This Day in AI Podcast GPT-5.2 Can't Identify a Serial Killer & Was The Year of Agents A Lie? EP99.28-5.2
253 snips
Dec 12, 2025 In the latest discussion, GPT-5.2 is underwhelming, failing to identify a convicted serial killer while its competitors shine. The hosts critique the overhyped 'Year of Agents' that never delivered, and why many users are switching to alternative AI models like Grok and Gemini. Disney's hefty investment to use Mickey Mouse in a new project also sparks conversation. Plus, tune in for a hilarious diss track that showcases the model's shortcomings while maintaining its swagger.
AI Snips
Chapters
Transcript
Episode notes
Barber Switched To Grok For Speed
- Michael's barber switched from OpenAI to Grok because it felt faster and less intrusive.
- He wanted direct answers and preferred Grok's speed and consistent responses.
Agents Need System-Level Work, Not Just Models
- Agentic workflows are possible but require more software and process work than new base models.
- Chris says the AI system layer, not model advances, is the main missing piece.
The Ivan Milat Vision Test Failure
- The hosts ran the 'Ivan Milat' vision test and found GPT-5.2 hedged despite on-image text identifying the serial killer.
- Gemini and Claude identified him and deemed him untrustworthy while GPT-5.2 refused to commit.