LinkedIn Presents.

I'm Rufus Griscombe, and this is the next big idea.

Today, the risks and rapture of superintelligence. Today's episode is a doozy.

Today's episode is a doozy.

I won't sugarcoat it.

Our guest today thinks there is a 70% probability that AI will achieve superintelligence

resulting in either human subjectation or extermination, the end of our story. And he thinks this could happen in the next 10 years.

Now, of course, there are many people who disagree. We'll get into that. Some think AI will never

achieve superintelligence. Others think it will take many decades. And that time will enable us to

build in the transparency

to keep AIs aligned.

That's a word you'll hear many times in this conversation, meaning working in collaboration

with humans to advance our interests.

But our guest today, Daniel Kokatelo, is worth taking seriously because he has an uncanny track record

of accurately predicting AI progress. In 2021, he published a scenario titled What 2026 looks like,

in which he laid out predictions for the next five years of AI progress. From the surge of advanced chatbots to US restrictions on the export of AI chips to China,

many of Daniel's predictions have come true, earning him a reputation as a reluctant profit.

After that success, he joined Open AI as a governance researcher only to depart in 2024 in a very public break over AI safety concerns.

Now free to speak his mind, Daniel has teamed up with other top analysts to produce AI 2027, a detailed forecast published a few weeks ago that provides a month-by-month roadmap of how

the next waves of AI development might unfold.

That's what we're going to talk about today.

But first, I think it's worth taking a moment to step back and think through what we're doing

...