News for hackers. Stuff that matters.

kirchoni

Re:Deep Dive into G-Eval: How LLMs Evaluate Themselves on Wednesday November 05, 2025 @10:50AM (#45824222)
Interesting 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.

Show HN: GhostCV – embed AI-readable text in your resume (without losing design)

https://www.ghost-cv.com/

I suspect a lot of resumes — especially the ones designed in Figma or exported from Canva — look great to humans but get completely butchered by resume parsers.Most of these systems don’t use vision or layout — they just extract raw text. And if the PDF is made of outlines, floating layers, or flattened design, there’s often nothing to extract So I built this thing: ghostCV - It takes your...

( Read More... | no comments | on HN )