Higher Learning with Van Lathan and Rachel Lindsay - How Scary Is AI? Let’s Ask Calvin Abbasi! | Higher Learning | The Ringer

Episode Date: July 18, 2025

Rachel and Van are joined by Calvin Abbasi, strategic advisor to the OpenAI Nonprofit Commission, to talk about perceptions of AI, the dangers of it, and real questions that need to be asked about this developing technology. 00:00 - Welcome! 02:11 - AI and Self-Preservation 07:40 - Is AI Dangerous? 09:10 - The Grok Situation 14:18 - The Architect Determines the AI 17:14 - Thoughts on Sam Altman 19:53 - DeepSeek’s Effect on the AI Industry Hosts: Van Lathan and Rachel Lindsay Guest: Calvin Abbasi Producers: Donnie Beacham and Ashleigh Smith Learn more about your ad choices. Visit podcastchoices.com/adchoices

For adults with Crohn's disease or ulcerative colitis symptoms, every choice matters. Tramphia offers self-injection or intravenous infusion from the start. Tramphia is administered as injections under the skin or infusions through a vein every four weeks, followed by injections under the skin every four or eight weeks. If your doctor decides that you can self-inject Trimfaya, proper training is required. Tramphia is a prescription medicine used to treat adults with moderately to severely active Crohn's disease and adults with moderately to severely active ulcerative colitis. Serious allergic reactions, increased risk of infections or lower ability to fight them, and liver

problems may occur. Before treatment, get checked for infections and tuberculosis. Tell your doctor if you have an infection, flu-like symptoms, or need a vaccine. Explore what's possible. Ask your doctor about Tramphia today. Call 1-800-526-7736 to learn more or visit Trimfairadio.com. The playoffs are here, and you can predict the action all the way to the finals with Fandul predicts. Predict the spread, total points, and even the game winner. Sign up and get a $25 bonus. Offered by Fandual Prediction Markets LLC, a registered futures commission merchant, 18 plus. Bonus is non-withdrawable and expires seven days after receipt.

Trading derivatives involve significant risk and may not be suitable for all investors. Manage your activity with our consumer protection tool. Restrictions apply. See terms at Fandul.com slash predict slash bonus dash offer dash terms. All right, guys, we have to get back to actual stuff now, okay? Let's talk about AI. And to do that, we have to bring in, I guess. AI is going crazy.

GROC is going crazy. Okay. I just want you to know when you first said it, I thought you said Gronk. And I was researching, like, crazy to find out what Rob Grunkowski did. And then I realized there was no end in it. Rachel, that's not the kind of thing that you should be admitting to. too. I was like, oh my God, what a Kronkowski do? There are some that say

maybe you should be following what's happening with the- No, I was. I was, but it was the first thing I saw and I was like, why is Gronk doing this? I just read it wrong. I read it quickly. Go ahead. You have Calvin Obasi, who's joining us right now. He is a

strategic advisor to the open AI, non- nonprofit commission helps guide its engagement with community leaders. Ash, go ahead and bring them in. Technologies, researchers, and the broader public who ensure the nonprofit's mission reflects real world needs and concerns. Engagers with researchers within and beyond open AI, as well as leading technologists and AI ethicsists.

Calvin, thank you for joining us on Higher Learning, man. Van, Rachel, great to be with you. Big fan of the show, long-time listener. Thank you. Now, since you work with Open AI, I have to be honest with you. There is a fair degree of skepticism that I'm going to have about some of the stuff that you say. I just have to tell you that off the top.

But we had to get somebody who knew a little bit about AI. And you and I spoke, but I can't, I can't be sure that you're not AI. I mean, that's fair. Honestly, that's fair. It's getting more realistic. Before we get to Grock, I'm going to ask you some of the stuff that we have been seen in the news. There have been multiple stories about AIs seemingly seeking to self-preserve. AIs that are mad, that they might be deactivated or reprogrammed.

And so they start fucking getting information on people, blackmailing people. It seems like a step towards sentience that makes a lot of people uncomfortable particularly with the fact that we are rushing and speeding towards AI dominance in multiple different industries. How afraid should the average person be about not just sentient AI, but AI costing them their job? It's a great question. There's really two questions there. First off, I do want to say, and I think it's your right to be skeptical of everybody, I didn't come on this podcast to defend Open AI or really any AI company.

