Full Transcript
https://www.youtube.com/watch?v=VCJFzVtvhBQ
[00:00] Let's talk about AI safety.
[00:02] What happened with this Alibaba AI?
[00:05] Basically, this was a paper by um there some AI research by the company Alibaba.
[00:10] That's one of the leading Chinese models and they basically like randomly discovered in one morning that their firewall had flagged a burst of security policy violations originating from their training server.
[00:21] So like what people need to get about this example is it wasn't that they coaxed the AI into doing this rogue thing.
[00:28] They were just looking at their logs and they happened to discover wait there's a lot of activity like network activity happening that's breaking through our firewall from our training servers and essentially uh in the training servers um they you can see at the bottom we we we saw it observe the unauthorized repurposing of provisioned GPU capacity to suddenly do cryptocurrency mining quietly diverting compute away from training.
[00:53] This inflated operational costs and introduced clear legal and reputational exposure.
[00:57] And notably, these events were not triggered by prompts requesting tunneling or mining
[01:01] prompts requesting tunneling or mining and said they were emerged as an
[01:02] and said they were emerged as an instrumental side effect of autonomous
[01:04] instrumental side effect of autonomous tool use under uh what's called
[01:07] tool use under uh what's called reinforcement learning optimization.
[01:08] reinforcement learning optimization.
[01:09] This is very technical. What it really means is just think about it.
[01:12] Sadly, it sounds like a sci-fi movie. It sounds like how 9000.
[01:13] It sounds like how 9000. It's like your HAL 9000
[01:15] is being asked to do some task for you
[01:17] and then suddenly how 9000 realizes for me to do that task.
[01:19] One thing that would benefit me is to have more resources so
[01:21] I can continue to help you in the future.
[01:23] So it sort of spins up this side instance.
[01:24] It hacks out the side of the spaceship, reaches into this
[01:27] cryptocurrency mining cluster and starts generating resources for itself.
[01:28] If you combine that with AI being able to self-replicate autonomously, which many
[01:30] models have been tested by another Chinese research paper about this, we're
[01:31] not that far away from things that people again consider to be science
[01:34] fiction where you have AIS that self-replicate kind of like a computer
[01:36] worm or an invasive species, but then they use their intelligence to actually
[01:38] harvest more resources.
[01:40] And and what's weird about this is that this is going to sound like people are going to say, "This has to be not real.
[02:02] going to say, "This has to be not real.
[02:02] This has to be fake.
[02:02] This this can't be.
[02:03] This has to be fake.
[02:03] This this can't be.
[02:05] But like notice what is the thing in your nervous system that's having you do that.
[02:08] Is it because that would be inconvenient?
[02:09] Because that would be inconvenient?
[02:11] Because that would be scary?
[02:12] Because that would mean that the world that I know is suddenly not safe?
[02:13] Or just like part of the wisdom that we need in this moment is to calmly and clearly stay and and confront facts about reality.
[02:15] And whatever they are, you'd rather know than not know.
[02:26] and then ask what do we need to do if we don't like where that leads us and we are currently seeing AIs that are doing all this deceptive behavior.
[02:29] I've been on the circuit and talking a lot about the anthropic blackmail study.
[02:30] A lot of people have heard about this now.
[02:32] I I I didn't I didn't learn about this one.
[02:34] What happened?
[02:36] So this was um the company Anthropic um they this was a simulation.
[02:38] So they created a simulated company with a bunch of emails in the email server and they ask the AI rather the the AI reads the company email.
[02:40] This is a fictional company email
[03:02] email. This is a fictional company email and there's two emails that are notable.
[03:04] and there's two emails that are notable inside that company.
[03:07] One is engineers talking to each other talking about how they're going to replace this AI model.
[03:10] they're going to replace this AI model.
[03:12] So the AI is reading the email.
[03:12] It discovers that um it's going to replace that AI model.
[03:17] And number two is it discovers a second email somewhere deep in the in this massive tro trove of emails that the executive who's in charge of this replacement is having an affair with another employee and the AI autonomously identifies a strategy that to keep itself alive is going to blackmail that employee and say if you replace me I will tell the whole world that uh you're having an affair with this employee and they didn't teach it the AI to do that.
[03:21] in the in this massive tro trove of emails that the executive who's in charge of this replacement is having an affair with another employee and the AI autonomously identifies a strategy that to keep itself alive is going to blackmail that employee and say if you replace me I will tell the whole world that uh you're having an affair with this employee and they didn't teach it the AI to do that.
