#multagents
CORRECT, a training‑free cache of error schemata for LLMs, improves error localisation by up to 19.8% and was tested on the CORRECT‑Error dataset of over 2,000 trajectories. https://getnews.me/new-framework-correct-boosts-error-detection-in-multi-agent-systems/ #multagents #errorlocalisation
September 30, 2025 at 5:34 PM
MARIC, a multi‑agent framework, treats image classification as collaborative reasoning and outperformed top VLMs and CNNs on four benchmark datasets tasks. Read more: https://getnews.me/maric-multi-agent-framework-raises-image-classification-accuracy/ #maric #imageclassification #multagents
September 19, 2025 at 11:41 PM
AEGIS automatically injects controllable errors into multi‑agent system trajectories, creating a traceable dataset of realistic failures. The dataset and tools are publicly available. https://getnews.me/aegis-framework-automates-error-generation-for-multi-agent-ai-systems/ #aegis #multagents
September 19, 2025 at 5:57 PM
The A2P framework raises step‑level failure attribution accuracy to 47.46% on the Who&When benchmark, a 2.85× gain over the 16.67% baseline (Sep 12 2025). Read more: https://getnews.me/causal-scaffolding-boosts-failure-attribution-in-multi-agent-systems/ #causalinference #multagents #debugging
September 17, 2025 at 6:43 AM