if the underlying moves nowhere, just up, or just down but get fucked when after going one way the market sharply reverses which was JS signature.

wild take: JS had a role to play in the new Trump tariffs

Modafinil ~ NZT-48 !! not medical advice !!

I revisit this Sam Altman post nearly every month:

wild take: JS had a role to play in the new Trump tariffs

Modafinil ~ NZT-48 !! not medical advice !!

I’ve saved hundreds of AI tool threads this year. Most are noise. But this one from a16z hits differently. Here's why: → It’s organized by use case, not just hype → It includes tools you’ve never heard of, but should → It goes beyond “chatbots and image gen” and into parenting, therapy, workflow agents, voice input, and learning tools This feels less like a thread and more like a snapshot of where AI actually fits into daily life. Share this with someone starting their first AI side project.

Profiles like this are getting $100M offers from Meta. Let me show you the EXACT career path that’s worth more than most IPOs: The Journey: • 2010-2012: Intern at Microsoft Research Asia (2 years) • 2014: 4-month research intern at Microsoft • 2017: PhD from MIT in Computer Science • 2017-2023: Staff Research Scientist at Google DeepMind (6+ years) • 2023-Present: Member of Technical Staff at OpenAI Translation: This person has been building the AI systems that are reshaping civilization. What makes someone worth $100M? 🔥 Deep expertise in foundational AI research 🔥 Experience at the 3 most important AI labs on Earth 🔥 PhD from the world’s top CS program 🔥 6+ years at DeepMind during the breakthrough era 🔥 Currently at OpenAI during the ChatGPT revolution The brutal reality: While most people argue about whether AI will take jobs, this person IS the AI revolution. Career lessons hidden in plain sight: ❌ “Follow your passion” → This person followed the hardest problems ❌ “Work-life balance” → This person worked at the epicenter of history ❌ “Job security” → This person built skills so rare they’re worth $100M ❌ “Climb the ladder”* → This person jumped between the world’s best labs The $100M reality: - Total compensation packages: Up to $100M over 4 years - Annual value: $25M+ per year for top talent - Base salaries: $2M+ confirmed - Stock/equity: The majority of the package For context: This is more than most Fortune 500 CEOs make. More than most hedge fund managers. More than most professional athletes. The uncomfortable truth: AI researchers are becoming the new athletes. The top 50 people in AI are worth more than entire companies. What this person probably knows: • How large language models actually work • The secret sauce behind ChatGPT/GPT-4 • Google’s next-generation AI strategies • The technical details that determine AI’s future Market reality: There are maybe 500 people on Earth with this specific combination of skills. Meta is offering compensation packages worth up to $100M to poach OpenAI’s top researchers. The new career math: Traditional path: 40 years × $200K = $8M lifetime earnings AI research path: 4 years × $25M = $100M Questions for your career: • Are you building skills that can’t be replicated? • Are you working on problems that matter? • Are you positioning yourself at the center of major shifts? The bigger picture: We’re watching the birth of a new aristocracy. The people who understand AI will own the future. Everyone else will rent it from them. My prediction: In 10 years, this salary will look cheap. Because AI researchers aren’t just employees. They’re the architects of the next economy.

