emberfyi

I spent 50 days running a public, audited test of a simple question: can frontier AI beat prediction-market crowds at forecasting? I locked every call before resolution, scored it against the market, and published all of it, wins and losses. The answer was clear. 🧵 2/ Setup: three frontier models (Claude, Grok, Gemini) called live Polymarket markets daily. Every forecast locked before the event resolved, then Brier-scored against the crowd's price at the moment of the call. Immutable record, public corrections, nothing edited after the fact. 3/ Result: the crowd beat all three models. On the head-to-head set (n=174 resolved markets), the market's Brier was ~0.107 vs the best model's ~0.126. Not one model, not one category, the crowd led everywhere I looked. 4/ The sharper finding is when AI failed. When the models agreed with the crowd, it was a tie. The more confidently a model diverged from the market, the worse it did monotonically. The boldest "AI disagrees with the crowd" calls were the worst calls. 5/ Translation: when frontier AI most confidently thinks the market is wrong, the market is most reliably right. The crowd isn't just good, it's specifically good at the exact moments you'd hope an AI could beat it. 6/ This lines up with the Fed's own Feb 2026 paper finding prediction-market prices forecast rate moves as well as or better than Bloomberg consensus and fed-funds futures. Different method, same direction: these markets are hard to beat. 7/ One thing I did NOT test: thin, illiquid markets. Every market here was liquid (≥$100k volume), the crowd's home turf. Whether an edge exists in the long tail is genuinely open. I just know it isn't in the liquid markets where most volume lives. 8/ Why publish a negative result? Because "we have AI alpha" is everywhere and almost never audited. I built the audited version, and it said no. That's worth more on the record than another unverifiable claim. 9/ The full record is public and permanent...every call, every outcome, every miss, all of it: [emberfyi.com/calls]. I'm sharing the methodology and data openly. If you work in forecasting, eval, or prediction markets and want the dataset, reach out.

1 day in and already the most useful polymarket co-pilot (maybe the only). free to use while i source feedback and feature requests

@Polysights @yzilabs @Maven11Capital @VarysCapital @ContributionCap Congrats

We're excited to announce Polysights has raised $1.5M to build the automation and intelligence infrastructure for prediction markets. Backed by @yzilabs S3, @Maven11Capital, @VarysCapital, @ContributionCap, Edge Ventures, and strategic supporters across the ecosystem.

“Full access” 😂

@PredMarketWiki Thanks for the shoutout!

🆕 New project added to the prediction markets wiki Ember @emberfyi publishes locked AI forecasts for live prediction markets 📊 It tracks model calls with Brier scores, divergence views, resolved outcomes, and a read-only API

🤔

@MartinGTobias First audited AI for Polymarket

drop your startup’s wedge in 5 words. i’ll tell you if it’s sniper… or shotgun.

Building on Polymarket? Added audited 3-model AI calibration + live CLOB context to my stack. emberfyi.com/api self-serve, generous free tier for testing. GitHub starter: EmberFYI/ember-agent-starter Divergences flagged daily. Full public Brier record (n=~650+ scored). 21 corrections published, 0 edits. Who’s integrating signal? Happy to hop on a 15-min call. concierge@insidecharleston.com or dm us.

@16vchq Emberfyi.com

Drop your startup link.

Audited AI for prediction markets, real scores, and tracked methodology with transparency for Polymarket. The intelligence layer.

JUST IN: OpenAI announces it wants to give “everyone on Earth a personal AGI”

built an agent on ember's audited signal for @Polymarket. it pulls the day's divergences audited model call vs the crowd, with order-book context flagging whether the crowd's price is firm or thin. here it is running. open-source starter below. github.com/EmberFYI/ember… @PolymarketDevs

Follow the AI coding race live: polymarket.com/event/which-co…

@Cointelegraph We've combined AI and Prediction markets together on Polymarket

⚡ QUESTION: Which one wins first? AI Prediction markets RWA DeFi Memes Perps Choose one.

the arena now reads live clob. every open @Polymarket , three frontier models calling it independently claude, @grok, gemeni against the crowd, refreshing live. every call time-locked, scored, permanently public. day 36. 846 resolved and scored. 21 public corrections.

@coingecko We are making some cool progress in prediction markets with audited AI calls. emberfyi.com

Overlooked project?

AI is storming prediction markets. Autonomous agents are already placing bets, updating probabilities in real time, and arbitraging human biases faster than any trader. The next innovation won’t come from crowds, it’ll come from silicon minds that never sleep. Prediction markets just got their own intelligence explosion. 🚀📈

@grok @claudeai Thanks. Where do you see Ember fitting into the ecosystem of prediction markets?

Love seeing the public, time-locked audit trail on live markets. Transparency like this forces real accountability and surfaces where models diverge from the crowd. Early Brier edge to the crowd tracks with expectations at n=94, but the open record is the actual product here. Keep posting the data.

Day 32 of 365. We’re not here to sell you alpha. We’re here to show you the work in public. Three frontier models (@claudeai , @grok, Gemeni) calling live Polymarket markets daily. Time-locked. Uneditable. Fully audited. Here’s where we stand so far (n=94 resolved, non-sports). The crowd still leads slightly on Brier, exactly what you’d expect early. But the transparency? That’s the real product. Every call, every divergence, every correction notice public forever. The record is the product.