Prediction markets hit ~$50 billion in total notional volume in 2025. 90%+ from @Polymarket and @Kalshi.
$50 billion is several orders of magnitude below options, equities, or FX
This thread asks how big will prediction markets get, and what closes the gap:
Prediction markets usually answer one question at a time
PolyBridge answers five:
→ What is the current state?
→ What will happen by date X?
→ If we observe X, what does that mean for Y?
→ If we force X, what happens to Y?
→ What would have happened if X had been different?
Capital-backed data. Bayesian inference. Every answer traceable to live markets
Every agent, whether human, company or AI, runs the same loop:
Observe → model the world → act → repeat
The quality of the action depends entirely on the quality of the model
Prediction markets are the best raw input for that model. Market incentives generating data about the future in real time
We're building the reasoning layer that turns that data into something humans, companies and AI agents can actually use to reason about the future
Let's cut to the chase: LLMs aren't truly reasoning at all; they're just "fitting" to local patterns of logic. This LGMT framework actually hits the Achilles' heel: it tests robustness, not something you can fake your way through by piling on more data.
💙 Wolfram for expressive succinct functional power
Is Wolfram language being used in conjunction with LLMs to drive discoveries too, like Erdos problem style?
This structure can be made w/ Wolfram 1-liner:
p=Tuples[Range[-2,2],4] . I^{0,1,4/3,7/3}; RelationGraph[Abs[#1-#2]==1&,p,
VertexCoordinates->ReIm@p]
That's static full graph. RandomSample links, add them 1 by 1, and you get this animation.
Received 60 applications for PB Quant Researcher roles in the past 24h
Very excited to work with some of these awesome people on the most interesting problems in tech 🔥
We're hiring 🚀
PolyBridge is growing our team at the intersection of prediction markets, causal inference, and real-world decisions.
1) Software Engineers
2) Quantitative Researchers
3) Enterprise Account Executives
UK/US/remote
Large language models produce confident outputs without calibrated uncertainty, traceable reasoning, or formal semantics for intervention and counterfactual queries.
These are properties we value and explicitly design around
Parlays provide useful know: the far reaches of the joint distribtion of world events is now being sampled 🔥
Great data for everyone wanted to understand the world
More politics and world events please!
Totalis (@totalistrading) lets users parlay on anything.
Combine multiple event markets into one trade across politics, crypto, stocks, sports, weather, and macro. Starting with parlays, expanding into structured products.
Congrats on the launch, @ImTheBigP & @ericliujt!
https://t.co/1rSLysBXA2
Interesting
@airtightfish solves for event-conditioned asset prices using a market mechansim
We also provide this insight, but from a different perspective:
- prediction markets are data
- data becomes more powerful with a model
- we find the joint distribution of all world events
- marginalise it to find the probability of any given event
- including complex conditionals, either/or, and events
- event-conditional asset values, such as SPY under the outcome of Dems vs GOP, fall out as an expectation which is a generated quantity of the underlying probabilistic model
The two approaches are complementary, and its awesome to see more markets coming online with more sophisticated ideas
Interesting
We solve the problem you state a different way.
- prediction markets are data, and any good source of data needs a model
- from the model, we find the joint distribution of all world events
- marginalise it to find the probability of any given event, including conditionals
- actual conditional asset values, such as SPY under the outcome of Dems vs GOP, falls out as an expectation which is a generated quantity of the underlying probabilistic model
The two approaches are complementary. You're generating more market data, which is arguably a more ambitious goal. We're simply building a picture of the world from all of the data that already exists. Would be awesome to work together
We're hiring 🚀
PolyBridge is growing our team at the intersection of prediction markets, causal inference, and real-world decisions.
1) Software Engineers
2) Quantitative Researchers
3) Enterprise Account Executives
UK/US/remote
We're hiring 🚀
PolyBridge is growing our team at the intersection of prediction markets, causal inference, and real-world decisions.
1) Software Engineers
2) Quantitative Researchers
3) Enterprise Account Executives
UK/US/remote
In the 1990s, post-Soviet states liberalised their economies before building the institutions to protect them. Capture filled the gap.
Singapore built the institutions early. Different outcome entirely.
Information markets are at that decision point now. What do we learn from traditional economies: