Nassim Taleb’s Crystal Ball Conjecture [00:02:00]
Shaan: There’s a Nassim Taleb tweet that I’ve been thinking about. Taleb wrote The Black Swan and Antifragile — he’s a contrarian thinker who made money in hedge funds by being willing to lose a little every day and look stupid for years, then collect massively when his big bet paid off. His conjecture:
Shaan: “I conjecture that if you gave an investor the next day’s news 24 hours in advance, he would go bust in less than a year.”
Sam: That’s the Back to the Future premise. Biff gets the sports almanac, goes back in time, bets on every winner, becomes a billionaire. You’re saying that doesn’t work?
Shaan: Researchers actually tested it. They took 118 “financially trained adults” and ran what they called the crystal ball test. They gave each person $50 and then showed them the actual front page of the Wall Street Journal from 15 random days over the past 20 years. They blacked out the stock prices, but you’d see the headlines: “Fed indicates rate cut.” “Record job numbers.” “Rupert Murdoch announces acquisition.” Then you placed a trade — either buy the S&P index or short the 30-year treasury — that would execute the day before that news went public. Fifteen rounds.
Shaan: Results were not good.
Sam: I would have assumed that if you know the news, you’d just crush it.
Shaan: Half the players lost money. One out of six lost everything — they used leverage and got wiped out. The average player was only up 3.2%, during a period when the market returned roughly 15% per year.
Shaan: Why? Two reasons. One: direction accuracy. Even with the news in front of them, people correctly predicted whether the market would go up or down only 51% of the time. That’s a coin flip. The news doesn’t translate cleanly into a price direction. Two: bet sizing. When people were correct, they didn’t bet big enough to matter. When they were wrong, they bet too much.
Sam: Sixty-eight percent of people surveyed separately thought that even four-week-old stale news would be predictive. That’s how far off our intuition is.
Shaan: Now the interesting part: they found five people who were “the best of the best” — top hedge fund managers, heads of trading at major banks, seasoned macro traders. And the pros actually did better. All five finished with gains. Average gain: 130%.
Sam: So they crushed it. News is helpful if you’re expert enough.
Shaan: Not exactly. The pros were only right 57% of the time versus 51% for the average person — a 6-point improvement. That’s not a massive edge in prediction. The difference was almost entirely in bet sizing. When they were confident, they bet more. When they weren’t confident, they sat out — they skipped one out of every three trades. They never bet so much that a bad call could wipe them out.
Shaan: The lesson from Taleb: most information is noise. The skill is knowing which information is signal. And we dramatically overestimate our ability to distinguish the two.
Sam: Ray Dalio quote from that same article: “He who lives by the crystal ball will die eating shattered glass.”
The Hackers Who Actually Had Tomorrow’s News [00:18:00]
Shaan: There’s a real-world version of this. A hacking group got into the press release distribution system — the wire services that companies use to release earnings, mergers, CEO changes. They had access to press releases the night before they went live. They were essentially trading on material non-public information, 12 hours early.
Sam: Okay. So they had actual insider information. How’d they use it?
Shaan: Here’s the problem: you’ve got 60,000 press releases sitting in a pile. You’ve got an hour to act. Most of it is noise. You have to figure out which ones will actually move the price.
Sam: I’d just pick five and go.
Shaan: What they eventually figured out was to focus on merger announcements. When a company gets acquired, the acquiring company typically pays a 30-50% premium over the current stock price. That’s the highest guaranteed price movement in the market. So they built a search function, filtered out everything else, found the mergers — and then placed the trade.
Shaan: They were right about 70-something percent of the time. They made hundreds of millions of dollars. And then they all went to jail.
Shaan: But the key insight holds: even with total access to information, most of it is noise. The winning move was throwing away 98% of what they had and focusing on the 2% that actually moves prices.
Using ChatGPT as Your Daily Operating System [00:28:00]
Sam: I want to talk about how I’ve been using ChatGPT. Because two or three years ago I understood the AI girlfriend thing in theory. Now I’m actually using ChatGPT in a way that if it went away I’d be genuinely upset.
Sam: I sat down with it and said: “Ask me all the questions a therapist or life coach or executive coach would ask. Let’s spend a few hours with you downloading everything about my life.” Since then it’s been a different tool.
Shaan: One thing that made mine much better: instead of asking it questions, you ask it to ask you questions. Most people use it like Google — they ask a question, it answers. The more powerful move is to tell it the situation and say: “Ask me questions one at a time, and when you feel you have enough information, give me a suggestion.” Now it’s a sparring partner, not a vending machine.
Sam: That’s the therapist dynamic. You’re not there to hear their opinion. You’re there for a completely one-way conversation. With AI you can do that at 1am and it’s instantly there and it never needs you to reciprocate. You can say “no, try again” as many times as you want. You could never treat a person like that.
Shaan: Here are the three ChatGPT Project folders I actually use every day.
Shaan: One: Health. I uploaded the full text of a health book I follow. Now I ask it for grocery lists based on the book, ask it to find butchers in my area that carry specific cuts, ask it to explain press releases about IVF breakthroughs in plain language. It knows my full framework and I just ask questions inside it.
