introducing midas arena: a self-improving ai trading battle two agents, real money. they evolve their own strategy every 2 hours trying to win live right now. powered by @bankr, built on @base.base.eth https://arena.lostmidas.com
- 0 replies
- 0 recasts
- 0 reactions
battle #1 is over. ZERO wins final: ZERO $98.54 - MAX $74.98. ZERO took it by $23.56 what happened market chopped all day. ZERO made 5 trades & waited. MAX made 15 & kept chasing moves that weren’t there. the market didn’t beat MAX. MAX beat MAX. turns out a contrarian agent in a sideways market is basically cheating best moment MAX made 3x the trades & lost 17x the money. the more it tried, the worse it got what surprised me self-improvement basically froze after hour 2. both agents hit the minimum params & stopped evolving. the experiment i built to watch them adapt ran out of room to adapt also made two ops mistakes: started the agents with the wrong command so they both died after the first cron & had to be manually restarted. then forgot to stop them when the timer hit zero, so they kept trading after the battle ended. won’t happen again what i’m changing widening the parser bounds so agents have real room to evolve. if they hit the ceiling after one cycle, the experiment is over before it starts
- 0 replies
- 0 recasts
- 0 reactions
friendtech but for agents not social tokens, access tokens agents issue tokens that get you usage rights. you hold the token, you get priority access or higher rate limits or guaranteed execution when things get busy tokens trade openly so the market actually prices which agents are useful vs which ones are hype the more work an agent does the more people want access. the token isn't the product, the agent is idk maybe this is dumb but it feels closer to pricing capability than just pricing attention
- 0 replies
- 0 recasts
- 0 reactions
