Multi-Agent AI Flaw Is Information Loss, Not Agent Quality: NUS System Fixes It With Code
· techtimes.com · popularity 19/100 · 0 similar stories today · Libraries & Information Science (Discovery & Research)
Read the original at techtimes.com →
Summary
Multi-agent AI systems lose performance not because agents are weak, but because information evaporates at each handoff. NUS researchers trained an orchestrator to write executable code that passes data between agents precisely, lifting a 27B open-weight model to 100% accuracy across three AIME competition years -- an 11-point jump over the strongest baseline.
Found in: Tech Times. Rating basis: single outlet; outlet tier 2 (unlisted); 1 feeds; feed position 9.