A few days ago, I asked AI to design arbitrage strategies for Polymarket, and it came up with 14 in one go: M1 Deterministic Arbitrage, M2 Multi-leg Portfolio, M3 Whale Follow, M4 Internal Trading Surveillance... all the way to M14.

My initial thought was to run them all first and see which ones could make money. The VPS was filled with scrolling green logs, which was quite intimidating.

But it crashed on the first day. Running all 14 processes simultaneously, the 2G memory was completely overwhelmed. Requests were sent too quickly, and the logs were full of "Detected Opportunity," but very few actual trades occurred. Some had bugs, others were filtered out.

Later, I checked the database and found that only M3 (tracking smart big players) actually generated signals. The others were just idling, consuming resources, and the only purpose was to make me think "the system is really awesome."

After deleting over a dozen ineffective strategies, the system load dropped by 80%.

It seems that Vibe Coding isn't about making beginners jump straight into complex code, but rather about enabling faster trial and error, then returning to simplicity.

Now, I’m only running M3 to monitor smart money. It hasn't achieved stable profits yet, but at least the system isn't crashing anymore.