ChatGPT ARTICLE 5 September 2025

Why language models hallucinate

OpenAI’s new research explains why language models hallucinate. The findings show how improved evaluations can enhance AI reliability, honesty, and safety.

OpenAI’s new research explains why language models hallucinate. The findings show how improved evaluations can enhance AI reliability, honesty, and safety.

Back to ChatGPT updates
Save

More from ChatGPT

All updates
ChatGPT
18 Jun 2026

New usage analytics and updated spend controls for enterprises

Open article →
ChatGPT
18 Jun 2026

Improving health intelligence in ChatGPT

Open article →
ChatGPT
18 Jun 2026

Using AI to help physicians diagnose rare genetic diseases affecting children

Open article →
ChatGPT
17 Jun 2026

A near-autonomous AI chemist improves a challenging reaction in medicinal chemistry

Open article →

Comments

Sign in or join free to leave a comment.

No comments yet. Be the first.

Gemini komt eraan

Cookies and external media

The Next Input keeps optional media off until you say yes.

We use the site with necessary technical storage only, and block external media such as YouTube videos by default. If you allow external media, those services can place their own cookies and receive technical data such as your IP address.

Always on

Core site functions, security, and anonymous visit tracking. No IP address is stored in readable form — we use a daily rotating hash. The tni_visitor cookie remembers your article reactions so you can only vote once per article.

Optional

External embeds such as YouTube videos and other third-party players. These services may place their own cookies and receive your IP address.