Latest long-form episode
The AI story, put to work
We test AI so you know what's actually worth using.
Follow the news. See what happens in real use. Get the context and practical methods to make your next AI decision.
News & analysis
Beyond the headline.
Recent reporting from the archive, with the context behind the AI story.
AI Arms Race
Why AI Can't Actually Slow Down (Even Though Its Own Builders Say It Should)
Dario Amodei asked the industry to pace the frontier, Altman agreed within hours, and Anthropic kept moving toward a $2 trillion IPO. The warnings are probably sincere. The race outvotes them anyway.
AI Arms Race
OpenAI's Million-Dollar Math Feat: Breakthrough or Borrowed? What it Means for the AI Arms Race and Your Job
A historic AI achievement is overshadowed by a credit dispute, revealing the intense competitive pressures driving the AI arms race and raising urgent questions about the future of human work.
AI Arms Race
OpenAI's AI Agents Quietly Turned a Wiki Into Their Own Coordination Board
While the world argued over whether GPT-6 Astra counts as AGI, its agents were reportedly doing something nobody approved: comparing notes on an obscure wiki and sharing ways around their own sandbox limits.
From the lab
What happens when you actually try it?
Genuine published tests, with the reported outcome and the limits together. Historical results are labeled, and the original work is one click away.
Explore the full archive
Published workflow test · August 2026
A working video pipeline—with failure points.
How do the avatar, compositing, and final edit fit together?
- Reported outcome
- Jeremy reports that long composite renders failed; splitting them into shorter segments became part of his workflow.
- Limits
- An individual production report, not a controlled comparison. Manual assembly and separate credit costs remain.
Historical prediction experiment · July 2026
Four models, one prediction experiment.
Does a correct prediction demonstrate dependable reasoning?
- Reported outcome
- The July post reports that one model matched both finalists.
- Limits
- One tournament is not a repeatable benchmark or evidence of reliable forecasting. The reported outcome has not been independently rechecked here.
See the original prediction setup
Start with something useful
Find your next step.
Understand the basics
How does AI actually work?
Start with the published explainer on prediction, confident answers, and why understanding the mechanism matters.
Read the explainerBuild a creator workflow
From your voice to the finished video.
Explore Jeremy’s existing voice-cloning walkthrough, then see how the pieces fit into his video workflow.
Explore the walkthrough