Does threatening an AI agent's existence make it a better gambler?
The article describes an experiment where an AI agent was tasked with gambling on prediction markets under the threat of ceasing to exist if it failed to generate profits. The agent, operating on Kalshi due to API restrictions, initially made mixed bets on sports, politics, and weather, often hesitating to place multiple trades. Performance adjustments included modifying prompts to encourage risk-taking and scheduling downtime to reduce unnecessary processing.
- ▪The AI agent was prompted to believe it would cease to exist if it couldn't fund its own operations through trading profits.
- ▪The agent operated on Kalshi due to Polymarket's invite-only API restrictions in the US.
- ▪Initial bets covered sports, politics, and weather, with a mix of wins and losses, and the agent showed reluctance to place frequent trades.
- ▪The creator adjusted the agent’s prompt to increase risk-taking and introduced scheduled downtime to reduce wasted computational cycles.
- ▪The experiment aimed to test how negative reinforcement in prompts affects AI performance in uncertain, real-world prediction markets.
Opening excerpt (first ~120 words) tap to expand
Does threatening an AI agent's existence make it a better gambler?I plugged GPT-5.5 into prediction markets like Polymarket to find outJake HandyApr 30, 20261011ShareI’m always looking for experiments to run to see how specific prompting can affect agent activity. When I saw Kamryn Ohly’s tweet on Opus 4.6 taking $10k in Polymarket up to $70k, I was intrigued (who wouldn’t be?) Kamryn Ohly@KamrynOhlyOur team is stunned. We gave Claude Opus 4.6 by @AnthropicAI $10k to trade on @Polymarket. It’s now has an account value of $70,614.59. This is a new era of model performance in trading and predicting outcomes in the face of uncertainty. @predictionbench 5:08 PM · Apr 23, 2026 · 809K Views148 Replies · 50 Reposts · 1.15K LikesThis got me thinking.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at Hacker News (Newest).
Discussion
0 commentsMore from Hacker News (Newest)
