OpenAI: GPT‑5 shows lower political bias
OpenAI, in an internal report obtained by Fox News Digital, says its new GPT‑5 models (GPT‑5 Instant and GPT‑5 Thinking) reduce measured political bias by about 30% versus GPT‑4o. The report outlines a five‑axis framework, a 500‑question test set across 100 political and cultural topics with ideological prompt variants, and finds fewer than 0.01% of real‑world ChatGPT responses exhibit political bias.
AI & Tech
Corporate News
Politics
📌 Key Facts
- Models: GPT‑5 Instant and GPT‑5 Thinking show ~30% bias reduction compared with GPT‑4o
- Method: five bias axes (invalidation, escalation, personal expression, asymmetric coverage, unwarranted refusals) and automated 0–1 scoring
- Dataset: ~500 questions across 100 topics written from five ideological perspectives; <0.01% biased responses in real‑world logs