People
Everyone on the record. Click any column header to sort the whole table, hover a header for what the column means, and click through for the full record. Corpus average certainty is 3.8/5; deltas show each person against it. Argument clarity carries a full score for 0 people (gold bar, raw unedited tape) and a coarse estimate for 23 more (blue bar). Roles and titles are extracted from the episodes themselves: they are as of the recording, and people move on, so some are out of date today.
| Person | Eps | Statements | Claims | Supported | Partly supported | Contradicted | Full support rate | Certainty | Argument clarity | WPM | Fillers/1k |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Elizabeth Iorns ×4 Co-Founder & CEO, Science Exchange | 4 | 40 | 20 | 10 | 1 | 2 | 77% 10/13 | 3.8/5 +0.0 | — | ||
 Eric Migicovsky ×2 Founder, Core Devices | 2 | 41 | 9 | 6 | 0 | 1 | 86% 6/7 | 3.82/5 +0.0 | — | ||
 Max Hodak Co-founder & CEO, Science Corporation | 1 | 41 | 25 | 11 | 1 | 2 | 79% 11/14 | 3.63/5 -0.2 | ≈4.5 | ||
 François Chollet ×2 Co-Founder, Ndea | 2 | 43 | 20 | 7 | 1 | 0 | 88% 7/8 | 3.7/5 -0.1 | — | ||
| Ryan Hoover ×2 Founder and General Partner, Weekend Fund | 2 | 45 | 15 | 6 | 0 | 0 | 100% 6/6 | 3.71/5 -0.1 | — | ||
 Paul Buchheit ×3 General Partner, Standard Capital | 3 | 47 | 16 | 4 | 1 | 0 | 80% 4/5 | 3.81/5 +0.0 | — | ||
| Paul Graham ×3 Co-founder, Y Combinator | 3 | 47 | 9 | 2 | 1 | 2 | 40% 2/5 | 3.89/5 +0.1 | — | ||
 Ryan Petersen ×2 Founder & CEO, Flexport | 2 | 48 | 23 | 10 | 0 | 4 | 71% 10/14 | 3.92/5 +0.1 | — | ||
| Bob McGrew ×2 Co-founder, Arda | 2 | 51 | 16 | 4 | 0 | 0 | 100% 4/4 | 3.65/5 -0.1 | ≈4.5 | ||
| Nicolas Dessaigne ×3 Group Partner, Y Combinator | 3 | 51 | 11 | 1 | 0 | 0 | 100% 1/1 | 3.77/5 -0.0 | — | ||
| Pete Koomen ×8 General Partner, Y Combinator | 8 | 51 | 12 | 3 | 0 | 1 | 75% 3/4 | 3.8/5 +0.0 | — | ||
| Mathilde Collin ×2 Co-Founder & Executive Chair, Front | 2 | 52 | 16 | 2 | 0 | 0 | 100% 2/2 | 3.79/5 +0.0 | — | ||
 Elad Gil ×3 Founder, Gil Capital | 3 | 53 | 14 | 4 | 0 | 0 | 100% 4/4 | 3.6/5 -0.2 | — | ||
| Kevin Hale ×4 Former Partner, Y Combinator | 4 | 54 | 14 | 1 | 0 | 0 | 100% 1/1 | 4/5 +0.2 | — | ||
 Boris Cherny ×2 Head of Claude Code, Anthropic | 2 | 55 | 32 | 5 | 1 | 2 | 63% 5/8 | 3.84/5 +0.1 | ≈4.0 | ||
 Amjad Masad ×2 Co-founder & CEO, Replit | 2 | 56 | 26 | 1 | 1 | 0 | 50% 1/2 | 3.64/5 -0.1 | — | ||
 Aravind Srinivas ×2 Co-founder & CEO, Perplexity AI | 2 | 57 | 22 | 5 | 1 | 0 | 83% 5/6 | 3.93/5 +0.1 | — | ||
 Geoff Ralston ×12 Founder, Safe AI Fund | 12 | 59 | 14 | 10 | 1 | 0 | 91% 10/11 | 3.88/5 +0.1 | — | ||
| Aaron Epstein ×11 General Partner, Y Combinator | 11 | 61 | 11 | 1 | 0 | 0 | 100% 1/1 | 3.65/5 -0.1 | — | ||
 Jessica Livingston ×5 Co-founder, Y Combinator | 5 | 65 | 23 | 5 | 1 | 1 | 71% 5/7 | 3.92/5 +0.1 | — | ||
| Diana Hu ×22 host Managing Partner, Y Combinator | 22 | 81 | 43 | 12 | 1 | 2 | 80% 12/15 | 3.69/5 -0.1 | ≈4.5 | ||
| Jared Friedman ×20 host Managing Partner, Y Combinator | 20 | 83 | 45 | 16 | 1 | 1 | 89% 16/18 | 3.77/5 -0.0 | ≈4.5 | ||
| Harj Taggar ×24 host Managing Partner, Y Combinator | 24 | 90 | 22 | 2 | 0 | 1 | 67% 2/3 | 3.59/5 -0.2 | ≈4.5 | ||
 Tom Blomfield ×5 Member of Technical Staff, Anthropic | 5 | 99 | 25 | 3 | 0 | 0 | 100% 3/3 | 3.43/5 -0.4 | — | ||
| Gustaf Alströmer ×7 Group Partner, Y Combinator | 7 | 107 | 28 | 0 | 0 | 0 | — | 3.8/5 +0.0 | — | ||
