Muss ich bei jedem KI-Update neu Prompten lernen?
Summary
In this 150th episode of the KI-Podcast, hosts Marie Kilk and Gregor Schmalzried answer listener questions about AI workflows, prompt adaptation for new models like GPT 5.6, and the differences between ChatGPT’s Chat and Work modes. They explore why AI struggles with tasks like generating accurate chessboard images or determining weekdays for obscure dates, attributing these errors to training data limitations and lack of logical reasoning. The episode also discusses using agent tools for complex web research and offers tips for better prompt engineering.
Topics discussed
Intro: Episode 150 and listener emails
Prompt robustness across model updates
ChatGPT Work vs. Chat: Features and agent behavior
Using AI agents for complex data extraction tasks
Why AI struggles with chessboard image generation
Why AI makes errors with days of the week
Outro: Rabbit R1 hack and AI-generated music
Listen ad-free on Castria