A very interesting paper with insights into understanding when and why synthetic data (although imperfect and biased) can boost the performance of statistical inference!! 📈

LLM self-improvement has critical implications in synthetic data, post-training and test-time inference. To understand LLMs' true capability of self-improvement, we perform large-scale experiments with multiple families of LLMs, tasks and mechanisms. Here is what we found: (1/9)

✋✋

Could I be added as well? I’ve worked a bit on uncertainty quantification in LLMs and BNNs 😃