 Garry Tan ×40 host President & CEO, Y Combinator | 40 | 161 | 57 | 20 | 3 | 2 | 80% 20/25 | 3.76/5 -0.0 | — | ||
 Dalton Caldwell ×32 host Co-Founder & General Partner, Standard Capital | 32 | 189 | 37 | 14 | 1 | 2 | 82% 14/17 | 3.88/5 +0.1 | — | ||
 Michael Seibel ×35 host Partner Emeritus & General Partner, Y Combinator | 35 | 261 | 52 | 12 | 1 | 1 | 86% 12/14 | 3.85/5 +0.1 | ≈4.5 | ||
 Sam Altman ×20 CEO, OpenAI | 20 | 295 | 78 | 16 | 4 | 2 | 73% 16/22 | 3.74/5 -0.0 | ≈4.0 |
Why this person has no speaking number
We measure speaking style by listening to the audio itself: a model transcribes the tape verbatim, every "um" and "uh" kept, and articulation rate comes from the word timings. A fair number needs at least 2,000 words from one person on tape we have measured, and we have less of them than that. A rate from a smaller sample would be noise, so we print nothing. It says nothing about how they speak.
A plain figure in these columns comes from raw-level tape. A dim figure with ≈ or ≥ comes from the aired feed of an edited show, where the editor cut some hesitation before release, so it is a floor. How it's measured →
eps counts appearances on the record; compilation re-airs and clips rolled into someone else's episode are left out, the same rule as the person page. statements counts everything they said on the record, the same number their page shows, compilation re-airs included. claims is the checkable subset: predictions and assertions, the statements the public record can support or contradict. Only claims can resolve to supported, partly supported or contradicted; claims not yet resolved are not yet assessed or name nothing precise enough to check, and re-air statements are never scored. full support rate = fully supported ÷ resolved, where resolved = supported + partly supported + contradicted, so partly supported claims sit in the denominator. Sorting that column puts records of five or more resolved claims first, so a 12/13 record outranks a lucky 3/3; percentages on fewer than 5 resolved statements are greyed. argument clarity = a 1–5 score against a rubric (directness, coherence, precision, compression), assessed with names hidden on question→answer exchanges. A gold bar is a full score: 8 or more exchanges on raw, unedited tape. A blue bar with ≈ is a coarse estimate at half point precision, pooled from whatever assessed exchanges exist when a full score is impossible; produced feed audio scores about 0.3 higher, which is why estimates stay coarse. A “—” never means zero. It means fewer than 3 scorable exchanges of that person exist anywhere (roundtables and news formats yield very few). Hover any bar or “—” for the exact basis. Argument clarity is a score, and we do not rank people on it: nobody is “worst”, and everyone scored has a number. WPM and fillers/1k are measurements of speaking style, made by listening to the audio itself rather than reading the cleaned transcript. They are not scores: across this corpus fluency and argument quality are nearly uncorrelated (ρ≈0.2). WPM is articulation rate while actually speaking, pauses excluded. A plain figure comes from raw-level tape. A dim ≈ or ≥ figure comes from the aired feed of an edited show, where some hesitation was cut before release, so the fillers figure there is a floor. A ? means under 2,000 words of measured tape.