A.I. Expert Answers A.I. Questions From Twitter | Tech Support | WIRED

I remember how on my first day in middle school I walked into one of my classrooms for the first time, and above the door was a long poster that stretched most of the way across the room with the quote: "If the human brain were so simple we could understand it, then we would be so simple we couldn't." It has stuck with me ever since, for the past 20 or so years, as something I was confident in believing to be true. I think this week that confidence may have been truly shaken for the first time.

Yeah the scary AI scenario in the upcoming few years is indeed not "Robot goes red-eyed and becomes evil". It's "We let AI control critical systems, and then discover it's still too naïve, stubborn, gullible or exploitable for the task"

This deserves to be watched by everyone who is interested in AI

As a computer scientist, this is SO refreshing! Someone that is actually knowledgeable about a subject talking about it! Thank you, thank you, thank you!!!!

AI has the potential to be the biggest double-edged sword in all of human history. I have absolutely no faith that we’ll be wise enough to know how to wield it.

Wow… Gary is an extremely intelligent and well spoken person in his field!

As a person that works with AI, it's refreshing to see a normal and real take on this subject.

For all the hyped up videos about how we'll reach singularity in 5 years, this is a breath of fresh air. Thank you Gary for explaining this like it actually is!

"People are easily fooled" is a mood. and SO true.

I hope one of the effects of A.I. is that it actually strengthens human reliance on / trust in in-person interaction and connection. When AI generates more false information, intentionally or otherwise, true human-to-human connections will be the most valuable, trustworthy thing in the world.

This is the type of guy I want in charge of AI development He understands the issues. He isn't diluted. He isn't just here to make money. He's realistic about the applications and the benefits.

It does have the ability to internally represent the world and solve problems. Me:You have five objects 1:dinner plate 2:box of cereal 3:cookie jar 4:basketball 5:hockey puck ;How would you arrange these objects to stack them vertically for maximum stability? GPT4:In the context of stacking objects vertically for maximum stability, the basketball presents a challenge due to its round shape, which makes it difficult to place other objects on top of it without a risk of them sliding or rolling off. If you still want to include the basketball in the stack, you could try placing it on top of the stack, as follows: Dinner plate (1): Place the dinner plate flat on a stable surface with the concave side facing up. Box of cereal (2): Put the box of cereal on top of the dinner plate. Cookie jar (3): Place the cookie jar on top of the cereal box. Hockey puck (5): Put the hockey puck on top of the cookie jar. Basketball (4): Carefully balance the basketball on top of the hockey puck. However, please note that placing the basketball on top of the stack significantly increases the risk of the stack becoming unstable, as the round shape of the basketball is likely to cause it to roll or slide off. This arrangement is not ideal for maintaining maximum stability, and caution should be exercised when attempting to create such a stack.

It concerns me that there's a lot of conversation about things like AI reducing the number of human cashiers needed in stores, but very little about what actually happens to those people who are no longer needed. I think Universal Basic Income needs to develop at the same rate as AI being used in place of humans in the workplace, otherwise people will be left with reduced options while companies get to simply save money.

"Babies are like little scientists" best quote ever

What a great episode, interesting questions and thoughtful answers!

Bro…. The fact about the Furby. 🤯 I was so amazed as a 5 yr old

I think with the current status of ChatGTP, essays written exclusively by the software do vary between A-C grades, skewering towards the lower end depending on subject. However, starting off with ChatGTP and then fine tuning it can produce some amazing A level efforts. And it won't be much longer before A level essays and articles are pretty much 80% of the output on first attempt.

This was awesome. I love that he mentions multiple times the risks and undesirable possible outcomes of AI. I feel like this isn't talked about enough by AI experts.

This guy is great. Do another one with him!

Wired is really updated with the content and what's the hype even in tech. Gary has a degree in psychology as well, which actually makes this some informative comments. I wonder what's his comment or thoughts on image models such as Bluewillow or Dall-e.