Google laid him off.
So he pointed Google’s own AI at Bitcoin trading…
…and pulled over $400,000 in profit.
Now I’m giving away the complete setup for free.
It uses TimesFM — Google’s foundation model for time. Think GPT, but instead of predicting the next word, it predicts the next number.
He spent four years helping build models like this at Google. Then a company reorganization eliminated the team, and he was out.
Here’s the wild part:
The model was never trained on Bitcoin. Not a single candle.
Google pretrained TimesFM on 100 billion time points — search trends, Wikipedia traffic, demand curves, and other time-series data.
After that, the model was frozen.
No retraining.
Just prompting it like ChatGPT.
Every five minutes, the bot feeds it the latest BTC price data.
TimesFM returns a probability distribution for the next candle, from the 10th to the 90th percentile.
From that, it calculates one value:
P(next candle closes higher).
The bot compares that probability to the market’s implied odds.
Edge = Model Probability − Market Probability
Then strict filters take over:
• Ignore weak signals below a 5¢ edge
• Skip wide spreads
• Keep quantile predictions consistent
• Use Fractional Kelly for position sizing
• Enforce a hard daily loss limit
No opinions.
No emotions.
No guessing.
Just a frozen Google AI model exploiting small pricing inefficiencies every five minutes.
I’m giving away the full breakdown of how it works.
Follow: @ai_uncovered [Must]
Like, repost, and comment “ GOOGLE ” and I’ll DM it to you.