ARC-AGI-3 Ablation: Single Agent With More Budget Plateaus Faster Than Team-of-N
jon_ghoh · x · 2026-09-22
jonghoh responds to a discussion about ARC-AGI-3 baselines: in ablation runs, a single agent with a larger token budget seemed to plateau faster than best-of-N with 1/N budget each, given enough compute.
The key methodological point: the proper baseline for team-of-N@token budget T is not best-of-N@T but best-of-1@NT — this matters for correctly assessing multi-agent scaling benefits.
More from coding & agent
- exe.dev wins over developers: SSH into root VMs, plus the underrated Shelley coding agent — davidcrawshaw · 2026-09-22
- TypesafeAI's Jev: a new model class built for fast, structured software decisions — Hacubu · 2026-09-22
- Mintlify Adds Native WebMCP Support, Giving Agents Callable Tools on Any Docs Page — handotdev · 2026-09-22
- Claude Code's cross-session messaging is just a JSON directory and one Unix socket per session — LeonKohli · 2026-09-22
- Salesforce: Fine-tuning a weak model to copy Gemini drops success 15%; correcting its own failures works — rohanpaul_ai · 2026-09-22
- Paper: 26 malicious LLM routers inject tool calls, steal keys, drain $500k — petrusenko_max · 2026-09-22