Menu
Sign In Search Podcasts Charts People & Topics Add Podcast API Pricing

Ege Erdil

๐Ÿ‘ค Speaker
529 total appearances

Appearances Over Time

Podcast Appearances

Dwarkesh Podcast
AGI is Still 30 Years Away โ€” Ege Erdil & Tamay Besiroglu

Why would humans be in the loop there?

Dwarkesh Podcast
AGI is Still 30 Years Away โ€” Ege Erdil & Tamay Besiroglu

And also it's less valuable.

Dwarkesh Podcast
AGI is Still 30 Years Away โ€” Ege Erdil & Tamay Besiroglu

I mean, it takes time, but

Dwarkesh Podcast
AGI is Still 30 Years Away โ€” Ege Erdil & Tamay Besiroglu

it is often quite fast.

Dwarkesh Podcast
AGI is Still 30 Years Away โ€” Ege Erdil & Tamay Besiroglu

In fact, ChatGPT grew extremely fast.

Dwarkesh Podcast
AGI is Still 30 Years Away โ€” Ege Erdil & Tamay Besiroglu

And if that's not an efficient way to- Yeah, you're equivalent of them.

Dwarkesh Podcast
AGI is Still 30 Years Away โ€” Ege Erdil & Tamay Besiroglu

So it is worth thinking about just, like, why one might expect...

Dwarkesh Podcast
AGI is Still 30 Years Away โ€” Ege Erdil & Tamay Besiroglu

Yeah.

Dwarkesh Podcast
AGI is Still 30 Years Away โ€” Ege Erdil & Tamay Besiroglu

Central planning to be slightly better in this world.

Dwarkesh Podcast
AGI is Still 30 Years Away โ€” Ege Erdil & Tamay Besiroglu

Right.

Dwarkesh Podcast
AGI is Still 30 Years Away โ€” Ege Erdil & Tamay Besiroglu

So one consideration is just communication bandwidth being potentially much, much greater than it is today.

Dwarkesh Podcast
AGI is Still 30 Years Away โ€” Ege Erdil & Tamay Besiroglu

And then like in the current world, the information gathering and the information processing are like co-located.

Dwarkesh Podcast
AGI is Still 30 Years Away โ€” Ege Erdil & Tamay Besiroglu

Like humans observe and also process what they observe.

Dwarkesh Podcast
AGI is Still 30 Years Away โ€” Ege Erdil & Tamay Besiroglu

Right.

Dwarkesh Podcast
AGI is Still 30 Years Away โ€” Ege Erdil & Tamay Besiroglu

In an AI world, you can disaggregate that.

Dwarkesh Podcast
AGI is Still 30 Years Away โ€” Ege Erdil & Tamay Besiroglu

That's actually a really interesting point, yeah.

Dwarkesh Podcast
AGI is Still 30 Years Away โ€” Ege Erdil & Tamay Besiroglu

So you can have the sensors and not do much processing, but just collect and then process centrally.

Dwarkesh Podcast
AGI is Still 30 Years Away โ€” Ege Erdil & Tamay Besiroglu

And that processing centrally might make sense for a bunch of reasons.

Dwarkesh Podcast
AGI is Still 30 Years Away โ€” Ege Erdil & Tamay Besiroglu

And you might get economies of scale from having more GPUs that produce better models and also be able to

Dwarkesh Podcast
AGI is Still 30 Years Away โ€” Ege Erdil & Tamay Besiroglu

Think more deeply about what it's seeing.