Base AI models can judge their own output quality without training

Original: Self-Evaluation Is Already There: Eliciting Latent Judge Calibration in Base LLMs with Minimal Data

Source: arxiv.org ↗