VettedGaps

OpenAI Platform & APIs

Marketplace · 186 pain cards
Sample

Sample from OpenAI Platform & APIs

1 of 186 pain cards, shown in full — score, summary and a real quote from the thread it came from.

21
LLM developers can't trust LLM-as-judge scores due to self-disagreement and drift

We rely on LLM judges to evaluate our agents, but the judges disagree with themselves—scoring the same case 5/5, then 2/5, then 5/5 within minutes.

An LLM judge disagrees with itself. I’ve had the same case, same everything, score 5/5, then 2/5, then 5/5 within fifteen minutes. So a single bad run is not evidence: before I look at execution I run the case a few times against the approved version and see how much it moves on its own.

Integrations & API
12 posts
Pro

185 more pains are available in Pro

The blurred cards below are placeholders. No locked title, quote, score or id is sent to the browser.

Page 1 of 8 · 186 pains · 25 per page

OpenAI Platform & APIs is a Pro marketplace

Create a free account, then upgrade to Pro to unlock all 186 cards in OpenAI Platform & APIs.

Create free account

Already have an account? Sign in