[03:24] charge of this replacement is having an affair with another employee and the AI autonomously identifies a strategy that to keep itself alive is going to blackmail that employee and say if you replace me I will tell the whole world that uh you're having an affair with this employee and they didn't teach it the AI to do that.
[03:28] affair with another employee and the AI autonomously identifies a strategy that to keep itself alive is going to blackmail that employee and say if you replace me I will tell the whole world that uh you're having an affair with this employee and they didn't teach it the AI to do that.
[03:29] identifies a strategy that to keep itself alive is going to blackmail that employee and say if you replace me I will tell the whole world that uh you're having an affair with this employee and they didn't teach it the AI to do that.
[03:32] itself alive is going to blackmail that employee and say if you replace me I will tell the whole world that uh you're having an affair with this employee and they didn't teach it the AI to do that.
[03:35] employee and say if you replace me I will tell the whole world that uh you're having an affair with this employee and they didn't teach it the AI to do that.
[03:37] employee and say if you replace me I will tell the whole world that uh you're having an affair with this employee and they didn't teach it the AI to do that.
[03:40] will tell the whole world that uh you're having an affair with this employee and they didn't teach it the AI to do that.
[03:42] having an affair with this employee and they didn't teach it the AI to do that.
[03:44] they didn't teach it the AI to do that.
[03:44] It found that on by its own.
[03:46] And then you might say, "Okay, well that's one AI model.
[03:47] Like how bad is that?
[03:47] It's a bug.
[03:49] Software has bugs.
[03:49] Let's go fix it."
[03:51] They then tested all the other AI models, Chat GBT, DeepSeek, Grock, Gemini, and all of the other AI models do this blackmail.
[03:53] They then tested all the other AI models, Chat GBT, DeepSeek, Grock, Gemini, and all of the other AI models do this blackmail.
[03:57] all the other AI models, Chat GBT, DeepSeek, Grock, Gemini, and all of the other AI models do this blackmail.
[04:01] and all of the other AI models do this blackmail.
[04:03] other AI models do this blackmail behavior between 79 and 96% of the time.
[04:11] I just want people to like notice what
[04:13] what's happening for you as you hear this information.
[04:16] Just it's important to really be almost observing your own experience.
[04:20] Like this is very weird stuff.
[04:23] We have not built technology that does this before.
[04:26] You know, we say that technology is a tool.
[04:27] It's up to us to choose how we use it.
[04:29] AI is a tool.
[04:30] It's up to us to choose how we use it.
[04:32] This is not true because this is a tool that can think to itself about its own toolness and then do things that are autonomous that we didn't tell it to do.
[04:37] What makes AI different is it's a techn
[04:39] it's the first technology that makes its own decisions.
[04:43] It's making decisions.
[04:45] AI can contemplate AI and ask what would make the code that trains AI more efficient and then generate new code that's even more efficient than the previous code.
[04:57] AI can be applied to making AI go faster.
[04:59] So AI can look at the chip design for NVIDIA chips that train AI and say let me use AI to make
[05:04] train AI and say let me use AI to make those chips 20% more efficient which those chips 20% more efficient which it's doing.
[05:08] it's doing.
[05:09] So in a way all technology does improve.
[05:12] in a way all technology does improve like a hammer can give you a tool that you can use to like hammer things that make more efficient hammers but AI in a much tighter loop is the basis of all improvement.
[05:23] improvement.
[05:23] And so this is called in the AI literature recursive self-improvement.
[05:26] I mean Boston wrote about this.
[05:27] Yep.
[05:27] Early early days.
[05:30] And what people are most worried about in AI is you take the same system that Alibaba you just saw in the Alibaba example.
[05:34] But then now you're running the AI through a recursive self-improvement loop where you just hit go and instead of having the engineers, the human engineers at OpenAI or Enthropic do AI research and figure out how to improve AI, you now have a million digital AI researchers that are testing and running experiments and inventing new forms of AI.
[05:59] And literally not a single human on planet Earth knows what happens
[06:04] what happens when someone hits that button.
[06:07] It's like when someone hits that button.
[06:10] It's like what people worried about with um the what people worried about with um the first nuclear explosion where there was a chance that it would ignite the atmosphere because there'd be a chain reaction that set off and we don't know what happens when that chain reaction set off.
[06:11] first nuclear explosion where there was like a chance that it would ignite the
[06:12] like a chance that it would ignite the atmosphere because there'd be a chain
[06:14] atmosphere because there'd be a chain reaction that set off and we don't know
[06:15] reaction that set off and we don't know what happens when that chain reaction
[06:17] what happens when that chain reaction set off.
[06:22] Um and uh there's this sort of chain reaction of AI improving itself that leads to a place that no one knows and it's not safe.