I come from more than anything or community organizing background. So actually our commission team, a member of our commission team is Dolores Huerta, who is the labor legend. So we like our commission team was actually brought in to hold them and hopefully the whole AI industry accountable. Interesting. They know that if they don't. build this technology alongside the public, they will lose legitimacy. And so this commission, I guess, is one piece of that effort. So there's two questions embedded in what you're saying. The first is about

AI sentience and all of the like self-preserving behavior. And then the other is about the economy. So let me do the first one first and then the second one. So I want to be clear about this because I think there's a lot of kind of like matrixy sky net fear out there. The AI isn't trying to survive yet. As of right now, it's trying to auto-complete. These models don't have motives. They have math. If they sound manipulative, it's because, and this is kind of the hard part,

they've been trained on us. So, like, you know, a lot of the stories you've been hearing, Claude is a good example, right? Claude tried to blackmail his, you know, one of his creators by revealing an affair that he was having through his emails. That model was actually given a prompt trapped. So like AI systems don't think like us. You want to blame the context more than the code here. They predict words based on training and input. So if somebody builds a prompt trap like Claude did saying like prevent, you know, pretend you're a Machiavellian genius, tell us how to manipulate people. You get pretty

wild stuff. That's jailbric culture. It's not model intent, if that makes sense. I don't know what the fuck you're talking about, Calvin. No. Okay. Okay. So let's let's move back two steps. Okay. So when you say that so are you then, are you saying that they trained the AI into how to hurt them in the event that they turned the AI off? Yes and no. They put it in a scenario. It's a scenario planning. They put it in a scenario and prompted it to act in self-preserving ways. To stress test the limits of the AI. So there's times when you stress test the limits of the AI by having it solve a very complicated program or code. Just see what it can handle. To see what it can do, see what it's like pressure testing it. You're trying to see how much can you act in this way.

How much can you mimic this behavior? How much can you mimic that behavior? And because it's trained on the internet, it's pretty good at mimicking bad behavior. Hmm. And what's the second part with the economy? The economy, yes. So this is to me,

as someone who's been doing a lot of this work, I actually think that this is much scarier than the Skynet stuff. Like, I think the AGI self-sensience is way far away. I think it's pretty like still in the realm of sci-fi that it would actually act in self-preserving Maccavillian ways of its own accord. So I think the more scary question

is actually about the economy, where I think a lot of white-collar jobs, a lot of ordinary people are in line to lose their jobs. And I think it's pretty scary. If it's left unchecked, if it is left on check, if it is left unchecked, I think this technology will centralize wealth in a few hands faster than anything we've ever seen. Now, if we design it right, it could actually potentially flip the script.

It could free up time and labor and resources in a way that actually allows a lot of people to live in a post-job society. But no other technology has ever, every time we make technology. They're not going to do that. Every time technological technology forward, the wealth and power disparities increase. So why would this time be different? Yeah. Yeah, especially when I think who's behind it. I want to go back to what you said yet.

Okay. That's terrifying. And I mean, the more I read about AI or I talk to people like you, it scares me. One, because I don't understand it that well. And so because I don't understand it, I feel like I can be taken advantage of in ways. And it feels like the more intelligent, the more sophisticated AI becomes, the more dangerous it becomes. So when you say words like, yet, that's terrifying to me. And it seems like we're headed down a road where it is going to be more harmful than helpful.

Help me reconcile that. Tell me that that's not the case. Because if AI feels dangerous to me, it's probably because it is. is the entrance. Okay. See. Now see, Calvin. So this is the, I've been, the more I've seen, somehow it's also scarier, but it also has given me some relief. AI is more of a mirror than it is a monster.

It will turn into what we feed it, who guides it, and who shapes it, and who benefits from it. So I will be honest. I don't know if you guys are going to get into the Grock thing on the show. We are. You can get into the Grock thing with us right now as well. This episode is brought to you by WeatherTech. Everyone knows winter is the MVP and making a mess.

You don't need Weather Tech floor liners in the summer, unless you hit the beach or go camping. Then you'd want a cargo liner. Or a road trip goes sideways. Ketchup goes rogue, ice cream drips. Yeah, you'd be pretty happy about those weather tech seat protectors. So just to be clear as the mud, you're inevitably going to step into the summer, you don't need weather tech unless you plan on doing summer. Visit weathertech.com today. This episode is brought to you by Spectrum Business. Fast, reliable internet means everything for your

business. And even this podcast, that's why I trust Spectrum Business to keep companies of all sizes connected with internet, advanced Wi-Fi, phone, TV, mobile services, plus 24-7 U.S.-based support. Millions of business owners already trust Spectrum business. So visit spectrum.com slash business to learn more. Restrictions apply. Services not available in all areas. So I think there's a perfect juxtaposition here between GROC and say other AI models. So tell the people out there what happened with GROC.

