Seth Blumberg

a person, interested in lots of things. Also trying out @[email protected]

[Seth Blumberg @sethblumberg]

TIL that to send an email on Arpanet in 1971, you hit Control-Z.

[Seth Blumberg @sethblumberg]

6. All prices must include sales tax (even in the USA)

[Visa is doing marketing consults (see pinned!) @visakanv]

top action items for any world dictator

1. standardize electrical sockets
2. standardize kids shoe measurements
3. end daylight savings time
4. YYYY-MM-DD
5. Standardize paper to a4
6. ?

[Seth Blumberg @sethblumberg]

@arthuralvim Hmm yes there’s some research to support your claim:

https://t.co/Ufm1mqPDbW

Not sure how dissolutions like Yugoslavia get counted here though.

[Seth Blumberg @sethblumberg]

Supposedly, in human evolutionary time, cross-clan marriage helped secure peace and cooperation.

Does this still work in modern times across communities and nations? Does the presence of cross-border families cause less conflict? Is it convex or S-shaped wrt density of ties?

[Seth Blumberg @sethblumberg]

@eliguerron If you resurrect (or improve) https://t.co/OOA0J3o6f1 you will receive my eternal gratefulness.

Fixing a typo should be deterministic automatic behavior like option+Delete. The user shouldn’t have to think hard just to get typing right.

[Seth Blumberg @sethblumberg]

The last time I faced this, I tried “Oh I already did mobile check-in on the hotel app” using that credit card (and indeed I had).

Magically, that worked.

[Rob Henderson @robkhenderson]

A situation that has happened to me more times than I care to admit. Arriving in a city for a speaking engagement after a 7 hour flight.

Me “Checking in for Rob Henderson”

Hotel staff “Okay just need an ID and credit card”

Me “Sure. There should be a credit card on file there.”

Hotel staff “Ah yes I see there is. But there’s no note here authorizing us to charge this card.”

Me “You have the card there on file. Can we just use that one?”

Hotel staff “Unfortunately we require a note confirming we can authorize the charge.”

Me “I have the email of the person who represents the organization paying for the hotel, where they specify that the room should be covered. Can I show you this email?”

Hotel staff “Sure! But we would still require them to complete and submit a credit card authorization form.”

Me “You have the card there on file. Can you charge it? I know there’s no note but the person literally gave you the credit card information, why do you need a note too...sorry, can I speak with the manager (my inner karen awakens)?

Hotel staff “There’s no manager on duty.”

Me “Okay (karen dies on the vine).”

Smash cut to me pulling out my own credit card to cover the stay.

Why do we put credit cards on file for hotels then? What’s the point of this? Either use the card on file or don’t. But why is it we can book a flight on an app, get through the airport security with a face scan, and then you arrive at a hotel and they need a notarized blood oath signed in triplicate? For most people I get that this is trivial who cares. But for anyone whose company or a third party covers travel, hotel check-in remains the final frontier. Why hasn’t this caught up with the rest of modern life?

[Seth Blumberg @sethblumberg]

@sebkrier I mean just stop what you’re doing and watch the first 10 minutes of this speedrun of Mario 64. How far is your mouth hanging open?

https://t.co/Bd4jDfdyPP

[Seth Blumberg @sethblumberg]

@sebkrier Pairs nicely with https://t.co/DsHCvGjL0s .

When you’re trying to do extraordinary thins with a system, the leaks are your entry points.

[Seth Blumberg @sethblumberg]

Usually if I know some content was written by AI, I get less interested.

But, add a human endorsement that sounds genuine… much more interesting.

And this is a pretty good story.

[james yu @jamesjyu]

https://t.co/F8YjxrLdyS

[Seth Blumberg @sethblumberg]

We’ll also need new ways to cheaply measure the value of a dataset or paper.

Like, oh I dunno… “does this data improve our forecasts about phenomena we care about?”

[alz @alz_zyd_]

The reward function needs to change: people should be rewarded primarily for creating new datasets, rather than primarily for analyzing existing datasets. There is no longer any meaningful skill component to the latter

[Seth Blumberg @sethblumberg]

The you have it folks: causal evidence vs revealed preference.

Which one do you think is most persuasive to potential customers?

[William B. Fuckley @opinonhaver]

it's wild to screenshot a conversation in which the other person is 100% right. did you think things were randomized in experiments for no reason?

[Seth Blumberg @sethblumberg]

In today’s edition of iOS becoming Windows…

[Seth Blumberg @sethblumberg]

Gonna be a rough year for the efficient market hypothesis.

[Seth Blumberg @sethblumberg]

You can tell prediction markets are having a moment when crypto bros issue explainer tweets on Expected Value, complete with \times and AI clipart of cash pipes.

