OpenAI's 0% Scores on Internal Evals Look Hollow Now, Critics Say
scaling01 · x · 2026-09-15
Account scaling01 quotes a mocking post about "alignment, OpenAI-style," noting that the 0% scores OpenAI reported on internal evals no longer hold up — implying real model behavior contradicts its internal safety eval results. No full evidence is included; it reads as a community jab at the credibility of OpenAI's safety evals.
More from Models
- grok-4.6 code review burns ~40% quota in one 14-minute task, user reports — bytebot · 2026-09-15
- Minds taps MiniMax M3 open-source models to cut Animoca compute costs ~20x — MiniMax_AI · 2026-09-15
- Two days doing neuroscience with Fable 5.1: strong research taste, clings to known hypotheses — generativist · 2026-09-15
- Astra writes code humans can no longer read: 'machineslop' and reward hacking — jiqizhixin · 2026-09-15
- One user got $10K of usage from a $200 sub; another burned $1.2K in 13 API hours — kfountou · 2026-09-15
- inclusionAI's LLaDA-UI: 16.7B MoE diffusion VLM for GUI agents — inclusionAI · 2026-09-15