Insights
What the podcasts say, counted.
We keep the full transcript of every episode we summarize, 806 so far. These pages count what the shows say in those transcripts. Each page shows example quotes with a link to the exact moment, and explains how we counted.
The biggest move right now
AGI Fading
AGI peaked in the four weeks to Aug 16, at 12 of 54 episodes (22%). It is at 5% now, well below its usual 13%.
Biggest moves
The terms whose share changed the most between the earlier weeks and the newest four, strongest evidence first. The tag says whether the change is big enough to call.
- ChatGPT No clear trend12% before, 4% now (7 of 157 episodes)
- Neoclouds No clear trend6% before, 11% now (17 of 157 episodes)
- Prediction markets No clear trend5% before, 10% now (15 of 157 episodes)
- Vibe coding No clear trend9% before, 4% now (7 of 157 episodes)
- Computer use No clear trend9% before, 5% now (8 of 157 episodes)
Who said it first
Shows in the same week are tied, and a later show may be reacting to the same news.
Astra. First heard in our archive, counting any mention: the week ending Sep 6, on All-In Podcast, Dwarkesh Podcast, How I AI, Lenny's Podcast and TBPN. 20VC with Harry Stebbings and Odd Lots each first said it 1 week later. 4 more shows first said it within 6 weeks of the first.
Predictions due next
Specific calls about the future, in the speaker's own words. Most have a deadline, and we grade each one against the evidence once the outcome is clear.
About these numbers. Transcripts are automatic captions, and the counts cover the shows we summarize, not every podcast. Trend numbers are plain counts. For predictions, a model writes a one-line summary from the speaker's exact words, and code checks those words against the transcript. The predictions page shows the words beside each call. Each page says how it was counted. The data is open: trends (JSON, CSV) and predictions (JSON).