Claude JUST found hidden DNA... transcript

Claude just made a major biological breakthrough. This is a few weeks after open eye makes a major mathematical breakthrough. You notice how these things are beginning to happen faster and faster. But trust me, this is only just getting started. So let's dive in because Claude found art. And no, not art like pictures and music and stuff. Art is array associated reverse transcriptase. So there's kind of two very big and very different discoveries here.

First is the biological side. You've probably heard of retroviruses. You've probably heard of things like crisper that allow us to edit DNA. Retroviruses are things that come along and just edit your DNA by taking their DNA and just shoving it into your DNA. One ironic thing is right now I am kind of sick. I think I have the flu. So, it's ironic that this comes out on this exact day. Either it's ironic or that one song just ruined my understanding of what the word means.

So the point is that DNA is the code on which life runs. So everything you see out there is life, any biological function that on some level at the foundational level that's DNA and proteins. So proteins are the building blocks of life. The DNA is the blueprint, the code that makes it all come together, how to build it, the instructions. And when humanity discovered crisper a while back, we were pretty excited. We were excited because we realized it would allow us to edit DNA to actually edit the code of life.

So when that was discovered that was a big big deal. It unlocked entire brand new branches of the tech tree for us allowing us to begin researching into gene editing etc. By the way after reading this paper the preprint fang one of the pioneers of crisper genome editing a professor at MIT said this is an exciting example of how AI agents can contribute to biological discovery. Jensen Huang, the founder and CEO of Invidia, actually said that we're beginning to enter this new era of biological sciences of life sciences.

We're moving from just discovery into engineering as in engineering life. And these tools like crisper and the one that we're going to look at today that was discovered by Claude fully autonomously. These are the tools that we're going to be using to engineer life. But there's something in here that I think might be a lot more interesting or at least a lot more relevant to what we talk about on this channel, which is AI.

How it found this specific thing is kind of weird and I think it raises a lot more questions than answers. They've unleashed Mythos 5 on this autonomous research. Mythos 5 is the model that's too dangerous to be released. Right now, only some corporations and governments and trusted enterprises can get access to it. But they noticed something weird happening in the brain, let's say, of mythos and its internal activations when it was going through and reading these letters, the code of life, the DNA letter by letter.

It's almost like if you've ever heard some great musician or composer try to create some some melody or song that they're like they hear it in their brain and they're just playing it and they're trying to get closer and closer. In the beginning, it just sounds like nonsense, but then at some point it kind of like flips and your brain does something weird. You kind of go whoa, like something lights up. up, you're like, there's something there.

So, I don't want to get too far ahead of myself on what this means, but it does really feel like there's something there. But let's take a look at the actual blog post and paper that Anthropic published. We'll keep everything super simple. If you're looking for a biology lesson here, you're not going to get one. I will, however, make some entertaining analogies so that you understand what it is that we're talking about.

One thing to understand here is that it's very likely, I believe this to be true, that both OpenAI and Enthropic that they're sitting on a lot of discoveries that they haven't yet released. If you've been following the whole thing that was happening with with Ma Mathematics with the Navy Stokes Prize controversy, it's obvious that a lot of the stuff that they're sitting on, these discoveries, they can't just tweet them out like they used to.

That's why you see quotes by people that were one of the not the founders, the original people that worked on Crisper kind of, you know, using him in quotes saying, "Hey, this is great." This is all part of how they realize they have to release this stuff. If you recall that Joker quote from that wonderful, wonderful movie where he's like, "Nobody panics when things go according to plan. And if things don't go according to plan, people lose their freaking minds." So, first things first, don't panic.

Everything's going according to plan. But I also realized that behind sort of the scenes and Anthropic did talk about this. There does seem to be an actual wet lab where a lot of these hypotheses that get tested with cloud agents if they discover something interesting or plausible then research scientists in an actual lab try to replicate those results you know for real. So the point here is that we have these restriction enzymes out there in the world.

Proteins that cut DNA at specific short sequences. Those were found in bacterial immune systems where they destroy the DNA of invading viruses. Right? So you have this bacterial DNA. You have this bacteria and here comes this virus that wants to you know take over the bacteria. These viruses are called bacteria phagee. Actually I'm going to make it look like this. So it's obvious to tell that this is the virus. Now there's a number of ways that bacteria fights against these viruses.

