Uncertainty-Aware Answer Selection Boosts Multi-LLM Reasoning Performance
Calibrated log‑likelihood scoring boosts multi‑LLM answer selection, yielding around 4%‑5% accuracy gains on GSM8K, MMLU and ARC benchmarks without extra checks. Read more: getnews.me/uncertainty-aware-answer... #multillm #uncertaintyaware
0
0
0
0