Is AGI on the Horizon?
Вставка
- Опубліковано 25 січ 2025
- We’ll explain why 2025 is already shaping up to be a huge year in A.I. - with models like OpenAI’s o3, Google’s Gemini 2.0 and DeepSeek, from China, stirring discussion that superintelligence is near.
This is a clip from episode 117: • Meta Goes MAGA Mode | ...
Additional Reading:
OpenAI Unveils New A.I. That Can ‘Reason’ Through Math and Science Problems
www.nytimes.co...
Hard Fork is a weekly look into the future that's already here. Hosts Kevin Roose and Casey Newton explore stories from the bleeding edge of tech.
Casey’s publication, Platformer: www.platformer...
FOLLOW US ON SOCIAL
TikTok:www.tiktok.com...
Insta:
Kevin Roose- www.instagram....
Casey Newton - www.instagram....
Bluesky:
Casey - bsky.app/profi...
Kevin - bsky.app/profi...
Twitter (X):
Kevin - x.com/kevinroo...
Casey - x.com/CaseyNew...
Threads:
Kevin - www.threads.ne...
Casey - www.threads.ne...
Subscribe to the audio only podcast:
Apple - podcasts.apple...
Spotify - open.spotify.c...
Amazon - music.amazon.c...
Google - podcasts.googl...
The New York Times - www.nytimes.co...
Credits
“Hard Fork” is hosted by Kevin Roose and Casey Newton. Produced by Rachel Cohn and Whitney Jones. Edited by Rachel Dry. Executive Producer, Jen Poyant Engineering by Alyssa Moxley and original music by Dan Powell, Elisheba Ittoop and Marion Lozano. Our audience is Nell Gallogly. Video production by Ryan Manning and Chris Schodt. Video QC by Isabella Anderson. Podcast Art Direction by Benjamin Wilkerson Tousley. Podcast Art & Animations by Julian Hespenheide. Additional Motion graphics by Phil Robibero. Thumbnails by Julia Moburg, Elizabeth Bristow, and Harshal Duddalwar. Special thanks to Paula Szuchman, Pui-Wing Tam, Nell Gallogly, Kate LoPresti and Jeffrey Miranda.
The fundamental silliness in most definitions of "artificial general intelligence" is that they specify a broad range of competencies that no single human possesses, and thus equate general intelligence in a machine with superhuman performance.
By the way, we already have superhuman intelligences that are poorly aligned with human interests. We call them corporations, and confer legal personhood on them. The path toward aligning super-intelligences with human interests begins with corporations. (And that is a disturbing observation.)
@@tomenglish9340 a corporation is run by many humans...
@@tomenglish9340 don't disagree with the alignment issue though. Just not superintelligent
Excellent as always, thanks, guys!
Thanks for this update, I love you guys :)
Thanks for the podcast. I always enjoy your perspectives.
If you are someone who is worried about AI safety like me, then this episode just comes with one body blow after another. The tech bros who have before indicated they will willingly constrain themselves to build AI safely are getting more devil-may-care, Trump and congress are very uninterested in safety regulation, the avenue for constraining compute is vanishing with the DeepSeek demonstration of cheap and capable models and we now seem to be headed to a sped up AGI timeline. Grim.
you missed quite a lot that happened last year already
Agreed
You said “what is super intelligence but a system that is better than any human” - that’s more AGI. Super intelligence is a system that has the intelligence of all humans combined and more…(but I’m sure you well know this, needless to say)
Humans are so quaint. It used to be that we said AI cannot make art or write poetry,; now it is superhuman at those. Now we say it can only do math and coding because of the reward function. Come on........we are in our primate winter ❄️
They didn't word that very well. The situation is that the reinforcement training techniques applied to o3 work well when there's a method for efficiently verifying that a response is correct, and not well when there's no such verification method. The capabilities of generative AI in making art and writing poetry are not the result of the same sort of training that makes the o-series models good at mathematical proof. In fact, in human preference testing, o1 is no better at writing than GPT-4o.
as a poet who uses cutting edge models every day for various tasks it's nowhere near superhuman poetry except in terms of sheer volume
You have low standards for art and poetry
AGI is on the horizon as much as artificial general consciousness is 😊 0:16
Are computers really that intelligent if you need a datacenter to recognize a cat or write a Python program where large parts of it already exist on the internet?
It will only be data center sized for a few years until they miniaturize it just like all previous computers
It is the 174th best coder. This is far beyond your assertions.
@ They should try it on 20 watts of power. Or better yet, 10 W. And only let it read a few computer science books. And make it drive to the test site. And make it feel stupid or angry when it gets a question wrong.
Congratulations to your family Casey!
no
Are you guys literally reading from your screens ? Like word for word ? Hard to watch
Open AI spent $1 million to score 87% on that coding benchmark. How much would it cost to have human programmers do the same thing?
They actually answered that question: it was $5
@@Misterchalmwhere??
@@dandushi9872 it was on their 12 days of Shipmas webcast I think
@@dandushi9872 the video where they announced o3 I believe
@@Misterchalm no, it was five dollars to perform a single problem on the arc AGI test. That is what the humans who baselined the test were paid per problem. I think there are 200 problems in the private test set would make $1000. However, I think a score as high as this was only achieved by stem graduates, and not average people taking the test, so they would cost more.
Altman doesn't know a thing about building AI/AGI. He's just THE influencer that triggered Microsoft into going all-in with a pair of two's, and by now everyone at the table buying more and more chips (see what I did there) to add onto the betting pile.
lol. You’re funny.
hehe didnt realise he was a gay. congratulations on your new relationship
A one and a two and a gay
@kevinnugent6530 heeehehehe bro stop playing.
he annoyingly talks about it all the time, make it stop
hearing about his "love" life is terrible, this is just making it way worse