Bits + Bips: Why You Can't Opt Out of the AI Guardrails Race

Transcript

Heads up everyone. We've now fully transitioned Bits [music] and Bips to its new dedicated channels. So if you're not yet subscribed to the new Bits and Bips channels on X, YouTube, Spotify, Apple Podcast, or wherever you get your podcast, go there and subscribe now. For the next few weeks, we'll play [music] segments of the full interviews on Unchained as a reminder. But for now, you'll only hear [music] those short segments of bits and bips on Unchained. No more full episodes. You can get links to all of the dedicated channels with the full episodes at unchangedcrypto.com/bits andbbits [music] spelled vips. Thanks again.

>> So I think one interesting point about the AI adoption thing is I wonder how much of this is going to start happening because they will be pulled along by consumers. And I say that specifically, Rahm, because you raised health care. I know tons of people who are now taking things their doctors say to them or like conditions that they have and putting it into AI and trying to get a first cut. Not just from the perspective of like, hey, AI, diagnose me, but before I go to a doctor, I want to understand more of what's going on here and have a vastly more informed conversation with the doctor than I would otherwise have. I think there's also been a lot of testing

of doctor's claims in ways that maybe were not there in the past. And you know, essentially my doctor thinks this, but let me put it into AI. Let me shop for a second doctor if I don't like it or don't think they're right. So, I'm wondering, and I don't have an answer to this, if that behavior is already observable in the health care space or if we're too early in the AI adoption curve to really see the impact here. But to me, there is a possibility industries will be forced to start adopting things because part of what I'm seeing in healthcare in particular, I'm like, you know, I I, as you can tell from my hair, was a mathematician by training originally. And so,

If nobody was paying attention during the World Cup, apparently Anthropics Claude Fable 5 solved the Jacobian conjecture, which is for non-mathematicians. I'm going to make this real simple. It's a roughly 90-year-old problem in algebraic geometry. It was proposed by Keller and it basically was talking about how spaces behave. And what Claude was able to do was provide a counter example to the conjecture. Like it actually came up with a specific knowable computable counter example to the conjecture saying this cannot possibly be true in general because here I found a specific time

where it's not true. And that is a huge result. This is one of the 25 most important problems hanging out there unsolved in mathematics right now. How long until we're seeing that with the quality of these models with regard to like healthcare rom you know good example if I got cancer right now I'm probably going to try to get it sequence to go slam it into one of these things and be like okay what do I do about this like we may already be very close and so I guess I'm asking the question of what is the pull function to increase the rate of adoption here >> well a doctor's using AI now for diagnostics doctors and get the skill level of like house now, right? So, it

has powerful diagnostics capabilities. So, I think that that's there, but we still have so much more to go. Like the promise is still ahead of reality. Even now with the latest models, there's still opportunities for these models to improve. And the Jacobian conjecture, it's a great accomplishment for math. But what real world impact does it have? real world impact is we found a way to target this cancer or here's a way you design a fusion reactor that produces net energy um or here's a way you keep your telmerous you know longer for Asian we still haven't seen that kind of impact which is actually what matters in the world

>> I mean have we not found some of those impacts I now I'm shooting from the hip here but haven't we had various results in like healthare They're already found by AI like >> there's some incredible protein folding you know >> what's the consequence so we've had novel stories the guy with Australia and his dog there's another guy who diagnosed something for with his daughter like there's some novel stories that are emerging and they have some nice local impacts um but you know we still have more to do >> always tons more to do but I I think the acceleration is is profound I across the board, every single discipline is being impacted.

>> I also wonder how much of that what is the right way to say this is adoptionbased. Like doctors, I'm not trying to insult you guys. I have a lot of doctors and nurses in my family, but doctors are old. And so I'm wondering how much of this is just an adoption question because I do think some of these impacts are large to like the alpha thing which is what Chris is talking about. >> Rom for you what it does is allows faster cheap molecule design novel proteins and specific like custom therapies for various diseases and mutations. So the impact is real. The problem is we've got to do all the

studies or get emergency authorization to use these things. It seems like it's more the bureaucracy holding that up. I >> I think the adoption is self- financing because if you're using AI, you have an unnatural competitive advantage. For example, Eli Lilly is using ANI AI and they're driving results. I know we're using AI in a lot of what we do that drives productivity. So if you are using AI, you're going to grow and take market share. And if you're not using AI, you're going to get competed out. So that's going to force adoption.

