ByteDance Paper: Keeping Full History Cuts Costs by 19% vs. Compression
AGI Hunt · wechat · 2026-08-31
ByteDance's Chain-of-Experience paper investigates how AI agents should handle past attempts during iterative solving. Contrary to intuition that compression saves context, experiments show retaining the full history with feedback signals achieves a 71.0% average accuracy across 8 models and 6 benchmarks, outperforming non-iterative solving (66.8%). Despite longer context windows, faster convergence reduces total API costs by 19%. The study suggests that feedback on "why it went wrong" holds more value than compressed result summaries, validating a "less is more" context strategy.
More from coding & agent
- Reflection's Beam, a 501B-parameter open MoE model with 1M context, lands on Hugging Face under Apache 2.0 — QGallouedec · 2026-10-16
- 'LLM code is sloppy' rebutted with real ffmpeg code: identical if/else branches in a top library — kingslyj · 2026-10-16
- Office-editing tool calls need file, worktree and unit: lessons from dsh-univer-office — FunDistribution6537 · 2026-10-16
- Architect MD workflow: Gemini one-shot outputs that beat Opus, via 5-lens plan critique — oikk01 · 2026-10-16
- Prove a coding agent actually ran the tests: check artifacts, not approvals — ComprehensiveShake76 · 2026-10-16
- Building an acceptance harness for AI spreadsheet edits: allowlists, recalc checks, idempotency — ukdicha-modakkk · 2026-10-16