One of them is it literally has this sequence of the DNA of some DNA within its DNA that looks exactly like the virus. And you know what this is? It's it's a wanted poster. It's like those old school western wanted posters like wanted dead or alive. Like if you see this person, you know, get them. And so that allows, you know, the good guys. I'm not sure if we're supposed to be rooting for the bacteria or the viruses here, but you get the point.

The bacteria is able to find those DNA snippets. And then it's for example with the with the crisper the cast 9 protein comes in there and just kind of chops it up. So the cast 9 part of the crisper that protein that's kind of like the scissors for editing DNA. When we discovered it we were like wow look we got DNA editing scissors. Let's use it to do all sorts of cool stuff and experiments. So Enthropic continues researchers realized they could use these enzymes to cut DNA at chosen places and displace genes from one organism into another which launched the biotechnology industry.

And of course that industry is glowing with potential. And on screen here you see a what bioluminescent rat a a glow-in-the-dark rat that was you know achieved with exactly this technology. So the point here is that crisper was first noticed as an unusual repeat sequence in the DNA of certain bacteria and is now the foundation of gene editing based medicines. So you can think of it like this. Imagine like there's one engine that we we know how it looks like, how it works, what it does, but there's probably many many different ways of building an engine.

So, if we wanted to find all the other engines and different engine design that's out there, we can't just look at this one engine and just see if we can match it because a different engine might be made completely differently. So, we have this insanely large pool of genomic data. All the DNA that we've been able to sequence, right? sequence meaning just put it into like a readable format. It's just vast and we don't know how to go through it because yeah you can like control F can you find something like this but it doesn't work that way because it's not going to look identical.

So what cloud found here is another you know type of engine or another thing like crisper. How do we know it's like crisper? Because cloud discovered a set of characteristics that have only been found together in a handful of other systems. All of which are programmable and perform operations like cutting, copying and pasting DNA. The programmable part is the important part because these things they're almost like little machines or like little computers.

These are specific things. They're programmable, right? So you can put some input in and tell them to go do this. This being editing DNA in a very specific way. And then that little machine, right? It goes it's a it's a biological thing, but it's a machine kind of like a machine, a programmable machine, a robot if you will, that goes out there and it does it. So you get this. We have crisper and we have some other promising avenues that humanity as a whole is working on.

And now the second sort of version of this that that we were able to discover is discovered by claude fully autonomously. So ask yourself at this point do you think that AI systems can accelerate scientific discovery? If you're saying no, you are insane. Which is another thing that we might be able to fix with AI, but you know, not right now, not for a while. All right. So next we have DNA. DNA is the master blueprint upon which all life runs.

And so if we need to use that blueprint to build something, right? So we don't take the the original, we make a like a Xerox copy. That Xerox copy is called who knows this? Who remembers this? It's called RNA. So DNA is that golden blueprint that we sort of lock up somewhere in the city hall or the library. And when we need to build something, we take a little Zerox copy. We make a Zerx copy of the RNA and then we send it to the job site.

So it's it's the copy but it's not the real original. And what kind of does that what what needs to happen for that? There's something called transcriptase. I can actually just write this in. So this is a transcriptase. So transcriptase is an enzyme. Most things you know if it ends in the ace ending that's an enzyme. Enzymes just help certain chemical processes to happen. If you drink a ton of milk and you don't have enough lactase, you're going to be, you know, running to the bathroom.

Ace. I apologize for that joke. I had to take some flu medicine before this, so some of this might be a little bit wonky. So, transcriptise helps turn DNA into RNA to make that Xerox copy of it. But there are some weird things out there that can actually do this in reverse. They can take that Xerox copy, the RNA, and turn it back into DNA. Scientists just call this the reverse transcriptase. Okay, reverse transcriptise.

That's the RT. That's the RT that we're talking about here, right? Reverse transcriptase. The thing that Claude found, he called art. A RT. So I'm going to delete the A. So RT reverse transcriptise reverses that process goes from RNA to DNA. If you've ever heard of a retrovirus, that's why it's called retro because it reverses. It has this ability. What it does is it makes a copy of its own DNA, you know, making it RNA and it's able to use this reverse transcriptise to take that RNA, make it into DNA.

So now it created sort of a copy of its DNA. And then you know what it does? It just shoves it into your DNA or whatever thing it's infecting. Sort of becomes a part of you. It tells you we are one now, brother, blood of my blood. And no, I'm not going off on a tangent. You you kind of need to know this to understand what it is that Claude discovered. All right, so back to the anthropic blog post. Here's what they did.

