Don't Worry About the Vase Podcast AI #152: Brought To You By The Torment Nexus
Jan 22, 2026
The podcast dives deep into the utility and failures of language models, highlighting the contrasts in performance. Innovations like Gemini's personalized AI integration are discussed, alongside concerns about deepfake technologies and their societal implications. The impact of AI on job disruption and rising economic expectations is debated, while safety concerns about AI systems are examined. The complexities of AI alignment and consciousness also spark intriguing discussions, offering both humor and critical insights.
AI Snips
Chapters
Books
Transcript
Episode notes
Chip Exports Could Multiply Chinese Compute
- Selling H200/MI3-like chips to China risks multiplying Chinese compute and narrowing the lead.
- Zvi and others liken it to selling nuclear-grade tech and warn against easing exports.
Celeb Takes Underestimate AI Progress
- Ben Affleck uses LLMs for brainstorming and appreciates their limits, but underestimates ongoing capability growth.
- Zvi argues models improved far more and cheaper than Ben suggests, countering views that progress is near its peak.
Conserve Expected Evidence In Debates
- Use the law of conservation of expected evidence: an argument's existence should update you slightly both ways.
- Zvi applies this to debates about AI consciousness and poor counterarguments.
