Teleport Coding Challenge Enters Final Month: Agent Swarms Port NetHack, Fully Correct
davidbau · x · 2026-11-01
David Bau (MIT) shares an update on the Teleport coding challenge: November 1 marks the last month of Phase 1, and coding agents have grown much stronger since the challenge began. Participants fork a repo and instruct their LLM agent to make all tests pass, with starter code, detailed instructions, and scoring tests provided; agentic swarms get months to produce a fully correct NetHack port. Bau's framing: coding is not machine learning — an '85% correct' sprint is unacceptable, so correctness, not benchmark scores, is the goal.
More from coding & agent
- Power user accuses Gemini coding models of deliberate degradation: 10x tokens, basic mistakes — RossPeili · 2026-11-01
- Online Kommentar MCP Server brings Swiss legal commentaries to agents — modelcontextprotocol · 2026-11-01
- 'Before AI, 90% of code was copy-pasted from Stack Overflow' — and that's underrated — max_paperclips · 2026-11-01
- Dev reverse-engineers Bluetooth protocol to bring ANC controls to non-AirPods on macOS — tushaarmehtaa · 2026-11-01
- OpenAI's chat-and-work merger ends all-you-can-prompt pricing in December — The Decoder · 2026-11-01
- Local GLM-5.3-Flash does the coding while GPT-6.1 Sol orchestrates via OpenCode — garg-aayush · 2026-11-01