Episode Details
Back to EpisodesGemini Detects Google AI Media. Agent Skills Go Cross Platform. Guidance Rescues Untrainable Nets. FrontierScience Reveals Gaps.
Published 10 months ago
Description
You can now verify Google AI-generated videos in the Gemini app.Anthropic publishes Agent Skills as an open standard for AI platforms.Guided learning lets “untrainable” neural networks realize their potentialGPT-5.2 dominates OpenAI's new science test, but fails at complex research tasks.
The AI news for December 19th, 2025
--- This episode is sponsored by ---
Find our more about our today's sponsor Pickert at pickert.de.
---
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:
You can now verify Google AI-generated videos in the Gemini app.
Source: https://blog.google/technology/ai/verify-google-ai-videos-gemini-app/
Why did we choose this article?
Practical content-provenance tool from Google — useful for anyone worried about deepfakes or edited videos. Signals wider industry emphasis on built-in transparency and gives a concrete step (Gemini app) to verify AI-generated video.
Anthropic publishes Agent Skills as an open standard for AI platforms.
Source: https://the-decoder.de/anthropic-veroeffentlicht-agent-skills-als-offenen-standard-fuer-ki-plattformen/
Why did we choose this article?
A potential interoperability milestone — an open 'Agent Skills' spec could let developers reuse capabilities across multiple AI platforms, lowering lock-in and accelerating composable agent development.
Guided learning lets “untrainable” neural networks realize their potential
Source: https://news.mit.edu/2025/guided-learning-lets-untrainable-neural-networks-realize-their-potential-1218
Why did we choose this article?
Research with practical training implications — shows a way to rescue architectures previously deemed 'untrainable' by using guidance from another network, which could reduce wasted compute and expand usable model families.
GPT-5.2 dominates OpenAI's new science test, but fails at complex research tasks.
Source: https://the-decoder.de/gpt-5-2-dominiert-openais-neuen-wissenschafts-test-scheitert-aber-an-komplexen-forschungsaufgaben/
Why did we choose this article?
New benchmark (FrontierScience) that stresses models at Olympiad/research level — valuable for evaluating where top models like GPT-5.2 excel and importantly, where they still fail on complex research tasks.
Do you have any questions, comments, or suggestions for improvement? We welcome your feedback at podcast@pickert.de.
--- This episode is sponsored by ---
Find our more about our today's sponsor Pickert at pickert.de.
---
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:
You can now verify Google AI-generated videos in the Gemini app.
Source: https://blog.google/technology/ai/verify-google-ai-videos-gemini-app/
Why did we choose this article?
Practical content-provenance tool from Google — useful for anyone worried about deepfakes or edited videos. Signals wider industry emphasis on built-in transparency and gives a concrete step (Gemini app) to verify AI-generated video.
Anthropic publishes Agent Skills as an open standard for AI platforms.
Source: https://the-decoder.de/anthropic-veroeffentlicht-agent-skills-als-offenen-standard-fuer-ki-plattformen/
Why did we choose this article?
A potential interoperability milestone — an open 'Agent Skills' spec could let developers reuse capabilities across multiple AI platforms, lowering lock-in and accelerating composable agent development.
Guided learning lets “untrainable” neural networks realize their potential
Source: https://news.mit.edu/2025/guided-learning-lets-untrainable-neural-networks-realize-their-potential-1218
Why did we choose this article?
Research with practical training implications — shows a way to rescue architectures previously deemed 'untrainable' by using guidance from another network, which could reduce wasted compute and expand usable model families.
GPT-5.2 dominates OpenAI's new science test, but fails at complex research tasks.
Source: https://the-decoder.de/gpt-5-2-dominiert-openais-neuen-wissenschafts-test-scheitert-aber-an-komplexen-forschungsaufgaben/
Why did we choose this article?
New benchmark (FrontierScience) that stresses models at Olympiad/research level — valuable for evaluating where top models like GPT-5.2 excel and importantly, where they still fail on complex research tasks.
Do you have any questions, comments, or suggestions for improvement? We welcome your feedback at podcast@pickert.de.