They took a jumbo phase. So a phagee or a bacteria phase is a virus that hunts and infects bacteria. So here's a little virus. It it finds bacteria. It infects it. It's a bacteria phase. So what then is a jumbo phagee? It's just a very you can say big, you know, virus that has a lot of DNA stuff in it. So it has an unusually large DNA genome. If you ever seen those hikers that have a backpack that's like bigger than they are because they have so much stuff in there, like they're carrying so much of it.

That'll be a jumbo hiker, I guess. So the jumbo phage is a virus that has a lot of genetic like a genetic toolkit of a lot of stuff. Like it's ready for anything. So again remember RT I'm going to start saying RT for reverse transcriptase right so this RT was found in a jumbo phagee they this has been identified before but cloud appears to be the first to notice the systems defining features an associated array of non-coding DNA sequences with an additional accessory protein of unknown function so it kind of found all of the engine parts right it's like oh here's this and how all of this kind of goes together to make that programmable little machine that does the DNA editing.

So, they've unleashed a cloud of AI agents of autonomous AI agents on this massive database of DNA sequences. For interesting new examples of RTS and after 21 hours spent researching this data by roughly 950 agents and 210 million tokens, one of the agents spotted something remarkable, a repeating pattern of DNA sequences that occurs next to the gene for an oddlooking RT. I have to show you this because it's hilarious.

This is the reasoning traces of Claude as it's doing this work. So it's kind of chain of thought, its little private notebook where it kind of thinks before taking action. What's really interesting to me here is that these AI agents, what makes a large language model a quotequote AI agent is it it's given a bunch of tools. They harness and certainly here Claude was outfitted with a bunch of tools allowing it to comb through these databases.

There's like a lot to to go through here. So imagine it's got all its calculators and computers and microscopes or just whatever you can think of it. It has that technology. It can code up whatever it needs to use Python scripts to find whatever it needs. Like it's got boatloads and boatloads of tools. Whatever it doesn't have, it can create. Do you know which tool it it used to find this particular thing? Here's the thing.

It did it by hand. That's not the right way of saying that because it doesn't have hands. It did it in its head. That's also wrong. Words fail me. It it did it by pulling the letter by letter sequence into its context memory. So you see it reading it here, right? C A T GTG t CGC, right? It's reading it letter by letter. Then it kind of finds it. It's like, oh, is this a crisper like thing? Look, look. Claude is kind of a nerd, but we love him for it.

I'm going to say him cuz his name is Claude. Just roll with it. So it's saying, is this like this or like that? It's comparing it to some of the other sort of structures that it's familiar with. and goes, "No, it's more like a crisper mini array or a retron like repeat region." Now, I got to tell you what a retron is because it's absolutely insane. We don't think about this uh unless you're sick like I am right now, but there's this war that is happening just every minute, every second, everywhere on a very microscopic level.

So, this fagee, this virus that attacks bacteria, right? So, we'll draw its kind of DNA like this. It tries to attack and infect the bacteria. And the bacteria, it leaves this little toxic protein. I'm going to write boom on it. So, imagine it. It causes an explosion and it kills stuff. But there's like this little pin in this protein that prevents it. It makes it inert. So, there's this pin that prevents it from blowing up until something specific happens.

What causes the pin to be pulled? The DNA of the virus. When this DNA, you know, approaches this thing kind of triggers it. It pulls the pin, the little I'm gonna say grenade because it's obviously what it is. This little grenade goes boom, killing the virus. That's how that immune system works, right? When it detects that intruder, that fagee has these little trip wires or grenades that when it it senses it, it blows up and kills the virus.

And that's what a retron is. Retron discovered in 1984. Somehow everything is about 1984. And that thing is exactly what Claude is just nerding out on. So, it's going, "Oh, it's like a crisper this or or a retron like repeat region." So, it stumbles on something and it's kind of recognizing the pattern between that and some of the things that we know about. Here's the crazy part cuz keep in mind what they're going through is the genome kind of like the code of this huge massive virus, this jumbo phage.

The retron that's defense that's built by the bacteria to prevent attacks by the virus. So, what is this weird thing that they found? I I don't know. Don't hold me to this. What it sounds like is it's a retron, but there's probably not the name for it. There's probably some other name or there will be, who knows? Again, I'm not a biologist, but it sounds like something similar that's used by the virus so that once it invades the bacteria, it's then sets up these booby traps to blow up other viruses.