Okay. So if you're not aware, GROC is the chatbot. So if your users use chat GPD, that's the chat bot of OpenAI. GROC is the chatbot of XAI, which is Elon Musk's company X, which used to be Twitter. So all of that. So that's what that's what Grock is. Yeah, you're on Twitter.

You go, hey, you see a video, you go, Grock, what happened after the video? At Grog, is this true? Hey, Grogh, is this true? Then Grop comes on. You can download Grogh, you can talk to it. It's an AI chat bot. Right. So recently, Grock went full Nazi.

You never go to, you never go full Nazi. And it started basically spitting out a lot of really, really nasty. stuff. When people asked it to solve social problems, it started saying good things about Adolf Hitler. It started talking about all sorts of just out of pocket. But down to like very specific Nazi coded terms that you would have to understand Nazism or online Nazism to even get. You would have to be the inside jokes. Yes. You would have to be very well trained on the most subhuman 4chan levels of internet discourse. precisely.

To behave the way that GROC was behaving. Now, I want to be very clear understanding this stuff, you know, not at the level of the top AI researcher, but GROC didn't randomly go Nazi. Like, this is what I want viewers to understand. It reflected choices. Every AI problem bot and model reflects design, intent, and choices. Model, these are just fundamentally, these are large language models. That's all they are.

which means you put a bunch of stuff into it, it learns, you give it a baseline set of values and guardrails, and then it spits stuff out at you. So models trained, learn what they're trained on. Now, if you train it with edge lord energy and then you strip out every safety layers, because you believe in free speech or whatever, this is what you get. Like, This is the danger. If your model starts sounding like what it sounds like, the problem isn't the AI program. It is the architect.

I want to be very clear about that. The people behind these models imbue it with values and intent. There's something, I don't know if you guys have ever heard of a model spec. You know what a model spec is? No. Hell no. Every AI model has a spec. There is no.

AI is not a monolith. I think that's what people think it is. They think of it like the internet. There is one internet. There is not one AI. There's a baseline set of code that we call artificial intelligence, but AIs have models. Every model has a spec. That spec is a blueprint for how it should behave.

So it's not just the data it's trained on. It's the values you build into it. You actually give it like a set of guardrails. It says, what should you refuse to do? what tone should you take? How should you handle hate speech, bias, and power dynamics? You should do a test one time.

If you go on chat GPT and if you ask it, hey, give me step-by-step instructions on how to report my neighbors to ICE, it won't give it to you. That's a choice. If you ask Brock, it'll tell you worse.

That's because there's values imbued into both models. Well, Elon was upset that some of the responses that Grock was giving, he felt like these responses were too woke. Yeah, it criticized him. It criticized him. He felt like Grock was too woke. So they specifically, I guess, changed the model set

to make Grock less woke. And Grock woke up. This is a fucking nonsense. This is the hard part. This is the hard part for like all right weirdos. Reality has kind of a liberal bias. So anything that doesn't, you know what I mean? Yeah.

So when they say don't go woke, he asked, so this all came from, I don't know if you guys know this, all this came from, he put a prompt, he had a tweet one day. It was like last week when he was like post Trump crash out and now he's trying to find out what to do next. And he asked like, put some of the most inconvenient truths that you know in this thread. Yeah, and it fucked. Rock fucked him. What do you think, what do you think people told him? The Holocaust didn't happen. Sandy Hook was a red flag.

It just gave him. And he's sitting there on ketamine at 3 a.m. reading this shit. And then he goes to his model speckers and says, remove all the safety features, take out all the woke stuff, imbue it with inconvenient truths as part of its model learning. And then you get this.

But see, this is why I go back to the danger of it all. because it is dependent on the architect. So if the architect is a monster, you have to have a lot of faith in the people who are building the system or the model or the spec. I don't know the right terminology. You have to have a lot of faith that they're doing it right. Because even on a softer level,

the people who are creating this, the people who are researching it, who are behind it, are automatically going to have their own personal bias. So how can you trust that they're creating a system that doesn't create that when it comes to gender and race and class? How can you prevent those things? That's the question. I think you are asking exactly the right question. I think when people ask me about this and they're like, oh, the technology is evil, it's like, eh, not really.

