The Pulse
4,971 transcripts, every word counted. What Silicon Valley talks about, year by year, and who said it on this show first.
The interview, measured
Computed from every diarized word: how much the hosts talk, how hard the questions press.
Share of all words spoken by the host. It peaked at 34% in 2019; the guest gets more of the mic every year since.
The pace of interrogation: pure math over every question mark in the transcripts.
How often claims get contested instead of accepted: from 1.6/10 in 2015 to 4.6/10. The interview got harder.
Drilling into vague answers instead of moving to the next question on the list.
Setups, compliments and promotional prompts as a share of questions: 69% in 2015 → 41% now. The bigger the show got, the fewer of them, except at the top of the prominence scale: Star Treatment →
Episodes with full speaker diarization. These feed the charts around this one.
pushback = contests claims · follow up = drills into vague answers instead of moving on · setup questions = setups, compliments and promotional prompts as a share of questions. Rated by an LLM with a published rubric; talk share and questions are pure math.
The show itself
Every processed episode by year published. Bars count them, the gold line tracks the median runtime. The show got longer as it got bigger.
The edit index: how produced the show sounds, year by year
The same guests, measured two ways. On the podcast feed's produced audio the "um"s and false starts were edited out in post production. For years the median guest shows ≈0.7 um/uh per 1,000 words, a rate no spontaneous human achieves. On raw YouTube recordings the median is ≈10.8. Of 132 guests we've heard on both, 114 show more filler raw (median +17.4/1k). This is why the speaking style measurements only use unedited recordings.
2017–2021 sits at ≈0. Credit the editing suite, not superhuman fluency. The early shows from the Skype era (2015–16) went out barely edited.
Same show, same years, unedited source: how people actually talk. Both charts share one scale.