>> It's also evident on the battlefield. >> Just ask Dario. >> Yeah. And I mean, not just there. I uh as I've been observing this right like the battlefield iteration as we start looking at the nature of warfare if you look at what's going on with the front lines with Ukraine pushing Russia back largely using almost entirely a dronebased strategy some of which are human piloted some of which are probably AI like piloted even if only over short frames right now I think we're seeing more and more of that iteration like today all right it's a cartoon So, let's not get over excited here, but did either of you guys see the Anderil ad

that dropped >> on X today? >> Not yet. >> All right. So, we've got self-flying like helicopter drones that are now engaging in like both target interception and like disabling anti-air defenses to protect like manned US military craft. These things, I think, are coming pretty rapidly as people adopt. So, I'm going to bring this all the way back to where we started. And Chris, this one's for you to start. >> Oh, go ahead.

>> If you're looking at what's going on in the frontier model space, and you've seen the special pleading from Open AI and Anthropic to like restrict these things and keep them out of people's hands, but then China just drops an openweight model that will do it all. How are you thinking about this from a national security perspective? It's it's tricky. Um that there are we're in a national race. Rahm said it beautifully earlier. We have to be a leader in this technology. Full stop. Um and we also have to build in the controls and the you know the end of the day your technology reflects your ideals and we need to make sure that it powers

our ideals and doesn't isn't used to to change our ideals or or go in the opposite direction of ideals. So look, we have to accelerate. We have to be able to outperform our adversaries on the battlefield. There's a massive race going on. Talk to talk to anyone in the defense space. And again, it's not only just AI. Um, you know, I think crypto has important national security aspects. And there's also an arms race going on in the quantum side. Why? Because AI is accelerating all this stuff. So, you got to own it. Um, you know, how do you how do you defend it? I don't know. Like Ukraine is a big laboratory. Everybody knows that. Um lessons are being learned every day and um and it's it's a very very dangerous

world, but we have to we have to really accelerate these technologies to understand them and and to win. >> Rahm, what do you think? >> Yeah. No, I agree. I think you just need policies um to embrace and advance AI. uh you know there's a a post by I think one of the members of the all-in podcast who works for the admin is highlighting the regulatory constraints on AI adoption and he's a part of the administration so you know you have to have the right policy in in place to you know accelerate adoption China is very serious about this opportunity like you know we have the United States has the edge on capabilities um but uh

there's a lot to And as we discussed earlier like the promise is significant. The potential for human impact is significant and possible. >> In finance we found that lit predictable regulated markets attract a lot more liquidity um because there's controls there's a common set of rules and that's where we've found the most success in finance. Uh in AI you know there's a lot of concerns about sharing your information with with some of these Chinese models. you know if they you know is regulation going to be important yes the right type of principlebased regulation will be because if you have models ultimately that you can't trust or hallucinate or you know you you know you can't trust your data I do think a

thoughtfully regulated principles-based industry wins um that that based on my lessons in finance but it's a different industry maybe I'm wrong >> so I'll give a point here on how this is so complicated pulling up some quotes from David Saxs right now on Twitter One, Kimmy K3 just fixed 15 critical security bugs that Codeex and Fable refused because of quote cyber guard rails. There's no reason to limit American models on tasks that Chinese models handle without issue. We're only making ourselves less competitive.

Following tweet, here's another example. Hugging Face tried to use American Frontier models to analyze an AI powered cyber attack, but the guardrails blocked requests containing real exploit payloads. So they switched to GLM 5.2 running locally. The guardrails actually impaired defensive security. And I think this gets to the point that both of you are raising, which is there's no way out of this race. It's going. You could choose not to run, but that doesn't mean the race isn't happening. The race is still happening. you just chose not to be involved. If you like this segment, please like, subscribe, and tune in every Monday at 4:30 p.m. Eastern time.

I'm Austin Campbell, the host of Bips and Bips, along with my friends Ram Alawalia and Chris Perkins, and our slate of exceptional guests. Every week, we're going to discuss macro, crypto, and the collision of worlds, covering topics that move markets and shape the financial landscape. If you hold crypto on your phone, your biggest vulnerability isn't your wallet, it's your carrier. AT&T, Verizon, and T-Mobile have been breached again and again, and SIM swaps are still one of the easiest ways for attackers to drain accounts. That's where Cape comes in.

America's privacy first mobile carrier. Same premium service, but Cape rotates the identifier on your SIM every 24 hours, deletes your call and text metadata after a day, and protects against SIM swaps with a 24word recovery phrase that only you control. You also get two middle toend encrypted secondary numbers for banking and signups, so you stop handing your real number to every app that asks. Go to cape.co/chained co/chained and use code unchained for 33% off your first six months.