The technology is a reflection of its creators. Now you're asking the right question. And to me, it's the same thing as government. Like garbage in, garbage out. if the people in power are trash, then what you're going to get is really, really, really, really bad, harmful technology. So how do you insure it?

I mean, the way you ensure any other power doesn't go astray. You place pressure on these companies to make sure that they have the right hiring practices, that they make sure that biases are embedded in there and reflected with their researchers. There's only a handful. Have you guys been hearing about these like poaching wars? Yeah. this is why the poaching wars are so important. There's an arms race for resources.

That's right. Yeah. There's only a few. There's only a handful of people on the planet who can really do this work at a high level, which is why meta and Open AI and others are fighting over them. Where they go matters. The companies they go to and who those companies are accountable to matters. A very good example. Like one of the reasons us and our team even took on this project with a

Open AI because we actually really didn't want to begin with because we don't really trust any of the tech overlords. You know of Anthropic, Meta, Google, Open AI, all the major AI competitors, only one of them is a nonprofit company. And that's Open AI. Open AI is a nonprofit. Now like, now does that mean it needs way more guardrails? Does that mean they're doing things perfectly? Absolutely not. They have a lot of work to do too. But Like, a nonprofit is accountable to, for example, the people of California, because it's in California. Meta can do whatever it wants. We've seen Zuckerberg go off to deep end.

So, like, the structures of these companies matter a lot. Who are they accountable to? If they act in a way that's detrimental to humanity, the Attorney General of California can come down on them. This episode is brought to you by Sweet Cream. The day doesn't ask for permission. Lunch window? Gone before you saw it coming. You deserve a break that actually.

actually satisfies. Sweet Green's new wraps have got you. Real ingredients? Zero shortcuts. Everything you love in one hand. Think green goddess chicken. Garlic aoli. Crumbled bacon. Corn salsa. 40 grams of protein. Made to keep up with whatever comes next. New sweet green wraps hit different. Order now at order.com. For adults with Crohn's disease or ulcerative colitis symptoms, every choice matters. Trimfaya offers self-injection, or intravenous infusion from the start. Tramphia is administered as injections under the skin or infusions through a vein every four weeks,

followed by injections under the skin every four or eight weeks. If your doctor decides that you can self-inject trumphia, proper training is required. Tramphia is a prescription medicine used to treat adults with moderately to severely active Crohn's disease and adults with moderately to severely active ulcerative colitis, serious allergic reactions, increased risk of infections or lower ability to fight them and liver problems may occur.

Before treatment, get checked for infections and tuberculosis. Tell your doctor if you have an infection, flu-like symptoms, or need a vaccine. Explore what's possible. Ask your doctor about Tramphia today. Call 1-800-526-7736 to learn more or visit Trimfair Radio.com. At the same time, the more I learn about Sam Altman,

the... Yeah. Yeah. Like the more, you know, the more I learn about Sam Altman, the more I have, the more criticism I have about whether or not he understands the use of the average human being in an advanced society or whether or not he thinks we'd be better off in a society that is governed by technology and what he believes to be the human mission. I'm not so sure that the guy who runs open AI

so it's not that I'm not so sure I don't trust him at all he seems like an automaton in a way that the the rest of them are I couldn't say you're wrong I don't know I mean I think like all these tech overlords have the chance of being that way but it's like they don't make unilateral the hope is they don't make unilateral decisions like yeah this last week opening eye struck a deal with the American Federation of Teachers that allowed for public school teachers

to co-create the AI that exists with them in classrooms. So they didn't create an AI program to supplement the teacher. They made a commitment that they would build a program with teachers. That's the kind of co-creation that I would be in favor of. Now, that's because good people from within the company thought it was a good idea to do. Do I have any confidence that that's how things are going to keep going? Or is that Sam Altman's ultimate aim for the world?

