Public transcript

Why Alexandr Wang was ALREADY over the AGI hype in 2019

By: This Week in Startups Clips
Length: 5 min
Language: English
Text from: the video's captions
Transcribed: 12 October 2026

Read the full transcript Open on YouTube

Notes in English.

Summary

This clip features hosts reflecting on a 2019 interview with Alexandr Wang, who argued that general AI timelines were overblown and that more compute alone would not produce AGI. Wang believed AGI would eventually happen but not soon, and that the simulated-evolution argument for AGI was effectively science fiction. The recording ends with an ad read for Uber AI Solutions.

Key points

The hosts note that Wang's view that more compute alone won't reach AGI contrasts with current heavy compute investments by Meta and xAI.
One host says the idea that enough chips in a room will figure things out without human coding always felt counterintuitive.
Wang believes general AI will happen eventually because humans are infinitely creative, but thinks timelines are very overblown.
Wang argues Moore's law is going to die and quantum computing is very far away, weakening the exponential compute argument.
He says infinite compute helps narrow AI, like mastering go, poker, or chess, but not general AI.
Wang calls the simulated-evolution argument for AGI the most plausible one, but says it is science fiction and not close to being validated.
The clip closes with an advertisement for Uber AI Solutions at uber.com/twist.

Questions it answers

00:00Introduction and framing

Why are the hosts revisiting this 2019 Alexandr Wang clip?

They want to highlight that Wang said more compute alone won't get us to general AI, which contrasts with the current approach of Meta and xAI investing heavily in compute power. One host says hearing a technologist agree made him feel less crazy about doubting the chips-alone assumption.

Meta and xAI have been investing heavily in compute power for more intelligent systems.
One host found it counterintuitive that enough chips alone would figure things out without human coding.

01:19Wang on AGI timelines

Does Alexandr Wang believe general AI will happen?

He believes it will happen in some sense, because humans are infinitely creative and most conceivable, non-physically-impossible technologies eventually occur. But he thinks the timelines people talk about are very overblown.

Wang believes AGI will happen eventually if humans survive.
He thinks common AGI timelines are very overblown.

01:52The compute argument

Why does Wang doubt that more compute leads to general AI?

He says Moore's law is going to die and quantum computing is very far away, weakening the exponential compute argument. He also argues it is unclear that even infinite compute would produce general AI, though it does help narrow AI like go, poker, and chess.

Moore's law is going to die and quantum computing is very far away.
Infinite compute helps narrow AI but may not produce general AI.
Mastering chess then go then impressionist painting is a different problem.

02:30Simulated evolution

What does Wang think of the simulated-evolution argument for AGI?

He calls it the most plausible argument, but says it is science fiction and we are not close to validating the hypothesis. Someone would still have to code and build those systems; it won't magically happen.

The simulated-evolution argument is the most plausible one to Wang.
He does not believe in general AI anytime soon.

03:37Sponsor read

What is advertised at the end of the clip?

The hosts advertise Uber AI Solutions, which sources, labels, evaluates, and scales high-quality data for startups, drawing on Uber's experience scaling rides, Uber Eats, and autonomous driving. Listeners can book a demo at uber.com/twist.

Uber AI Solutions offers data sourcing, labeling, evaluation, and scaling.
Book a demo at uber.com/twist.

Notes

Context from the Hosts

The hosts introduce a 2019 clip of Alexandr Wang, highlighting that he argued more compute alone will not get us to general AI.
They note this contrasts with today's approach of companies like Meta and xAI, which are investing heavily in compute (e.g., 20,000 Nvidia chips in one building).
One host admits not being a programmer but says the idea that enough chips will "figure it out for us" always felt counterintuitive, since humans historically had to design how software works.

Wang on Whether General AI Will Happen

Wang believes general AI will happen in some sense, because most technological things humans conceive of that aren't physically impossible will eventually occur if humans survive.
However, he thinks the timelines people discuss are very overblown.

Problems with the Compute Argument

A common argument is that Moore's law will keep going and provide exponential compute. Wang counters that Moore's law is going to die and quantum computing is very far away despite recent press releases.
He argues it is not even clear that infinite compute would produce general AI.
Infinite compute helps narrow AI: playing out scenarios in go, poker, or chess, or throwing agents into random video games as OpenAI does.
But taking a system that mastered chess and asking it to master go or impressionist painting is a very different problem.

The Simulated Evolution Argument

One vogue argument says enough compute could create artificial life by simulating evolution, starting from bacteria and growing a human-like species with a big brain.
Wang calls this the most plausible argument but says it is science fiction, and we are not close to being able to validate the hypothesis.
Someone would still have to code and program those systems; it won't magically happen.
Conclusion: Wang does not believe in general AI anytime soon.