Getting more and more tradable systems designed by LLMs. This is what I’ve learned actually works for me. 👇 Step-by-step breakdown of the process that turns AI ideas into real systems I can paper trade. 1/ I start in ChatGPT (o3 model). I use a custom bio where I describe myself as an experienced trader—this primes it into “expert mode.” Then I share a picture from a real trading session and describe the context. In this case: intraday breakout from a volatility “squeeze.” I explain the type of data I have + what kind of realistic performance I expect (after slippage & fees). Then I ask GPT to brainstorm and rank strategies by: - difficulty to implement - likelihood of working live 2/ That prompt usually delivers surprisingly good ideas. I pick the one that makes the most sense intuitively. Then I ask GPT to: - write a full, detailed trading plan - create step-by-step implementation instructions for a programmer to code it into a backtester 3/ These instructions go into Claude Code (my coding agent). Claude takes the logic and turns it into Python. I run the code, but only on strict In-Sample data. This first version is rarely perfect—but that’s okay. 4/ Then I ask Claude to: - show screenshots of the last 10 trades - plot all entries & exits I use prompts like “think deeply” or “audit the logic for bias.” After some iterations, I ask Claude to summarize the implemented logic in plain English. This is key for the next step. 5/ I paste the summarized logic back into ChatGPT. GPT often spots errors, edge cases, or logical holes. Its feedback then goes back to Claude for code fixes. 6/ Once we’re closer to a clean build, I run a full audit: - Entries - Exits - Stops - Position sizing - Indicators Claude checks each piece individually, not all at once. Why? LLMs are more accurate on small, focused tasks. 7/ When code is solid, I test Out-of-Sample and on different markets. The equity curve I shared is from QQQ on full history - The system was created only on a few years of SPY. That’s an important check against overfitting. 8/ The whole process took me nearly a day. Not “automagic,” but powerful. Yes, I might still find issues in paper trading—that’s part of the game. But every iteration trains me (and the LLM) what to look for next time. 9/ Is this system overfit? I don’t think so. The core logic is dead simple—but I probably wouldn’t have thought of it myself. Core idea: - Detects tight ranges (volatility “squeeze”) - Trades breakouts with stop, profit target, and EOD exit Clean, robust, and intuitive. 10/ Conclusion? This workflow is absolutely worth trying. But treat LLMs as junior analysts, not black-box gods. Use critical thinking. Stress-test every idea. The more experience you bring, the more value you’ll get from it.

Hot Take Ta Ra Rum Pum >>> F1

The 10 principles to build an AI Company Capital: 1. This is the biggest opportunity since the internet. Bigger than cloud. Faster than mobile. 2. AI isn’t one market — it’s eating all markets. Software, services, labor, infrastructure. 3. The value is in the application layer, not the models. Tools people actually use will win. 4. Foundation models are moving up the stack. Compete by going narrow and deep, not wide. 5. The best products start customer-back, not model-forward. Solve one job completely. 6. This is the Agent Era. AI will shift from chatbots to swarms of agents executing work. 7. Speed is everything. There’s a vacuum in every vertical. Whoever ships first, wins. 8. Distribution is built-in now. You’re launching into a 5.6B-person internet. If it works, it explodes. 9. The model doesn’t matter unless the product is great. Startups win on UX, not architecture. 10. Your moat is your workflow. Own the end-to-end journey. Don’t build tools, build outcomes.

Never have feedback loops felt this snappy. I was exploring an AI-therapy app this morning and honestly was feeling quite luke warm about the experience. At one point, the “Leave feedback” box popped up. I wrote, “It’s okay, but you’re not making me dig deep enough.” Interestingly, the app took my feedback immediately and suggested a topic we could dig into. It came up with the topic on its own, gleaning from our previous chats so it resonated and I took the bait. Within minutes we were peeling back layers, my personal NPS for the app shot up, and I caught myself grinning at a bot. I went from luke-warm to being a “fan” within minutes! That’s when it hit me: this was the fastest feedback loop I’ve ever lived. Forget the old routine – collect reviews, list them in a spreadsheet, prioritize and add to the product roadmap. No, not anymore! We’re stepping into an era where products read your note, reroute the experience, and make themselves better before you’ve even left the screen. Wild. The future of product building is going to be very different! And only the ones that fully embrace an AI-first experience will last.

An old parable: "The most important part of the kingdom is the king, the most important part of the king is the king's head, the most important part of the king's head is the king's vision." "If the vision goes, the head goes. If the head goes, the king goes. If the king goes, the kingdom goes." "But if the vision remains, then nothing is lost." Therefore the vision is the kingdom.

Trading easy. Making money from trading is the hard part.