Xin Qiu

Catastrophic forgetting in LLM fine-tuning with Evolution Strategies isn't irreversible or unique to ES, and can be largely avoided with a simple regularization technique.

LLMs actually *do* improve time series forecasting, especially for cross-domain generalization, overturning prior doubts with a massive 8-billion observation study.

Fine-tune quantized LLMs without backpropagation using QES, a method that achieves high-precision results at low-precision cost.