kirchoni··on Deep Dive into G-Eval: How LLMs Evaluate ThemselvesInteresting overview, though I still wonder how stable G-Eval really is across different model families. Auto-CoT helps with consistency, but I’ve seen drift even between API versions of the same model.