Runescape Bench talk: benchmarking AI agents on long tasks in an MMO economy
mariofilhoml · x · 2026-10-19
Max Bittker's Runescape Bench talk is now on YouTube, covering benchmark results, the design of the evaluation harness, and his economic swarm experiments with agents inside the game's virtual economy. The retweeter calls it highly recommended for understanding how AI agents handle long-horizon tasks, and reminisces about learning to program in Ultima Online as a kid (and getting scammed).
More from coding & agent
- The recipe behind Claude's all-code video: add cliptude MCP to generate hour-long videos — drrickio · 2026-10-19
- Claude made an entire video in pure code: 2 prompts, 13 scenes, 2,148 frames, no video model — drrickio · 2026-10-19
- "The agent says done — then my real job starts": devs compare verification workflows — falafel777 · 2026-10-19
- lexfridman-cn open sourced: one Python command builds 506 static episode pages — vista8 · 2026-10-19
- Lex Fridman Chinese companion site goes live alongside its open-source repo — vista8 · 2026-10-19
- All 506 Lex Fridman podcast episodes converted to bilingual reads with local ASR, now open source — vista8 · 2026-10-19