Time-aware RL (RPG) beats vanilla RL by 85.7% on modded NanoGPT speedrun
sherryyangML · x · 2026-10-14
Researchers discuss applying time-aware RL (RPG) to autoresearch and MLE agents:
- On the modded-nanoGPT speedrun, RPG beats vanilla RL by 85.7% at the final budget
- Reward-rate RL achieves lower pretraining loss faster than vanilla RL
- Core idea: choosing the next experiment means allocating a limited research budget — agents should weigh an experiment's potential payoff against its runtime
- With autoresearch gaining momentum (Karpathy, Quoc Le, Jeff Dean), time-aware RL could be a key ingredient
Paper link in the original thread.
Related event: Time-Aware RL Method RPG Boosts Efficiency by 85.7% Over Vanilla RL(2 posts)→
More from coding & agent
- Creator builds custom GoPro ingestion tool with Codex: 4-camera array with WiFi trickle ingest — OpenAIDevs · 2026-10-15
- Claude Code 2.1.300 cuts system prompt tokens by 58.5%, drops forced question-asking directive — ClaudeCodeLog · 2026-10-15
- Claude Code 2.1.300 ships 88 CLI changes, drops AskUserQuestion directive from system prompt — ClaudeCodeLog · 2026-10-15
- Nietzsche's 'Last Men' meet coding agents: losing open technical discourse to machine intelligence — akbirthko · 2026-10-15
- Tahoe AI opens Tara, a one-prompt plug-in connecting agents to its Tahoe-100M biology datasets — anshulkundaje · 2026-10-15
- sgrep cuts coding agent LLM costs 21.3% on SWE-bench versus ripgrep, fully local in Rust — udmrzn · 2026-10-15