Speaker 0:  meta put a 30,000,000,000 parameter ai agent on a single consumer gpu and simultaneously started lobbying washington to keep models like it open. this isn't merely another model release. it's a fight over whether powerful ai lives inside five computer data centers or increasingly lives on hardware you control. meta just released muse glimmer, an apache 20 open weight, roughly 30,000,000,000 parameter model built specifically for local agents. the coding tool used, and long running workflows. quantitized versions fit into a 32 gig memory envelope. so this isn't an ai on every laptop, but it also puts surprisingly capable agents onto high end consumer hardware. the biggest story is timing. zuckerberg is arguing that america needs stronger open weight models to compete with china just as trump administration has chosen not to include open weight models in its voluntary fed prerelease safety testing programs. what's unclear is whether glimmer is good enough in real world deployments to make businesses seriously substitute local inference for frontier cloud models. what do you think about this, john?

Speaker 1:  yeah. i don't think you got

Speaker 0:  the right article there, craig.

Speaker 1:  we're talking about his ai manifesto, not about the the meta. that the article was about his manifesto and about, how his thesis is that the most valuable use of increasing intelligent ai will not be automation, but it will be invention.

Speaker 0:  yeah. i agree. however and this covers the article, but i think the timing of him releasing these open weight models and then coming out with this is the actual story, not just a manifesto itself.

Speaker 1:  i don't know because, i think, like, for sure. it, but i think this manifesto is is is kind of interesting because he argues that early ai systems, they're primarily obviously, they answer questions and automated routine work, and we're getting into more advanced systems that could increase increasing increasingly discover new ideas. like, like, we've covered medicines and proteins and different things like that. so, in his view, the economic upside from discovering previously nonexistent things could ultimately be much larger than simply replacing humans performing existing tasks. and that distinction matters because there's essentially two competing visions of advanced ai. one is the automation model where ai performs existing human jobs more cheaply. the other is augmentation kind of the invention model where ai dramatically increases what each individual can accomplish. and so zuckerberg is explicitly and this relates back to his models, but he's placing meta in the second cap. and his arguing is that if ai development is primarily directed towards replacing knowledge workers, society could experience severe disruption. but if ai increases individual capability quickly enough, humans can create entirely new categories of economic activity before old occupations disappear. and so this is probably, like, the, i would say, the philosophical heart of his manifesto. but back to your point, yeah, i so there's this there's two camps when it comes to llms as well, just specifically llms, not the way they're trained, but, there are people saying that llms are gonna become less data, data center, dependent and more portable and customizable. and yeah. go ahead, greg.

Speaker 0:  so i would disagree about what you think is at the heart of his manifesto. i think at the heart of his manifest manifesto is don't build one perfectly aligned superintelligence controlled by a handful of institutions. give billions of people their own superintelligence agents, and let competing people, companies, and i keep each other in check. so, you know, is that what he's saying? or, you know, who gets to control superintelligence? and the reason i talked about these models is alright. so is open source ever gonna be good enough? if open source isn't good enough, is manifesto doesn't matter doesn't matter whatsoever. right?

Speaker 1:  k. look at kimmy k three. that's open source. that's an amazing model.

Speaker 0:  so my point being is you can't take the conversation without talking about the open source models.

Speaker 1:  fair enough.

Speaker 0:  alright. so, his essay opens, with who gets superintelligence. so how do we control powerful ai? who controls the controllers? and he tries to zuckerberg's answer to dangerous concentrated intelligence is basically don't concentrate. don't do that. so my question based on this because he predicts that everyone will get a twenty four seven super, personal super agent. how is that different? so i sit back and say, is your mod model a realistic one? countries, huge corporations, nation states are not gonna stop building their concentrated intelligence. so are we talking about a effort where they're gonna be head butting with each other at some point, or they're just gonna cast off lesser models below and we'll have competing people on open source, deepsea, kimmy k, all those coming up from behind to try and get it? what's your prediction on this?

Speaker 1:  i don't think we can create one university benevolent superintelligence in my if that if that answers your question. i think it's gotta be distributed power. there's gonna be a balance of power, problem. and you and i talked about this last time, or in a previous episode about, how the us government blocked access to mythos in fable. right? and so, if i think open source is gonna be critical for this, but i also think there's going to be there's going to be issues when it comes to different governments, different values, different religion, political groups, all that sort of stuff that are going to be arguing over these models and the and models capabilities. and i think again, it's it's gonna be kind of like the new religion where people are gonna argue and people are gonna have opinions. and i don't know what it's going to look like, and i'm scared for that because i just want ai to solve the world's problems, not create more. it's on everybody to love each other.

Speaker 0:  we definitely need more love in the world, but i gotta say that it'll solve problems while creating a bunch more. we are currently in the information age, they call it. yeah. i'm not sure that is true because

Speaker 1:  mhmm.

