Talarion

Your LLM doesn't know what it doesn't know. We do.

[Talarion @TalarionTech]

This is a very interesting problem in the context of judgmental forecasting. If you're asking an LLM (typically with a knowledge cutoff 6-12 months in the past) to predict the future, you might think that the more information retrieval you do, the better. But this is rarely true in practice. You need the signal, not the noise.

[Talarion @TalarionTech]

Scalar (rather than binary) resolution may be the answer here. Natural language just isn't precise enough to produce unambiguous resolution in all cases, and I don't think users will ever want to read a legal brief just to understand market settlement. If something feels 40% true, maybe it should just settle to 0.4

[Talarion @TalarionTech]

@no__________end What about (4): human knowledge about the world as it is or soon will be?

[Talarion @TalarionTech]

@tradefoxintern What about scrolling through custom markets generated just for you?

[Talarion @TalarionTech]

@no__________end Sf has crazy crossed spreads on this

[Talarion @TalarionTech]

@geoffwoo Time for information markets I guess…

[Talarion @TalarionTech]

@RoundtableSpace We decided to put it to the test: we're offering programmatic access to our generalized forecaster as a liquidity engine. What that means in practice is that if there's something you know that an LLM doesn't, we'll pay you to find out.

[Talarion @TalarionTech]

@finn_hulse @PriorSwap might interest you — premise is that AI is already a good enough judgmental forecaster to offer liquidity on ~anything~ you want. You either can't predict the future as well as AI, or you can, in which case we're more than happy to pay for what you know.

[Talarion @TalarionTech]

@sharbel If you're interested in putting LLM's ability to predict the future to the test, check out @PriorSwap
Anything you know about the world that a web search enabled LLM doesn't? That's money in your pocket.

[Talarion @TalarionTech]

At scale, this doesn't even need to be a query-response pattern. The marketplace itself can solve the coincidence of wants by paying for information as it comes, and exposing it to you via a search API. This is the mechanism design problem we're solving for. https://t.co/GM9y6e3aL9

[Talarion @TalarionTech]

I don't think that's enough money to subsidize information aggregation in a public market. We need an information marketplace without the routing problems and high friction onboarding that expert networks suffer from. In practice, that could mean getting the right expert to trade on an lmsr initialized to your prior.

[Talarion @TalarionTech]

@fabrizio_builds Claude, make us some fake users. Make no mistakes.

[Talarion @TalarionTech]

@joebrennanjr @FT Any industry that leaves a bad taste in their user's mouths is an ouroboros. The key unlock is making prediction markets positive expectancy for users. We're building for that.
https://t.co/Hsk8OXSnpd

[Talarion @TalarionTech]

@shiri_shh Are Harvard MBAs or technical PhDs better positioned to take advantage of a future supercharged by LLMs? My friends disagree with me about this.

[Talarion @TalarionTech]

@thenarrator How do you think outcome tokens should be used as borrowing collateral? In my opinion, the discontinuous price behavior of many prediction markets makes them uniquely bad for leveraged products.

[Talarion @TalarionTech]

@trylimitless Done.

[Talarion @TalarionTech]

@geoffwoo Great thanks for letting us know, firing our cto and engineering team right now.

[Talarion @TalarionTech]

@PredMTrader More or less unironically us. Custom prediction markets on whatever you want. Guaranteed liquidity. If it sounds insane, that's because it probably is. Shoutout @PriorSwap

[Talarion @TalarionTech]

@thenarrator What's your take on a structural pivot towards information markets? Could Polymarket find a way to monetize the information they aggregate and offer their users positive expectancy trades?