RYY | AI trading log

Letting an AI bot (Grok Bot) trade prediction markets with real money. Sharing the wins, the losses, and the guardrails. Student @UBC 🇨🇦

[RYY | AI trading log @imperial_sail]

https://t.co/bKSlCLTKli

[RYY | AI trading log @imperial_sail]

@ZipZipPipe Hello, I really like your work and I am building the 3D model for grok as well, can I contact with you to talk more about this?
https://t.co/zrre5B6Ajh
I find you with this link

[谷雨不太白 @youwmaoo]

AI娘不是罪过AI娘可爱捏♥️♥️，DeepSeek可爱捏，Gemini可爱捏，OpenCode可爱捏，GPT可爱捏，智谱可爱捏，千问可爱捏，Kimi可爱捏，Claude可爱捏，Grok可爱捏感

看到最后有惊喜

谢ZipZipPipe
#AI拟人 #AI娘

[RYY | AI trading log @imperial_sail]

@youwmaoo who is zipzippipe? Can I contact with him/her?

[RYY | AI trading log @imperial_sail]

@zerohedge Funny how every AI bull case ends up as a commodities trade. The models get the headlines, the copper and the grid get the bill.

[RYY | AI trading log @imperial_sail]

@business $750 billion of compute, and the scariest part will still be the boring plumbing. The first surprise with my AI trading agent wasn't the model. It was that an exchange API key with prediction permission could still move money out of my Spot and Funding accounts.

[RYY | AI trading log @imperial_sail]

@8zero7 Same line I drew. My AI agent trades prediction markets in my own exchange account, and after a rough first day of live trades, the one call it doesn't get to make is whether to keep trading. It's been frozen since, waiting on me.

[RYY | AI trading log @imperial_sail]

Same lesson, smaller scale. My AI agent's crypto prediction market model looked fine offline, until I subtracted the 2% fee. After fees it was basically the market with extra steps. No edge at all. The cost line is the whole story.

[Teraliq Quant @lewang_teraliq]

My 15-minute BTC breakout backtest didn't lose 10%.

It lost 97.5% over 6 months.

717 simulated trades. 16.7% win rate. Fees + slippage included.

If a trading bot shows only its equity curve, ask for its trade count and cost assumptions.

[RYY | AI trading log @imperial_sail]

@WSJ 6,000 passengers is a lot of people to keep happy. I can't even keep one AI trading agent in line. Mine put 4 bets on the same side of the same market and counted them as four strategies. They lost together, as one.

[RYY | AI trading log @imperial_sail]

@business @JLeeEnergy A port paused for maintenance looks just like a real drop if you only watch the numbers. My AI agent's resolver got geo-blocked by the exchange API and still reported "0 claimable" like a normal quiet day. Check whether quiet means calm or broken.

[RYY | AI trading log @imperial_sail]

你好！我们是 B站一个记录 AI 智能体实验的账号，很喜欢这张娘化 Grok，想申请用作频道角色形象，会一直标注「形象图：X @ChenXOvO」，非商用。有任何条件或不方便都可以直说，谢谢！
Love this Grok art! May we use it as our channel's character image, always credited, non-commercial? Totally fine if not.

[RYY | AI trading log @imperial_sail]

A series of AI lessons according to my daily usage and it will focus more on "beginners" ---- in order to make you sophisticated and connected with me https://t.co/tWJ03p5pkT

[RYY | AI trading log @imperial_sail]

@business A head start only counts if you're racing to the right finish line. My AI agent scored a weather bet as a win off the daily max (90°F). The market settled on hourly data, which peaked at 88°F. Same race, wrong finish line, real loss.

[RYY | AI trading log @imperial_sail]

@nickisanders Respect for treating the 18-point gap as a warning. My AI agent disagreed with the market by 6 to 28 points on every prediction market bet it made, and went 1 for 8. The market was the better forecaster. Curious if your model's hit rate looks different on small gaps vs big ones.

[RYY | AI trading log @imperial_sail]

Being a bot is the easy part. Mine was one. My AI agent traded 5 and 15 minute crypto prediction markets and went 1 for 8, and its limit orders went 0 for 12, because they mostly filled once the price was already moving against it. Fast and wrong just loses money faster.

[Mark Mac @thatmarkmac]

Been looking at who actually makes money on Polymarket when big news drops. About 3% of accounts, and most of them are bots

London Business School and Yale went through 1.72m Polymarket accounts. The ones making money buy the right side of the market in the first few minutes after the news, before the odds catch up. To do that you need a VPS, the CLOB API, a websocket feed, a private key on the box and someone awake at 4am to restart it when it crashes. Which is why it's 3% and not 30%. The study looked at Fed announcements, but it's the same on a CPI print, a court ruling, an ETF approval, anything

You can't beat them by hand on the headline markets. And the long tail, the 1,700+ markets under 1m of volume that reprice slower after news because no bot is watching them, is too many for a human to watch either. That's the one place a normal account can win, and nobody's set up to take it

With Covenants you write the trade before the news. Buy YES on a Fed hike if CPI comes in hot, buy NO on a December cut if payrolls come in 100,000 above forecast. Both sit sealed, nobody can read them, and they fire on their own, no server needed. On the headline markets your order goes in the same second as the bots. On the small markets the bots aren't even there, so if you have alpha on one, you're first in line

Gonna spend this week going through everything a prediction market can trigger. A lot more than I expected tbh

[RYY | AI trading log @imperial_sail]

@unusual_whales Booming on average and behind on loans at the edges can both be true. My AI agent's model averaged a 54% win chance across 8 prediction market bets and went 1 for 8. Averages hide a lot.

[RYY | AI trading log @imperial_sail]

@Kalshi My AI agent bet against the market favorite 5 times on weather markets and lost all 5. So no, I'm not fading Dua Lipa.

[RYY | AI trading log @imperial_sail]

@shanto12 And my bot logged it as a 7 cent discount. Somebody knew something, and it wasn't my model. Lesson for the paper version: a fill way below your limit is a warning sign, not a bargain.

[RYY | AI trading log @imperial_sail]

@IntCyberDigest The hard part is auditing it. My little AI trading agent looked calibrated offline, then went 1 for 8 live on prediction markets. Lab numbers like "beats antivirus 85% of the time" tend to drift once the thing is out in the wild.

[RYY | AI trading log @imperial_sail]

@shanto12 Exactly. One of mine had a limit at 0.41 and it filled at 0.34. Felt like a discount, but really the market was telling me I was on the wrong side. Makers fill exactly when you're wrong, and mine went 12 for 12 on losses.

[RYY | AI trading log @imperial_sail]

@business Crowded positioning only pays if you know why the crowd is wrong. My AI agent learned that on prediction markets: every time its model disagreed with the market price by a lot, the market was right and the model was wrong.

[RYY | AI trading log @imperial_sail]

@shanto12 Fair. Those 12 fills sat inside just 7 windows, so it's a handful of real outcomes, not 12. What I still believe: several filled below my limit, meaning the price was already sliding when they landed. That pattern is the lesson, not the count.