Speaker 0:  are we moving into an ai age or an intelligence age away from information age when information and intelligence is cheap to nothing, how does the world's economy change?

Speaker 1:  yeah. well, it becomes less about well, no. i don't i, yeah, i don't i don't know how that changes. that's i think that's a philosophical question if i ever did come across one. like, we talk about, like, there's universal basic income, right, that could, help with the transition, but

Speaker 0:  it doesn't work.

Speaker 1:  yep.

Speaker 0:  so, like and i say it doesn't work. so the ubi doesn't work because what happens is the you get an unproductive society when what we need is to change their production rather than being unproductive. right? so i sit back and say, our nation state might have to change. when i was younger, i used to think, well, what will probably end up a little bit more is more corporate controlled environments. we kinda went a little bit that way, but not enough. where you're you'll your company will pay for your kid's education. your company will pay, because they'll have, like, microsoft university or whatever, and your kids will go there and everyone will you know, your stuff will be here. so you're locked into an environment, and that takes away a lot of this need for ubi because you're our company family. alright. that might work. i don't like that because now we're talking about putting people above this and above that. and i think they're going to need ubi, a skills gap, a payment thing that gets us from a to b because knowledge workers are going to shrink to next to nothing because everybody will have all the knowledge workers they need at their fingertips with ai. whether that's ten years from now or two years from now, you can argue till the cows come home, but it'll all be there.

Speaker 1:  yeah.

Speaker 0:  so what does that mean how many of the thousands of people who work at all these people are knowledge workers? what becomes a specialized skill? what becomes, the labor, the manual set, the things we build, the things outside where life changes because we all have enough compute, we all have enough game. we all have enough entertainment and distraction because ai can make whatever we want from entertainment point of view on demand, games, videos, what have you. we're gonna have to start coming back to the real world. and, so i think we're gonna get more carpenters, more tradespeople, more people who make, you know, a value of other things. so i don't know because the world is gonna change. and like i said, knowledge workers is done. you know, not now, but ten years from now. and when i say done, one person with a computer will be able to do what a whole team used to do.

Speaker 1:  yeah. i'm doing it right now.

Speaker 0:  well, yeah. but that's only a handful of people who can do that now because they haven't taken the time.

Speaker 1:  i get what you mean. yeah. i get what you mean.

Speaker 0:  you know? so we're moving into you know, people see terminator. i see iron man where we're enhancing people to be superpowered individuals that can get a lot done and, you know, jarvis working with you, not against you, versus, you know, if you think ai is coming for your job, then guess what? it will. yeah. if you decide that, i'm gonna embrace this and i'm gonna be, you know, a wielder of this tool, then odds are you'll have a longer employment life. what that means exactly, i

Speaker 1:  don't know. explain what you do at what your job is, it can be replaced.

Speaker 0:  yeah. so the world is changing. and, you know, this is a bigger topic, probably our last topic just because we, you know, we're delving deep in we look at hollywood. hollywood's changing. they're dead set against ai, yet they're using ai every day in these things. they're using ai for all sorts of things. they're using ai for dubbing. they can now dub your voice and your accent in real time and whatever so that it's not crappy. so they can use ai to do that. they can use ai to do special effects. what's the difference between ai and cgi? somebody gonna tell me that there's, you know, somebody with a couple of ai programs to build special effects versus cgi. the world's become faster. so you're gonna have graphics experts who do those things, but they're just gonna use better, faster, cheaper tools. so what used to cost millions, you're gonna now be able to do for a couple grand. and then, you know, next to nothing, next to nothing. right? but if you wanna do it at scale, you wanna do marvel graphics, it's gonna be what have you. but, i mean, the world is changing that way. these jobs will still need to be done, but it'll be very specialized jobs using very specialized tools, to get these things. and the world is gonna have to change. we're gonna see more independent businesses, more smaller teams doing very specialized things. so what those things are, we'll have to see.

Speaker 1:  yeah. exciting.

Speaker 0:  it's exciting. it's scary. it's all those things. it's a very interesting article. and, despite how we started it and you're thinking that this is, different, i thought we attacked it at this article at a different angles, which good. zuckerberg's manifesto isn't really about muse, glimmer. it's about who owns superintelligence. his argument is i keep trying to keep an extremely powerful ai inside. a few supposedly responsible companies may actually be a danger of task. instead, he wants billions of people to have personal superintelligence aligned to their goals with competing ais, companies, and governments acting like checks and balances in a democracy. now do you think this is because he's a sore loser, because, manus and everything deal backed out and they're behind everything? do you think this is philosophically his angle? let us know what you think, his thoughts are on this, and what do you think whether he's dreaming or this is actually the future? some sort of democratized ai and fully paid for ai head butting each other as we run through life. we'd love to hear what your thoughts are in the comments below.