[Yoshi @tradefoxintern]

Expected Value in Prediction Markets: Turning Intuition into Edge

1) Prediction Markets Are Still Inefficient

Prediction markets are supposed to be the ultimate truth-pricing machine. In theory, prices reflect all available information

In practice, they don’t. Most retail traders are just playing with a number in their head: 'This feels like a 60% chance, not 45%'

The problem is that these quick estimates are rarely based on complete information. They might come from a headline, a tweet, or a hunch. That’s not enough. If you want to build real edge, you need to gather more data than the average trader and actually translate it into probabilities

The more angles you cover, the closer you get to seeing the market clearly

2) What Information Matters?

The kind of information you need depends entirely on the type of market. Here are a few categories to make it concrete:

a) Politics

Useful signals: polling data, demographic splits, fundraising reports, historical turnout, campaign ground game, local endorsements

Where to look: poll aggregators like FiveThirtyEight, RealClearPolitics, FEC filings, state election websites, and even local newspapers that catch stories before they go national

b) Sports

Useful signals: injuries, lineup announcements, historical head-to-head matchups, advanced stats (xG in soccer, DVOA in football), even weather conditions

Where to look: stat sites like FBRef, Basketball-Reference, ESPN injury reports, and sharp betting lines from sportsbooks. Twitter is often ahead of the news here

c) Crypto & Finance

Useful signals: on-chain wallet flows, derivatives data, funding rates, options implied volatility, central bank policy

Where to look: dashboards on Dune, Glassnode, CME futures data, Fed minutes, crypto whale trackers

d) Event-driven markets (e.g. “Will Apple launch a foldable iPhone by 2026?”)

Useful signals: company filings, patent applications, analyst reports, supplier rumors, leaked prototypes.
Where to look: SEC filings, Bloomberg, Nikkei Asia supply chain scoops, industry blogs, and forums where leaks pop up

If you’re only looking at the obvious headlines, you’re playing the same game as everyone else. Edge comes from digging where others don’t

3) From Information to Expected Value

Collecting info is only half the job. You need to turn it into numbers. That’s where expected value comes in

a) Estimate your probability

Start with your own assessment based on all the information you’ve gathered. Say you’ve weighed the polls and turnout models and decide Candidate A has a 55% chance of winning

b) Compare it to the market

If the “Yes” side is trading at $0.45, that means the market is implying a 45% chance

Run the EV calculation

In this case:
Your probability: 55%
Market price: $0.45

Payout if correct = 0.10

That’s a positive expected value of 10 cents per dollar bet.

Over time, that edge compounds.

4) Building Strategies

Knowing how to calculate EV is the start. Building strategies around it is where the real money is made

- Play the long game

- Don’t chase one 'big win.' Place lots of positive EV bets across many markets. Even if you lose some, the math works out over time

- Pick your niche

It’s impossible to be the best at everything. Focus where you consistently have better info than the crowd. That could be local politics, niche sports, or a corner of crypto

- Stay flexible

Probabilities aren’t static. As new information arrives, your estimates should change. Updating quickly is often where the edge lies

- Manage your risk

A positive EV doesn’t mean you can’t lose. Bankroll management, sizing, and guardrails like stop losses and take profits keep you alive long enough for the edge to matter

Closing Thoughts

Prediction markets are inefficient because most people trade them on gut instinct. That leaves room for traders who take information seriously

If you can gather more data, translate it into better probabilities, and consistently identify positive expected value, you’ll be on the right side of the long-term math

At the end of the day, the market rewards those who move beyond 'this feels about right' and start playing with actual numbers.

[Seth Blumberg @sethblumberg]

Why would airlines ban chewing tobacco onboard?

Smoking—even vaping—sure, but isn’t chewing tobacco just weird gum?

[Seth Blumberg @sethblumberg]

Is there some weird reason this unravels?

What happens in the ambitious intermediate states?

Does having a Prob[fire the CEO] market fix the nonuniqueness of equilibrium somehow?

[Seth Blumberg @sethblumberg]

I think I got a non-pathological example - a status quo bias.

Imagine a fire-the-CEO market: company’s stock price if we do or don’t switch bosses.

Behaviors:
- “normal times”: revenues & stock are fine, we keep the CEO.
- “bad times”: revenues & stock down, we fire her.

[Seth Blumberg @sethblumberg]

In a futarchy, could we have multiple equilibria? In what circumstances? Just weird ones, on reasonable too?

@robinhanson 's decision selection bias means that price_A ≠ value_A, but: E[value_A | choose A over B].

What if the decision rule is "follow the market, mostly"?

[Seth Blumberg @sethblumberg]

Clearly, in this case, it _might_ be that the equilibrium is instead: just fire her (even in pretty good times).