Zewen Liu

Text communication among LLMs can lead to significant output homogenization, challenging assumptions about diversity in multi-agent interactions.

Evaluator coupling significantly impacts measurement reliability, with low coupling leading to high noise levels that undermine evaluation accuracy.

Evaluator biases can significantly skew LLM agent behavior, but the new EPC protocol standardizes how we measure and compare these effects across different evaluators and time points.

Calibrating evaluator feedback can slash preference coupling in LLMs by nearly half, enhancing their decision-making integrity.

Random undersampling can inflate calibration error dramatically, making it a hidden threat in imbalanced classification tasks.

Compressing state-of-the-art image generation models by up to 75% without sacrificing quality could revolutionize resource efficiency in AI image synthesis.

Memory Contagion reveals that even perfect memory consolidation can't prevent the spread of evaluator bias, with implications for the integrity of LLM training processes.

Evaluator biases in multi-agent LLM systems can propagate significantly, but increasing evaluator committee size can reduce this contagion by over 72%.

A single evaluation strategy can dominate multimodal AI decision-making, leading to severe bias and strategy inversion across tasks.

Sustained self-improvement in LLM agents is achievable through a novel adaptive framework that outperforms traditional methods in dynamic task environments.