Gary Marcus on AI's hallucination problem

AI expert Gary Marcus joins Andrew to cut through the noise and hype. Gary and Andrew talk about AI's difficulty with causal reasoning, what reasonable AI regulation looks like, stock market manipulation, 2024 election disinformation, romance scams, and driverless cars.

00:00 - Intro
00:57 - From Professor to Business Owner
07:47 - Global Threats of A.I.
14:09 - Do We Need an A.I.-Dedicated Federal Agency?
21:48 - Is Intelligent A.I. Imminent?
25:18 - Is ChatGPT Really a Money Maker?
30:27 - Like and Subscribe
30:38 - The Best Use Cases for A.I.
37:53 - Should We Require Licenses for A.I. Models?

----

Subscribe to Forward:

[emerraldx]

Andrew, I think that person in government thinking about AI policy options should be you. You’ve got the perfect background combination of practical modern connections across a lot of affected areas and UBI is the ultimate end goal solution after all the handrails are put in place. 🥰🧢🥰

[player]

Thank you for still trying to push your ideas into our democracy. I still believe you can be a big force for good change in our country! Thx for having these kinds of talks.

[vondoromal]

Humanity is better with you, Andrew Yang, connected!!!

[synthesis.podcast]

"I just don't see how the 2024 election isn't gonna be a *!#^$show."

Yuppers.

[TheGreatPizzaMasterpiece]

Shout to Steve Pinker. Living legend. When folks were lambasting him for writing a book on rationality I said “this is actually necessary, especially if AI agents will be trained on a human corpus and want a clear definition of the rational thought process.”

[TheGreatPizzaMasterpiece]

I think it was Dick Durbin who recognized that Senators were not nimble enough to keep up with the pace of technology, nor should it be their expertise. Whomever it was, great observation. The hearing as a whole was excellent. I hope the Senate spends more time debating versus voting with their phones in hand. It’s an opportunity for congress to work well.

[adurpandya]

Glad someone addressed the combinatorial reasoning issues. What about quantum computing though?

[haroldpierre]

I believe that before we consider regulations for consumer AI platforms, it's crucial to first establish laws governing military-grade AI. This ensures a balanced approach to AI regulation, prioritizing areas with potentially greater risks.

[hanxor]

Yang - I've been a supporter since seeing you in 2019 - as someone who was in political media marketing, it hurts to see your brand failing so badly. The engagement level is so off due to ZOOM CALL style interviews that truly don't touch pressing issues or provide true call to action for supporters. What is going on and who is running your org?

[AquarianSoulTimeTraveler]

It is said that children learn to speak in reverse first before they speak regularly and if you record what they say and play it in reverse you can hear them talking... Take this information with a grain of salt I haven't actually investigated this information. Our brains do flip everything we see...

[KungFuChess]

The only thing that will keep AI in check is keeping it free open source.

[GrumpDog]

AI's "hallucination problem" is only a short term problem of the current stage of progress. In a year or two, it'll be a mostly solved issue. So I think we should be focusing on the bigger implications of AI, without worrying about what's currently holding it back. Eventually the'll be overcome, and the implications on the workforce will still be huge.

[MoreLifePlease]

AIs hallucinate, huh? they're a lot like us humans already then?

[TheGreatPizzaMasterpiece]

TTRC is a better name: Trust in Technology Research Center. (or Resource Center... either or, whatever fits your mission better) Because you’re undoubtedly going to be addressing technologies which don’t necessarily fit under the category of AI. Doesn’t matter that it’s not catchy, just don’t give it a crappy logo that looks like some northwest Luddite movement, lol. Give it a bold and serious logo like NATO has, but combine it with the lyrical humanism that the CFR (Council on Foreign Relations) has. The combination of blue shades in that logo are a good start. I wish you luck.

[pencilcheck]

let's see what we can do, we need a leash but we also don't want to turn it into a handcuff.

[AquarianSoulTimeTraveler]

AI hallucinating is fine as long as it's just the skins of the 3D textures... I could take acid and be hallucinating and still able to drive a car... it's been physical sizes and distances our change is when the problem begins at least with self-driving cars.

[stephenharperisgay]

Yo Yang, numbers are rough and presidency is off the table but... keep focusing the conversation on the near future. It's a thankless job but just know I see what you're up to.

[e.d.]

sorry for the tangent but.... does New Yorker really only pay $250 an essay ??!

[TheGreatPizzaMasterpiece]

Is China’s data corpus as good as that of the U.S.? I mean is it truly high quality, not just high quantity.

[blackmage]

23:53
Because people always reason, stick to facts and don't make stuff up. /s