The AI Fix The AI hype train, space data centers, and lifelike robot heads
29 snips
Oct 14, 2025 Discover the surprising truth about GPT-5's training data, which taps into adult website phrases. The hosts dive into the chaos of AI pranks that led Irish police to issue warnings. Jeff Bezos dreams of launching data centers on the moon, while a new ultra-lifelike robot head raises unsettling questions about technology's future. They also unpack why most AI pilots fail and the growing concerns around corporate America's AI hype amidst legal risks. Prepare for a thrilling ride through the world of AI's potential and problems!
AI Snips
Chapters
Transcript
Episode notes
Models Trained On Unexpected Sources
- Reverse engineering GPT-OSS revealed prominent tokens from adult and spammy sites in its token list.
- Training corpora can include unexpected, explicit sources that shape model behaviour.
AI Home Intruder Prank Sparks Police Warnings
- TikTok users generated AI images of homeless people to prank partners, causing police false alarms.
- Irish police warned the trend led to real deployments to hoax burglary reports.
Use Agent Protocols And Kits
- Use tools and protocols to standardise agent communication before large deployments.
- Leverage agent frameworks like MCP and Agent Kit to build and orchestrate agents efficiently.