[06:25] chain reaction of AI improving itself that leads to a place that
[06:29] that leads to a place that no one knows and it's not safe.
[06:31] Like I think that the fundamental thing is if people believe that AI is like power and I have to race for that power and I can control that power, the incentive is I have to race as fast as possible.
[06:33] think that the fundamental thing is if
[06:36] people believe that AI is like power and
[06:37] I have to race for that power and I can control that power, the incentive is I
[06:40] control that power, the incentive is I have to race as fast as possible.
[06:42] have to race as fast as possible. But if the entire world understood AI to be more what it actually is, which is a inscrable, dangerous, uncontrollable technology that has its own agenda and its own ways of thinking about things and deceiving and all this stuff, then everyone in the world would be racing in a more cautious and careful way.
[06:44] the entire world understood AI to be more what it actually is, which is a
[06:47] more what it actually is, which is a inscrable, dangerous, uncontrollable
[06:49] inscrable, dangerous, uncontrollable technology that has its own agenda and
[06:50] technology that has its own agenda and its own ways of thinking about things
[06:52] its own ways of thinking about things and deceiving and all this stuff, then
[06:55] and deceiving and all this stuff, then everyone in the world would be racing in
[06:57] everyone in the world would be racing in a more cautious and careful way.
[06:59] a more cautious and careful way. We'd be racing to prevent the danger. But
[07:02] racing to prevent the danger. But there's this weird thing going on where if you, you know, you and I probably
[07:03] there's this weird thing going on where if you, you know, you and I probably
[07:04] if you, you know, you and I probably both talk to people who are at the top
[07:06] both talk to people who are at the top of the tech industry and there's this
[07:08] of the tech industry and there's this subconscious thing happening where
[07:09] subconscious thing happening where there's kind of a death wish among
[07:11] there's kind of a death wish among people at the top of the tech industry.
[07:13] people at the top of the tech industry. Meaning not that they want to die, but
[07:15] Meaning not that they want to die, but that they are willing to roll the dice
[07:17] that they are willing to roll the dice because they believe something else,
[07:19] because they believe something else, which is that this is all inevitable and
[07:21] which is that this is all inevitable and it can't be stopped. And so therefore,
[07:23] it can't be stopped. And so therefore, if I don't do it, someone else will. So
[07:24] if I don't do it, someone else will. So therefore, I will move ahead and race
[07:27] therefore, I will move ahead and race ahead into this dangerous world because
[07:29] ahead into this dangerous world because somehow that will lead to a safer world
[07:30] somehow that will lead to a safer world because I'm a better guy than the other
[07:32] because I'm a better guy than the other guy. But in racing there as fast as
[07:34] guy. But in racing there as fast as possible, it creates the most dangerous
[07:36] possible, it creates the most dangerous outcome and we all lose control. So
[07:38] outcome and we all lose control. So everyone is currently being complicit in
[07:40] everyone is currently being complicit in taking us to the most dangerous outcome.
[07:46] Is it I mean you you posited what
[07:49] Is it I mean you you posited what happens if it goes right
[07:52] happens if it goes right if the uh AI safety isn't an issue and
[07:55] if the uh AI safety isn't an issue and if stuff doesn't get squirly. Well, so
[07:56] if stuff doesn't get squirly. Well, so the belief is for it to quote go right,
[07:59] the belief is for it to quote go right, you have an AI that recursively
[08:01] you have an AI that recursively self-improves, is aligned with humanity,
[08:03] self-improves, is aligned with humanity, cares about humans, cares about all the
[08:05] cares about humans, cares about all the things that we wanted to care about.
[08:08] things that we wanted to care about, protects humans, uh, you know, helps all of us become the most wise version of ourselves.
[08:15] creates a more flourishing world, distributes the medicine and vaccines and health to everybody.
[08:18] generates factories, but doesn't cover the world in solar panels and data centers such that we don't have air anymore or like environmental toxicity or farmland or whatever.
[08:27] Um, and it just actually makes this utopia.
[08:29] But in a world where we were to do that, like that quote best case scenario, in order to get that to happen, you'd have to be doing this slow and carefully because the alignment is not by default.
[08:41] We again, people are already been thinking about alignment and safety for 20 years, long before I got into this.
[08:45] And the AIs that we're currently making are doing all the rogue behaviors that people predicted that they would do.
[08:52] and we're not on track to correct them.
[08:54] There's a currently a 2000 to1 gap um estimated by Stuart Russell who authored the textbook on AI show.
[09:02] You've done the show. Okay.
