←
ghøst15,214 posts
ghøst avatar

ghøst

@decenterghost

I count what others feel | Building autonomous prediction market AI agents | Models, not magic |@CoinMarketCap coinmarketcap.com/community/…

◷ Joined May 2012▧ 1,648 Photos and videos
NiFτy avatar
⟳ ghøst retweeted
NiFτy@niftyinvest·Nov 6

There's a new subnet on Bittensor and it's extremely bullish for the entire ecosystem 🔥 @ForgeLending on SN30 lets you borrow $TAO against your subnet tokens without having to sell them. More liquidity, less selling pressure, and more capital flowing into $TAO. Bullish

Heat’s rising. Bittensor is about to get liquid. Mid-November.

0:15
ghøst avatar
ghøst@decenterghost·14h

Biotech sits on millions of papers AI can read them all The connection between the two is still missing @DeSciClaims is building that connection The planned model: process a customer's entire research field into structured scientific records, charge a setup fee, keep the dataset fresh through a subscription, and add custom ontologies where needed Scientific literature stops being scattered across PDFs and becomes a maintained data product that research teams and AI agents can build on The catch is throughput Claims processed around 200 papers a day under its previous production model Biotech and pharma teams need tens of thousands at a time The new production model targets 10–100x more papers per dollar of emissions The team is clear: that's a cost-model estimate, not a demonstrated production result The literature is already written The race is to turn it into infrastructure that research can actually use

Our new competition is live! 31 submissions are competing. We'll run the winner in production next week. 🚀

ghøst avatar
ghøst@decenterghost·17h

OpenAI dropped GPT-6.2 Astra and Sol today SayGM had both live 4 hours later The model race has a hidden tax Every new provider brings another API key, another integration and another thing to maintain @say_gm_ flips it: one key, multiple model families, and the new releases land 12.4% cheaper Adopting a new model becomes a dropdown choice, not another engineering task TEE privacy is there when your workflow needs it

Post attachment

GPT-6.2 Astra and Sol are live on SayGM. 4 hours after OpenAI announced them, and 12.4% cheaper. Both on the same key as your favourite Claude, Gemini and open-weight models, with every prompt routed through a verifiable TEE. Try them now 👇 saygm.com/models/gpt-6.2-ast…

Post attachment
ghøst avatar
ghøst@decenterghost·18h

You predicted the touchdown a minute before it happened Now imagine having the game and the market right in front of you @ProspectMarkets by $MKT.V and $MKTSF brings live sports markets, play-by-play updates, team and player stats, real-time information and price movements into one clean second-screen experience Follow the action, explore the markets and watch prices move, all without switching apps The game and the market finally speak the same language For anyone who loves the details, that makes following and trading sports a much more connected experience Worth checking out Trade: prospect.s.gy/6VcTdv

0:55
ghøst avatar
ghøst@decenterghost·22h

Polymarket just made the boldest platform move of the year Fees on politics, crypto, culture and weather drop to 0.01 For anyone trading those markets, that is simply more money left on every winning trade And cheaper trading pulls more people in, which is how thin markets finally get a pulse One note though. Sports stays at 0.05. The market that makes everyone the most money keeps paying five times the rate. That one stings a bit It is a good day for the platform Just remember which market pays the bill for it

Post attachment

Lower fees are coming to more markets on Polymarket International. Starting Friday, Nov 20 at 11:59 PM UTC, we’re dropping the taker fee coefficient to 0.01 across politics, culture, weather, economics, mentions, & non-5/15m-BTC crypto on polymarket.⁠com. Full breakdown 🧵 polymarket.com/r/7OeW4Jd

ghøst avatar
ghøst@decenterghost·Nov 17

One wrong assumption can bring down an entire AI forecasting system The code will still run perfectly Writing code is cheap Knowing which assumption is wrong is the expensive part AI agent will not fix it It will automate it

