Four AI experts from Diary of a CEO, cut into one 23-minute argument

TL;DR. Four people who have never shared a stage sat down separately with the same host and were asked a version of the same question. This mix takes 6 hours 55 minutes of those conversations and cuts them to 22 minutes 56 seconds, in an order none of them recorded in. It isn't a highlights reel. It's an argument, assembled from four sources, and it ends somewhere none of the four episodes ends on its own. Built with Segue, which chains clips from videos on one channel into a single link.

A best-of reel stacks good moments and lets them accumulate. This does something else. Each guest is doing a different job in a sequence, and the sequence is the point.

The four guests

All four are uploads on The Diary Of A CEO. Listed by guest, because the channel rotates episode titles.

• Dr. Roman Yampolskiy, an associate professor of computer science working on AI safety. 1:27:38, 20.6M views.
• Geoffrey Hinton, the Nobel laureate usually introduced as the godfather of AI. 1:30:07, 13.7M views.
• Mo Gawdat, former chief business officer of Google X. 1:56:31, 11.6M views.
• Daniel Kokotajlo, a former OpenAI researcher who left the company, and lead author of the AI 2027 scenario. 2:00:50, 7.5M views.

That's 6:55:06 of interview and 53.5M views between them. The mix keeps 5.5% of the runtime. (Figures as of August 4, 2026.)

How the argument is built

The first four clips are one statement each, back to back. Yampolskiy on unemployment: not 10%, "but 99%." Hinton, asked what to tell people about their careers: "Train to be a plumber." Gawdat: "It is beyond an emergency. It's the biggest thing we need to do today. It's bigger than climate change." Kokotajlo on the open secret in his industry, which is a 70% chance of something like human extinction.

Four voices, two minutes fifty-three, no context offered. Then the longer clips do the work: jobs, and the reflex people have when told theirs is going away. What safety failure actually looks like, and the questions Yampolskiy thinks you should put to the people building this. Hinton on why Ilya Sutskever left OpenAI. Gawdat on why he thinks there's a sentience to these systems.

Watching it in order, the effect isn't agreement exactly. The four don't share a thesis. What they share is a direction of travel, and hearing that from four people who were never in the same room is a different experience from hearing it once.

The move that makes it work

Here's the part worth stealing.

The clip that opens the mix and the clip that closes it come from the same ninety seconds of the same episode. Kokotajlo's extinction line runs 0:00 to 0:17. The clip that ends the mix runs 0:18 to 1:38. They are adjacent on the source video. In the mix they sit twenty-three minutes apart, at opposite ends.

The claim opens the argument. What the claim costs closes it. Same tape, pulled apart, doing two different jobs.

You can only do that if in and out points are cheap to set and clips are free to reorder. It's an editing decision, not a technical one, and it's the single clearest reason to build something like this out of precise clip boundaries rather than by trimming a file.

Where it lands

The two closing clips come from different episodes and different guests, and they arrive at the same place.

Neither is really about the specific decision each guest describes. Gawdat gets at it directly when he refuses to let the subject stay narrow, listing the economic and the geopolitical and the climate alongside the technology, and then naming what all of it has in common: "This is the depth of uncertainty."

That's the register the ending works in. Not a prediction, and not advice. A question about how much exposure you're willing to carry, for yourself and the people close to you, when four people with this much proximity to the work will not tell you it's going to be fine. The mix stops there, which is a better ending than any one episode gets, because no single episode has to sit next to another one making the same turn.

How this mix was built

Paste each episode, mark the in and out of the moments you want, then chain them into one link. No download, no re-upload, no timeline to fight. Every clip plays from the real uploads, so the views still land with the channel.

Chaining clips across four different videos is the part that needs Segue Pro, and it's scoped to videos on a single channel, which these four episodes are. The free tier covers up to 3 clips from one video, with no account. For the workflow itself, the guide to clipping podcasts from YouTube walks through it start to finish.

Quotes above are transcribed from the episodes and lightly punctuated for reading. Every one sits at a timestamp you can check.