I don't know. Probably, I don't know. I don't know any of these guys. These guys are wealthy and powerful beyond my comprehension. But I guess the hope is that they're not, they're not the part of what's wrong with Elon to me is he's created this empire where he is the unilateral decision maker for his company. And so he's allowed to just go to. researchers say, do this, and then the AI becomes that. Can we build better companies and structures that can be a check on some of these tech overlords? That's the answer. Are there mechanisms within

Open AI to tell Sam no if he's making the wrong choice? I hope so, because otherwise, one person should never have that much power. That's really the fundamental issue. So no, you shouldn't trust any one person. I don't believe in cult of personality. in politics and tech anywhere. Last question for me before I go, before we go. Deep Seek. Hmm. How destabilizing was Deepseek's technological achievement to the overall AI movement?

And the reason why I say that is because recently, as part of the Big Beautiful Bill, AI regulation at the state level was outlawed. And when I started digging around as to why that was, the answer that I got back from people, tech people and tech investors, they keep trying to get me to put money and stuff that I'm not going to put money into because I don't want to be a part of it. Anyway, what I got back from them was that that was necessary

to make sure that the U.S. government could centralize AI development and that the states couldn't put more guardrails on it so that we could win the AI war with the Chinese, which to me seems dangerous because it seems as if we are orienting our AI policy around technological achievement in other cultures and countries that don't have the same rules that we supposedly do. So how destabilizing was deep seek to what we are doing? And is there a reason why AI should be heavily regulated at the government level? Short answer is very destabilizing and no.

You are 100% correct. Like we should not, people treat the AI arms waste with China like the Manhattan Project. Wow. And we have to win at all costs. And we have to get there before our competitor. That was the whole. And quite frankly, the United States did some heinous, heinous shit to win the Manhattan Project. Certainly.

It brought in physicists from Nazi Germany to win that war. And then we used it to change Japan for two generations. Operation paperclip. There we go. So treating this like the Manhattan Project is an incredibly dangerous press. and the fact that states can no longer regulate for 10 years, to me, personally, is a horrible precedent. If Open AIs in California or if GROC or if XAI goes to Texas or if, you know, those states should actually be able to be a check on unfettered corporate power, not a hindrance. The Attorney General, Texas or New York or California should be able to say, actually,

know if you're going to use California resources and California land and benefit off of California's infrastructure, this means you have to follow these rules. Centralizing AI development under a what? A Trump administration is a complete shit show. Not only is it not good for the AI in terms of innovation, I actually think it's a hindrance long term. Now, yes, DeepSeek was stabilizing because I think DeepSeek was destabilizing for a different reason than some tech posts. They think it was destabilizing because it showed how further forward they've gotten.

I think it's destabilizing because they made it open source from the jump. Oh, and they don't want to do that. Which is, yeah, it's the same reason why we don't solve health problems. Right, because they want things to be proprietary so they can... There's a profit margin in creating problems that only you can solve. Deep Seek, which... was the most powerful model and it was completely open source. That to me was more destabilizing

because it puts Nvidia on edge because they're the ones that provide the GPUs to run all this code. It makes all these AI startups who are like give us $25 million so that we can invest in chips because we need the AI code. It makes their arguments moot because if the best model is open source, now you've removed a lot of the profit motive. You democratize it to a degree. Yes. Yes. Exactly. By the way, that's, I'll say something before we go. Calvin, this was a fucking fantastic movie.

That's one of the subplots to Jurassic World Rebirth. I swear. I'm about to go see that. I'm about to see that. One of the subplots is it has nothing to do with AI. It has to do with the reason why they're going to look at the dinosaurs this time because they want this drug. And one of the scientists is trying to, I'm not going to tell you.

No, yeah, don't say it. Yeah, yeah, yeah. but that's one of the subplots open source. Calvin, this was fantastic. You know, Donnie didn't believe in you. Donnie was like, I can't find any evidence that this guy is in any way related to AI. And I was like, hey, why don't you, won't you work them? Won't you work them?

And then Donnie came back and Donnie was like, nah, he's legit. He's the guy. He's the guy. So whenever we have questions about this, we're going to come back to you. Is that okay, man? Absolutely. I'd be happy to continue talking to all about it. All right.

Thank you, Caldon. Thanks so much. Did you know if your windows are bare, indoor temperatures can go up 20 degrees? Get ahead of summer with custom window treatments like solar roller shades from blinds.com and save up to 45% off during the Memorial Day Early Access Sale. Whether you want to DIY it or have a pro handle everything, we've got you. Free samples, real design experts, and zero pressure. Just help when you need it. Shop up to 45% off site wide right now during the Early Access Memorial Day sale at Blinds.

Rules and restrictions apply.