Astra 自主交易三天：200 美元变 1600
给 Astra 一个券商账户跑三天，47 个假设里自己砍掉 41 个，最终资金翻了八倍。
GPT-6 Astra killed 87% of its own ideas in three days. Wall Street pays an analyst $650K a year to do the same thing twice as slow Gave it access to a brokerage account and three days with no intervention $200 turned into $1,600, and that's not the main number. The main one: 41 out of 47 hypotheses never made it to the market One of the survivors still bled $60 on day one. Not because the filter is bad, but because slippage is invisible on a backtest, you only see it on a live market That's the control point. The system isn't rubber-stamping everything, it actually takes risk and actually gets it wrong, just at small size while it's still learning The next night a similar setup didn't even make it to execution. The model read its own log and blocked it itself A retail trader tests an idea with money. Here the money only sees what already survived 40+ deaths on paper The full cycle behind these numbers is broken down in the post below ⤵
这条是引用着另一条写的：「让 GPT-6 Astra 自主交易：三天 200 美元变 1600」