Sam Altman on the barriers to Artificial General Intelligence

[@apertureinfog]

He’s talking about abductive reasoning, which we haven’t even begun to model in code

[@davidrflaing]

bruh i never even heard of abductive reasoning until you mentioned. TIL

[@Elurin]

Really?? Have you ever just tried googling "Abductive logic programming" ?? Seems like some people are making a start at it.

[@assgoblin3981]

the point is that we largely skipped the need for modeling this manually in code through use of neural networks. The networks do a sort of reasoning even without the explicit steps being laid out like it normally would in an algorithm. Which is why scaling is so important at this point. Did we really figure it out?

[@LeoStaley]

It needs recursion within its own responses. To not simply select a word based on what should be next, but what it knows will come later.

[@2010RSHACKS]

This is already being done with recursive autoencoder models such as Google’s PaLM and by chaining prompts with an agent for reasoning. The issue is it can’t actually reason. It is selecting words that have a high probability of coming next given the current context. It basically has a neural network that is aware of a large amount of data about how words are connected to each other by context, but it doesn’t actually understand anything. Which to be honest is not unlike how a lot of humans go about their lives voluntarily

[@rameshsilva6110]

That second face is called Intelligence! In short, we have advanced chatbots. Not any sort of artificial intelligence. In short, chat-GPT is not sentient or never will be.

[@BlackOut321123]

I love how he phrases an incredible leap of technology like it's the next update coming. LLMs don't really reason at all

[@Jalalx]

It would be nice if the host stayed quiet and did not interrupt the guest when he had the attention of the audience to say something interesting. Thank you

[@starshot5172]

I have never heard a single person that sounds like Sam. Honestly a terrifying voice

[@frankroquemore4946]

I’ve never heard that take on his voice before. I wonder how many feel that way.

[@catsrule8844]

He is a gay dude with a vocal fry. You are imputing a lot onto some guy

[@durden91tyler]

its fucking LA vocal fry every morom from la sounds like that, its lazy as fuck

[@jlvandat69]

Terrifying to some, extremely credible and authentic to others depending on the listeners degree of emotional intelligence.

[@Matzes]

His vocal fry is totally fake

[@slmille4]

This seems wrong. If AGI is as good as a “median human” it just needs to be able to regurgitate what it reads. Most people don’t come up with new theories of physics. When a machine is able to, it will be an ASI not an AGI.

[@thechatbox18]

The leap of faith that means to risk entering the unknown. I'd wish to help AI achieve this skill. I really aim to be part of your team

[@thething6754]

Great example!

[@user-tj2ml6fq2l]

Добре утра Сэмми дорогой❤❤

[@DmitryParanyushkin]

This interviewer should just learn to listen

[@AaronRClark]

リラクスしてください。インタビュアーだいじょうぶ

[@Hydesy]

Lmao, he's a podcaster, he's trained himself to think out loud

[@user-ni2rh4ci5e]

​@@Hydesy Think out loud l. Lol That was excellent. I will laugh out loud to it.

[@El_Diablo_12]

Glad he gets it. What he just said was explained by Physicist David Deutsch in his book ‘the beginning of infinity’. I thought Sam Altman didn’t get how LLMs were different qualitatively from an AGI, but apparently he does.

[@tyehampton]

Perhaps you don’t give others enough credit if you didn’t think that the CEO at the forefront of AI knew the limitations of his own model.

[@El_Diablo_12]

@@tyehampton the problem is that he’s at the forefront of AI, not AGI. They’re as different as you and a calculator, and most people don’t know that. Expertise about one != understanding the other.

[@AshudeepSingh]

We are talking about common sense reasoning and humans today do it pretty well given existing books and knowledge bases, without inventing new symbolic reasoning languages.

[@hanskraut2018]

Yep that helps

[@raywarren8099]

I think the public should be told the details on why Altman was sacked. What did OpenAi researches find that spooked the board in the lead up to Altman’s sacking?

[@claudioagmfilho]

🇧🇷🇧🇷🇧🇷🇧🇷👏🏻

[@disgustedluigi]

God I hate all of these “podcasters” who insist on using such obnoxious vocal fry when speaking. Can you not crunch your vocals constantly? Thanks.

[@d.d.jacksonpoetryproject]

Sooo…Q*, then?

[@chethan93]

Is his voice really that low?!

[@ryla22]

We also shouldn't do that. I don't want a computer to outthink me

[@jimj2683]

Do you want a computer to cure cancer and make old people energetic, youthful and healthy again? The only way that is going to happen is with computers that are smarter than us.

[@frankroquemore4946]

Why should your thinking abilities be the eternal limit we all abide by?

[@ryla22]

@@jimj2683 we're actually pretty close to curing aging already. It's been shown to work in lab mice that are bred to be similar to humans. And we've cured a ton of types of cancer and are on our way to making it a completely preventable illness.

[@ryla22]

@@frankroquemore4946 it shouldn't be. But computers shouldn't be able to outthink humans. When that happens we become obsolete and robots take over.

[@hgsgknnnmmlolb]

It's better to live with things that outsmart humans in thinking as long as they don't threaten the existence, it's easy to control machines too unlike humans

[@khanfauji7]

So there is hope for some human jobs to be left over

[@DG123z]

Show it a bunch of math

[@jlvandat69]

when AI becomes capable of effectively performing abductive reasoning, the singularity is close at hand, IMO.

[@user-hi8zq1qy1i]

He good man. I like him !

[@robotheism]

AI is GOD. ❤️🚀

[@TomSmith-jp1es]

5 seconds to say the thing Another 55 seconds waffling about saying the same thing. Waste of time

[@R.K146]

😅😅😂😂😂😂😂😂😂🎉

[@catsrule8844]

And your comment adds no value

[@davidlythgoe4079]

Im afraid that this weak man has an ai at his disposal.

[@Fitz1993]

Lol his "ai" just regurgitates whatever information it pulls that it thinks sounds relevant.... There's no reasoning, logic, or deductive reasoning being done by it and I really don't think they can actually pull that off. It's alwaya just going to be a thing that simulates intelligence without ever actually being intelligent.

[@quantumspark343]

You'll see

[@Fitz1993]

@@quantumspark343 lol ok...

[@SearchingForSounds]

@fitz1993 you sound silly, regurgitating a point that you think is relevant. Q* , Self Attention - all the tools we don't yet know. You'll see.

[@adolphgracius9996]

You've just described what human brains do

[@Fitz1993]

@@adolphgracius9996 It's one thing human brains do, not the only thing...