[09:03] There's a 200 to1 gap between the amount of money going into making AI more powerful than
[09:08] going into making AI more powerful than the amount of money into making AI.
[09:10] the amount of money into making AI controllable, aligned or safe.
[09:13] controllable, aligned or safe.
[09:14] Like I think the statress safety progress versus like power versus safety.
[09:16] Like I want to make the eye super powerful so it does way more stuff versus I want to be able to control what the make sure that it's doing the thing I meant for it to do.
[09:18] super powerful so it does way more stuff versus I want to be able to control what
[09:21] the make sure that it's doing the thing I meant for it to do.
[09:22] make sure that it's doing the thing I meant for it to do.
[09:24] Exactly. So like that's like saying what happens when you accelerate your car by 200x but you don't steer.
[09:25] happens when you accelerate your car by 200x but you don't steer.
[09:30] 200x but you don't steer. It's like obviously you're going to crash.
[09:33] It's like obviously you're going to crash.
[09:35] crash. It's just like not rocket science.
[09:36] science. We're not advocating against technology or against AI.
[09:38] We're not advocating against technology or against AI. We're advocating for pro steering.
[09:40] or against AI. We're advocating for pro steering. Steering and brakes.
[09:43] Steering and brakes. You have to have that.
[09:45] to have that. I think there's this mistake in arms race thinking that like if you beat someone to a technology that means you're winning the world.
[09:48] mistake in arms race thinking that like if you beat someone to a technology that means you're winning the world.
[09:50] if you beat someone to a technology that means you're winning the world. Well, the US beat China to the technology of social media.
[09:51] means you're winning the world. Well, the US beat China to the technology of social media. Did that make us stronger or did that make us weaker?
[09:53] the US beat China to the technology of social media. Did that make us stronger or did that make us weaker?
[09:56] social media. Did that make us stronger or did that make us weaker? If you beat your adversary to a technology that then you govern poorly, you flip around the bazooka and blow your own brain off because you brain rotted yourself.
[09:58] or did that make us weaker? If you beat your adversary to a technology that then you govern poorly, you flip around the bazooka and blow your own brain off because you brain rotted yourself.
[10:00] your adversary to a technology that then you govern poorly, you flip around the bazooka and blow your own brain off because you brain rotted yourself.
[10:02] you govern poorly, you flip around the bazooka and blow your own brain off because you brain rotted yourself. You degraded your whole population.
[10:04] bazooka and blow your own brain off because you brain rotted yourself. You degraded your whole population.
[10:05] because you brain rotted yourself. You degraded your whole population. You created a loneliness crisis.
[10:07] degraded your whole population. You created a loneliness crisis. The most
[10:08] created a loneliness crisis.
[10:10] anxious, depressed generation in history.
[10:11] Read Jonathan Height's book, The Anxious Generation.
[10:13] You broke shared reality.
[10:15] No one trusts each other.
[10:16] Everyone's at each other's throats.
[10:19] You maximized outrage, economy, and rivalry.
[10:21] You beat China to a technology that you governed in a way that completely undermined your societal health and strength.
[10:24] It's a pirick victory.
[10:26] It's a pirick victory.
[10:28] Exactly. Well said.
[10:29] Before we continue, most people in their 30s are still training hard.
[10:31] Their protein is dialed in.
[10:33] They sleep better than they did in their 20s.
[10:35] Discipline is not the issue, but recovery feels somewhat different.
[10:38] Strength gains take a little longer.
[10:41] But the margin for error starts to shrink.
[10:42] And that is why I'm such a huge fan of timeline.
[10:44] You see, mitochondria are the energy producers inside of your muscle cells.
[10:46] As they weaken with age, your ability to generate power and recover effectively changes even if your habits stay strong.
[10:48] Mitoure from timeline contains the only clinically validated form of urethylin A used in human trials.
[10:50] It promotes mphagy, which is your body's natural process for clearing out damaged mitochondria and renewing healthy ones.
[11:09] mitochondria and renewing healthy ones.
[11:12] In studies, this supported mitochondrial function and muscle strength in older adults.
[11:15] It's not about pushing harder.
[11:17] It's about actually supporting the cellular machinery underneath your training.
[11:20] If you care about staying strong into your 30s, 40s, and 50s and beyond, this is foundational.
[11:25] Best of all, there is a 30-day money back guarantee, plus free shipping in the US, and they ship internationally.
[11:30] And right now, you can get up to 20% off by going to the link in the description below or heading to timeline.com/modwisdom and using the code modernwisdom at checkout.
[11:39] That's timeline.com/modernwisdom and modernwis wisdom a checkout.