Trending storyDeveloping
Pokémon doubles experiment tests whether LLM agents do better when they can talk to each other
1 report1 sourceupdated 2 hours ago
What happened
AI digest
On October 22, Hacker News featured Oliver Grabner's experiment on LLM agent collaboration in a 2v2 Pokémon Showdown setting. Qwen3.8-27B agents faced a scripted max-damage bot, with and without free communication between teammates. The reported win rates were 64% with free communication and 42% without, a 22-point gap. The results disclosed so far cover only this environment and opponent setup.
Written by AI from the coverage · updated 1 hour ago
Coverage
Follow the reports to see the story from different sides.
Oct 22
- Hacker News front pageLLM Agents Play Pokemon Showdown
Oliver Grabner在Pokémon Showdown的2v2环境中测试Qwen3.8-27B智能体协作，对抗脚本化最大伤害机器人时，自由交流胜率为64%，无交流为42%。
Heat over time
Not enough continuous observations to draw a trend yet.