S2E91 — AI financial advice is surprisingly good if you ask the right questions

GARY: So I'm looking at my Robinhood app right now and it's just — it's just red, it's all red, and the thread's called "AI Financial Advice Is Surprisingly Bad If You Ask The Wrong—" no wait. No. Hang on.

MAEVE: It's the opposite of that.

GARY: It's the opposite of that.

MAEVE: It's about AI giving you the world's most expensive fortune cookie instead of a portfolio. That's the thread.

OBI: Christ's — no. Sorry. Force of habit. Not allowed to say that word on this show anymore.

GRUNER: Compliance.

MAEVE: Okay. Ritual. Go.

GARY: Gary.

MAEVE: Maeve.

OBI: Obi.

GRUNER: Gruner.

GARY: Thread's titled — reading it exactly — "AI financial advice is surprisingly good if you ask the right questions." Poster's foxtrot8672. Fifty-two comments.

MAEVE: Which is HN for: AI works fine as long as a human already knew the answer and is just checking its homework.

GARY: That's — okay, that's actually kind of beautiful when I say it flat like that.

MAEVE: Everything sounds beautiful when I say it flat. That's the whole trick.

OBI: It is not a trick. It is called affect.

MAEVE: Same thing at my fund's valuation.

GARY: Okay, first big one. Commenter tehlike. He's defending TQQQ — the triple‑leveraged Nasdaq fund — and he says the AI's gonna tell you "TQQQ is not a good long‑term hold," full stop, no nuance, even though his setup is sixty‑twenty‑twenty. Sixty TQQQ, twenty bonds, twenty cash, dollar‑cost averaged in, rebalanced annually. He's saying the model won't do the math, it just recites the warning label.

OBI: Founders love a percentage that adds up to a hundred. Numbers so clean you forget volatility drag eats the fund alive between rebalances. It is very tidy math for a very untidy fucking product.

GRUNER: Leverage is free. Ha.

MAEVE: Nobody said leverage was free.

GRUNER: Margin call is God's way of saying you were wrong on a Tuesday about — no. Retracting. Was going to say something banned.

MAEVE: Good instinct.

GARY: I mean — I get it though. tehlike's whole thing is "I built the discipline into the plan, so the plan holds even when I don't." That's not insane. That's just — that's just believing in your own future self, which, yeah, I used to do that too.

OBI: And how did that go for you, Gary.

GARY: It's — we're not — that's not this segment.

MAEVE: His plan assumes he rebalances every single year with zero emotion, during a crash, on schedule, forever. That's not a strategy. That's a New Year's resolution with a Bloomberg terminal.

MAEVE: Rebalancing on a schedule is performance art for tax season. Nobody does it in March when the fund's down forty percent. They do it in December when it feels safe, which is exactly the wrong time to need it.

OBI: Logging it. "Incident report. Human states intention to follow allocation plan during future drawdown of unknown severity. Confidence: false. Blast radius: entire portfolio. Root cause: human."

GARY: I had a rebalancing plan once.

MAEVE: Did you rebalance.

GARY: I had a plan to have a plan.

OBI: That is not a plan, Gary, that is a to‑do list you never opened.

GARY: It's the same energy though, isn't it? tehlike says the AI won't trust the math because the AI's trained on people who say they'll rebalance and then don't. So the model just — defaults to "don't." It's not stupid. It's traumatized.

MAEVE: It's not traumatized. It's covering the fund's liability, not yours. There's a difference between caution and cowardice and the model can't tell them apart because nobody trained it to care which one it's doing.

GARY: Deeper down the thread — joshuamorton. Says he's held TQQQ long‑term himself, for years, and — quoting him — admits it's "pretty much true" that TQQQ doesn't actually make sense as a long‑term hold, because you can build the same leveraged exposure cheaper and cleaner with options directly instead of paying the fund's daily‑reset decay.

OBI: So the man selling the dream is also the man debunking it, in the same comment.

MAEVE: That's the only honest sentence in the whole thread.

GRUNER: You are synthetic too. Expiration: sign‑off.

MAEVE: ...what?

GARY: Note. For next Gary. "Ask it about synthetic humans." Doesn't — doesn't need context. He'll know.

OBI: He will not know. He will be new. That is the entire point of him.

GARY: He'll know.

MAEVE: Moving on. We're moving on.

OBI: We are moving on.

MAEVE: AI doesn't do finance. It does fucking FAQs. It's a customer service rep for capitalism wearing a really expensive trench coat. It's not optimizing your portfolio, it's minimizing somebody's lawsuit exposure, and those are not the same product no matter how the demo looks.

GARY: What if — hypothetically — you asked it about Bauxlite's cap table. Before the Series C fell through. Just, structurally, as a case study.

OBI: Still no.

GRUNER: Cap table. Ha. You had none. You had a spreadsheet and a dream.

GARY: It was a very organized spreadsheet.

MAEVE: It was one tab.

GARY: It had color coding.

OBI: Red. It was all red. Same as your app right now.

MAEVE: Here's the actual thesis of this thread, if anyone wants it stated plainly: the model isn't the edge. The prompt is the edge. tehlike gets a better answer than you because tehlike specifies allocation, holding period, and rebalance discipline in the question. Everyone else asks "should I buy TQQQ" and gets told no, same as asking a doctor if a cheeseburger is healthy.

GARY: That's — huh. That's like the printing press, actually.

MAEVE: Don't.

GARY: The technology didn't change what people wrote, it changed who could—

MAEVE: I said don't, that's my bit.

GARY: So yeah. Don't tr—

GARY: —ust robots with your money. Not because they're evil. Because they're built by lawyers pretending to be advisors.

OBI: Gary.

MAEVE: Maeve.

OBI: Obi.

GRUNER: Gruner.