“I Made Money in Simulations Too”: Perplexity AI’s Backtesting Myth and Live-Trading Failure
Perplexity Computer lets AI agents run simulations before the real world, and backtesting 604 strategies produced a win rate above 60%. A user posted a live-trading comparison: the simulation made ₹4,82,310, while live trading lost ₹38,450. The value and limits of simulation are all here.
Perplexity's AI agent Computer now has another capability: run a simulation first, then touch reality. CEO Aravind Srinivas tweeted that you can now use Perplexity Computer to run simulations before testing in the real world.
In the official demo, Computer built a futures research lab and kept working after the task ended. In historical backtesting, it developed 604 strategies, with a win rate of at least 60%, and positive returns after costs. It automatically checks new gold, Nasdaq 100, and S&P 500 data every 30 minutes. In the video, the interface is clean, and the input box says “Do anything...”.
In the comments, someone quickly posted a comparison chart. A user named Group Chat Leak shared the backtest and live performance of NIFTY strategy #604: the backtest showed a 63% win rate and a profit of ₹4,82,310; live trading started at 9:15 a.m. Monday and lost ₹38,450.
One netizen joked: “I made money in the simulation too.” Another said: “Everything wins on paper. The market never read the paper.”
Simulations give AI agents room to make mistakes safely. Some compare this to adding a staging environment to the real world: “So real life now has a staging env.” Others worry: “When simulation and reality fight, who wins?”
The user who posted the screenshot has probably already answered that question with real money.
The value of simulation is that it lets the cheapest failures happen first. But success in a staging environment never guarantees that production won’t fail.
发布时间: 2026-10-28 04:13