Benchmarks Built on Sand. ChatGPT Chats Exposed in Search. Omnilingual ASR Expands Coverage. BSI Warns on Prompt Attacks.

Show notes

The AI news for November 11th, 2025

--- This episode is sponsored by ---

Find our more about our today's sponsor Airia at airia.com.

---

Would you like to create your own AI-generated and 100% automated podcast on your chosen topic? --> Reach out to us, and we’ll make it happen.

Here are the details of the day's selected top stories:

The big AI lie?
Source: https://www.all-ai.de/news/topbeitraege/ki-luege-studie
Why did we choose this article?
A rigorous meta-study that questions the validity of most AI benchmarks — essential for anyone who relies on published model comparisons or safety claims. Practical takeaway: treat benchmark rankings skeptically and push for better evaluation standards.

Broken trust: This is how Google reads your ChatGPT conversations.
Source: https://www.all-ai.de/news/top-news24/chatgpt-google-problem
Why did we choose this article?
A concrete privacy incident with immediate implications for users and deployers of ChatGPT-like systems. Important for risk assessment: shows how integrations (e.g., web search) can leak sensitive prompts into third-party analytics.

Meta's new AI system understands and transcribes spoken language in more than 1,600 languages.
Source: https://the-decoder.de/metas-neues-ki-system-versteht-und-transkribiert-gesprochene-sprache-in-mehr-als-1-600-sprachen/
Why did we choose this article?
A potentially transformative speech model that covers 1,600+ languages including many low-resource ones. Relevance: impacts accessibility, localization, and research on few-shot learning for new languages — watch for dataset and evaluation details.

BSI warns: Even leading AI providers fail to protect against prompt attacks.
Source: https://the-decoder.de/bsi-warnt-selbst-fuehrende-ki-anbieter-scheitern-beim-schutz-vor-prompt-attacken/
Why did we choose this article?
Official security warning from Germany's BSI about evasion/prompt attacks — a practical alert for developers, security teams, and regulators. Takeaway: defensive measures around prompt filtering and model hardening remain immature; plan accordingly.

Do you have any questions, comments, or suggestions for improvement? We welcome your feedback at podcast@pickert.de.