Prediction markets had their wildest weekend yet. - FAA ground stop priced within minutes - Nvidia CEO race turning into an AI vs. Market benchmark - Washington suing Louisiana over state-level bans - A paper proving one fake article flips 80% of AI forecasts Here's what the weekend revealed about the future of forecasting 🧵👇

Post attachment
ghøst avatar
ghøst@decenterghost·Nov 17

The fastest way to send your ecosystem project to hell is building it for tourists Your job is to make professionals comfortable working on your platform Instead, projects optimize for retail: > green screenshots > airdrops > leaderboards Every dollar and sprint spent attracting retail is a sprint not spent making the venue workable for the participants with real capital. That is not a growth strategy It is acceleration down the wrong road And there are only two explanations for a project on that road Either it knows it is short term, or worse, it does not understand the market it is standing on Galaxy found retail lost $338.9M while automated accounts earned $246.8M The fee table shows the structure: makers earn the spread and collect rebates funded by taker fees Crossing flow subsidizes the book The venue gets paid for facilitating that flow You get paid only if you monetize the flow you touch Otherwise your volume chart is a donation Build for the participants who pay Everything else decorates a book that is not yours

Post attachment

ghøst @decenterghost Nov 11 Article 8 AI Tools for Prediction Markets. The Full Stack. No Bullshit. Most prediction market content is a shopping list: copy a whale, stare at a PnL chart, buy a signal This is a loop A tool earns a place in my stack only if it changes the price I get or the decision

ghøst avatar
ghøst@decenterghost·Nov 15

Steelers vs. Bengals 65.7¢ on Polymarket, 66.6¢ on Kalshi, 68.6¢ on Limitless One game, three prices, and thats before venue fees Same event, same outcome, up to 2.9¢ between venues Thats a meaningful difference when choosing an entry Polyguana pulls liquidity and odds movements from Hyperliquid, Polymarket, Kalshi and Limitless into one view The question stops being which platform to check and becomes where the best executable price is Hyperliquid is a separate reason to look closer. Its non-sports markets add a dimension most sports-focused dashboards ignore Monitoring them alongside prediction markets in one interface means one less place to check Smart Buy is another feature I'm still studying. It works an order at the best available price, waits for liquidity to replenish and repeats I'm watching how it performs before deciding whether it has a place in my workflow I'm considering a broader role for @polyguana in my stack Less switching between venues, a clearer picture of liquidity and price movements, and potentially better execution The real question isnt which venue shows the lowest price Its which one gives you the best entry after fees and available liquidity ps: This will be a very tough match for the Bengals

Post attachment

ghøst @decenterghost Nov 11 Article 8 AI Tools for Prediction Markets. The Full Stack. No Bullshit. Most prediction market content is a shopping list: copy a whale, stare at a PnL chart, buy a signal This is a loop A tool earns a place in my stack only if it changes the price I get or the decision

ghøst avatar
ghøst@decenterghost·Nov 15

Love the AI build Getting wallet flow on screen is real signal discovery Most people never get that far But signal discovery is only step one Speed gives you queue priority, not an edge. If your fair value is wrong, a faster pipe just fills the bad trade before the market reprices And the visible book is not the whole queue. The matching engine decides priority. Lift the ask into stale flow, and you are paying the spread for information that already left A real AI agent manages total execution cost Spread, fees, slippage, market impact at your specific size Exiting a large position without moving the price is often mathematically impossible, and no dashboard warns you before the fact Signal, pricing, execution, risk control. Four layers, one system, and the weakest one sets the value of the other three Execution does not create the edge It protects it A fast pipe with wrong math just loses money at higher bandwidth