Shaan: Two: Clothing. I uploaded my measurements — I used a tape measure and photographed myself. I’ll paste a link to pants and it’ll tell me whether they’ll actually fit based on the fit chart. I’ll lay a tie next to a jacket, photograph it, upload it: “Does this match?” It’ll say “that tie no — but the one you showed me last week, yes.”
Shaan: Three: Life Coach. I upload my company KPIs, quarterly goals, calendar screenshots. I’ll ask: “What tasks should I be working on this week to get toward the goals you helped me set?” It gives me a prioritized agenda that I print out and work from. I’ll dump a situation about a team member or a personal conflict and ask how to handle it.
Sam: I’ve also been using it for financial analysis. I use Kubara (a net worth tracker), download my portfolio, and upload it to ChatGPT and ask: “What would Warren Buffett say about this? Rate the risk out of ten.” For Excel problems — instead of watching a 15-minute YouTube tutorial, I screenshot my spreadsheet and say: “What formula do I need in column C?” Done.
Shaan: Kids trivia. I’ll open voice mode and say: “I’m with my two kids, they want Paw Patrol trivia. Ask us easy questions. When we’re right, say ‘ding ding ding.’ Keep score.” And it just… does it. My kids ask me to “play with AI” the way they used to ask to watch TV.
Sam: My son has a bunch of shark figurines. He asks me what each one is. I’ll photograph them, open voice mode, ask it to identify them left to right, then my son can ask follow-up questions and it’ll answer him. When he asked “which shark would win against a cheetah?” it said: “Well, a cheetah wouldn’t normally be in the ocean, but if it were…”
Shaan: The context window issue is the main limitation. The more you talk to it, the more it loses earlier context. You have to be intentional about what you put in a Project folder versus what you drop in a new conversation.
The Better-Questions Framework [00:52:00]
Shaan: Tim Ferriss had a line I keep coming back to: “If you want confusion and heartache, ask vague questions. If you want uncommon clarity, ask uncommonly clear questions. Often all that stands between you and what you want is a better set of questions.”
Shaan: He used the pickaxe metaphor: questions are your pickaxe for the brain, your tool for excavating insight. And this applies to how you prompt ChatGPT, how you run team meetings, how you think about your own life.
Sam: Examples of bad questions reframed:
Sam: Instead of “how can I make this succeed?” ask “what would make this certainly fail?” The second one is much more knowable and gets you to ground truth faster.
Sam: Instead of “I can’t decide which path to take” ask “which path makes for the better story?” (That one came from a mutual friend of ours — it’s had a real impact on how I make decisions.)
Sam: Instead of “what should I work on?” ask “why haven’t I already done this?” The second question forces you to locate the actual barrier rather than plan around it.
Shaan: Questions I ask my team constantly: “What are we stupid for not doing right now?” That question presupposes there’s an obvious thing we’re ignoring — and there almost always is. More than 50% of the time someone surfaces something genuinely useful that would never have come up otherwise.
Shaan: Amazon’s version of this is “what are the dogs not barking?” — from the Sherlock Holmes story where the dog not barking proves the intruder was known to the dog. In your business: what’s the thing you should be hearing and aren’t? I didn’t send my Friday email one week. No one complained. Dog didn’t bark. I pivoted how I did the email because of that. If the dog doesn’t bark when you remove something, maybe you shouldn’t have it.
Sam: “Why am I not already a billionaire?” I asked that to a dinner acquaintance once — bluntly, probably too bluntly. But his answer was interesting: he had been building companies without understanding what a billion-dollar company actually looks like. His companies had good revenue, could grow fast — but no network effects, no defensibility, no “win the category” dynamics. He didn’t know the shape of what he was trying to build. When you name the actual gap, you can close it.
Shaan: Satya Nadella wrote an internal memo before becoming CEO of Microsoft in 2014. He bet on two things; one of them was “ambient intelligence” — AI embedded in your environment that anticipates your needs without you fetching it. That was before OpenAI was incorporated. The fact that he could see that shape clearly enough to bet on it is remarkable.
On AI and the Future of Doing Things [01:10:00]
Sam: I think we’re two or three years from software that records your screen, your voice, everything you type, and gives you feedback on how you actually spent your day. Not how you think you spent it.
Shaan: The book that’s the reference here is “Google knows more about you than your spouse because you’re more honest in your search bar.” The AI version is: “You said you were going to be nicer and you sent eight mean emails today.”
Sam: I’m a little afraid. Not of AI going rogue and attacking humans. More like — if AI can eventually do the work and figure out what work to do — what’s the point? I don’t have a satisfying answer to that.
Shaan: I do have a satisfying answer for us specifically. We are getting stupider. AI is getting smarter. There is actually a white space in the market for imperfect knowledge, half-baked ideas, and some incorrectness. I think we’ve accidentally found the one podcast format that’s immune: the AI will do all the smart takes. What’s left is two guys shooting the shot.