Stitching episodes across context resets: length-adaptive GAE preserves early credit
anirudhg9119 · x · 2026-10-14
Engineering details for long-horizon episodes: long trajectories don't fit in one context, and auto-compaction splits them at artificial boundaries — treating those as terminal biases credit. MIRA resets context at every decision but stitches meta-reasoning decisions back into one logical episode across all iterations; only true termination stops bootstrapping, so credit flows as if the episode were never split. Advantages use length-adaptive GAE so the earliest decisions in a long episode keep most of their weight on the realized return.
More from coding & agent
- TV noise crashes voice agents from 38.7% to 8.6%; full duplex is the fix — SIGKITTEN · 2026-10-14
- TypeSafe AI's Jev Claims 100x Faster Decisions, Can Replace 9 LLM Call Types — Arindam_1729 · 2026-10-14
- Gemini 4 Argon High tested with 'Horse prompt' via OpenCode V2 and Antigravity agent — ZealousidealTown1974 · 2026-10-14
- Buoy benchmark: Claude runs apps 6x faster and cheaper with MCP tools than none — LovesWorkin · 2026-10-14
- LMArena Launches Code Arena for WebDev Model Benchmarking — arena · 2026-10-14
- msgblast: Open-Source Mac App to Prompt Multiple AI Agents at Once — msg · 2026-10-14