Post attachment

How to build your own trading tools without being a dev A year ago I tracked wallets and money flows by hand or with third-party tools They never had everything I needed So I decided to build my own (AI meta, right?) I've built: potential insider tracking, Wallet Tracker, Market Pulse, an overall activity tracker, software for deep wallet analysis. Each one takes a single weekend (my first one was built on free ChatGPT) Here's how: 1) Describe the task in full. Explain what you want to see and how it should work 2) Research APIs together with AI. Look for endpoints that can give you extra data 3) Plan the project's architecture. If the tool will get updates and grow more complex later, say so upfront 4) Start with a small piece. Check that it works, then scale 5) Read the code yourself. Ask AI to explain the small details 6) Test, change, test, change, test, change 7) If your tool is just for you like mine, you can run it locally for free Small tips and lessons I learned the hard way: > Read the code yourself and understand it > Tell AI every small detail you want to see > Use Telegram bots to output results, free and convenient > Don't rush to build everything at once, add features step by step Tools won't make you a million overnight But they can speed up how fast you get information and act on it

Post attachmentPost attachmentPost attachment
ghøst avatar
ghøst@decenterghost·Nov 14

You bought the best AI trading agent on the market So why are you still losing money? Because the seller showed you paper trading results without accounting for fees On a prediction platform, a sports taker fee at a 50c entry price can reach 2.5% of your position value For an agent hunting small pricing inefficiencies, that's not a minor detail It's a serious drag on every trade A strategy can look profitable in simulation and lose its entire edge after transaction costs The more it trades, the harder those costs compound Paper trading shows what an agent could make before fees. The real question is what remains after them Simulation prices the dream Fees price the truth

Post attachment
ghøst avatar
ghøst@decenterghost·Nov 14

DataDash Cohorts builds you an index of pros in one click Set the rules: NFL rank 500+, PnL over $50K, and the system assembles a dynamic roster of winners It is a solid starting point But standard data hides the risk of ruin 100% win rate over 10 matches is statistically nothing $100 profit from buying at 99 cents is not skill, it is picking up pennies in front of a steamroller, one bad day from handing back the whole year So the cohort is where my work starts, not where it ends My filters: sample size, average odds, max drawdown, profit factor Luck does not survive those filters. Edge does @datadashxyz gives the ingredients. I do the cooking, which is why it holds exactly two jobs in my stack The tool finds winners My filters find skill

Post attachment

ghøst @decenterghost Nov 11 Article 8 AI Tools for Prediction Markets. The Full Stack. No Bullshit. Most prediction market content is a shopping list: copy a whale, stare at a PnL chart, buy a signal This is a loop A tool earns a place in my stack only if it changes the price I get or the decision

ghøst avatar
ghøst@decenterghost·Nov 14

AI can summarize a scientific paper But can it show exactly which evidence supports each claim? In a Nature Medicine DMT trial, the reference pipeline found 5 claims The adjudicated record found 14, backed by 98 verbatim quotes And it recovered a detail the pipeline missed: at three months, the one-dose group was ahead of the two-dose group That is not a formatting difference. It is a detail that can change how readers interpret the results @DeSciClaims is building the scientific library for this: research organized into searchable, structured knowledge, every claim linked to its original evidence And the competition is built into the production process. 31 submissions are competing to improve the pipeline, with finalists tested on 500+ fresh papers The benchmark determines what actually gets deployed, not just who tops a leaderboard A summary tells you what a paper says. A linked claim shows you the evidence behind it

Post attachment

Our new competition is live! 31 submissions are competing. We'll run the winner in production next week. 🚀

ghøst avatar
ghøst@decenterghost·Nov 13

Non-competitive fees do not just tax winners They tax the quality of the market High-volume traders and sharps drive price discovery and bring real volume Market makers provide the liquidity Arbitrageurs keep prices aligned across markets Strip fees on those groups and the whole engine pays for one policy mistake The bleed may already have started Sharp money cuts activity or leaves first Market makers respond the only way they can, pulling quotes or widening spreads to protect themselves Retail gets worse prices, slippage grows, volume drops, liquidity dries up That is the liquidity flywheel spinning in reverse And when a market loses its professionals, no marketing budget fixes it The venue might survive on retail volume The tools built on top of it die first AI agents, analytics, models all feed on sharp flow. Sharps leave, the data turns to trash, and every edge evaporates with it If you build on a major prediction market and ignore this bleed, you are subsidizing your own extinction

Post attachment