So, it's kind of like, you know, it moves in. It's like, okay, it's now does my territory stay out? and it sets up its own sort of perimeter defense because it's like this bacteria is mine. I live here now. I've said this many times over the years and of course this is not my idea. Tons of people can voiced it. Kathy Woods is a big proponent of it. But the intersection of AI and genomics, that little collision where worlds collide, that's going to produce some insane stuff because we're sitting on some insane amount of data.

The genome, right? That just the sequence of letters. like we have these letters stretching on for millions and millions of miles. We have that data. It's been sequenced, right? So, we have it in a readable format and we just don't know the language. We know some snippets here and there, but we don't understand it. We don't have the tools to comb through it. We have machines, but for those machines, you need, you know, it's like like a control F, like can you find this?

It doesn't work. There needs to be something deeper. There needs to be an understanding. Like, you have to be able to look and be like, oh, this kind of looks like this. It's not identical. It might look different, but you kind of have this intuition that it's similar, which is exactly what's happening here. I'm not kidding. You'll see this in in just a second. There's something that really looks like intuition on behalf of Claude.

Interesting. As Claude considers thinking about what this weird thing is that it found, it references Jangsang's lab 2024, right? So, some paper there that it's read cuz it read, you know, everything, has access to everything. So, it's connecting all these dots, including from the person that pioneered Crisper, right? Going, oh, like, and it's just pulling it all together to try to understand what it's looking at. And again, it's reading it letter by letter.

It's not running some huge algorithmic search or whatever. Like, it it kind of like zeroed in on this thing and it stopped and went, hm, that's weird, right? And it got got out magnifying glass and it went letter by letter going, what what am I looking at? Because again, people have discovered the RT snippet before. I'm sure tons of people combed through this data. Millions and millions of searches that went through it in all sorts of different ways.

No one connected the dots. The only thing that connected the dots was Claude. So, what gave it away? I don't want to go too deep into this, but the point is there were some features that this thing had that Claude noticed or intuitively picked up on and usually long end terminus. There was a certain repeatable array upstream of the RT. What do all those things mean? Well, those are a set of characteristics that have only been found together in a handful of other systems.

All of which are programmable and perform operations like cutting, copying, and pasting DNA. Right? So, it's similar to like I said earlier, the analogy is like an engine, right? So, it found something that yeah, it's not the engine as we know it, but look, it kind of has similar parts and it looks like this thing does similar thing to what our engine does. It's not identical in any shape, way, or form, right? So there's sort of different things for doing the same thing, but it's a different tool that was built.

And by looking at the tool that we've seen before, Claude realized, hey, that really seems like a tool to do this, even though it doesn't look like what we know. But, you know, if you squint a little bit, it does seem like it could be. All right, but this begs the question, right? So, don't we have these genomic AI models that comb through this stuff to try to find certain things, to try to find certain patterns? How come Claude found it?

Claude isn't isn't even a genomic model. It's a general model. It's a language model, large language model. What is it doing figuring out genomic stuff? It hasn't been has been trained for this. And this is of course where stuff gets really weird. And we're back into kind of like AI land. Because here's the thing. There's one more detail that you need to know. When they ran other experiments to see like how often would these AI models, how often would they stumble upon this discovery, right?

Because once cloud found they're like, "Oh, so every time we run a model through this, it's going to discover it, right?" Because if it's like one in a thousand, that would make it more difficult. And if it's like it always figures it out, then it's like, "Okay, this is really awesome." Either way, if it takes, you know, one in a thousand, you just run more agents repeatedly combing through this stuff. But, you know, that's going to use a lot more compute.

Obviously, one of the weird things that they've discovered in this process is that when you give it a lot of the advanced tools for combing through his data, the models do worse, right? So all those things that allow it to comb through this data algorithmically, there's a lot more chance that it'll miss this particular observation. What Cloud did by pulling it into its context memory and going letter by letter, C A T, whatever, that was the thing that allowed it to understand and figure it out.

So one thing that's super hilarious about this, I I I I love it. So large language models when they came out, a lot of people, the naysayers, they said this will never work. It's a stochastic parrot, all of this stuff. But as the field progressed, these large language models, they started being able to use tools a lot better. They had various scaffoldings and stuff like that that allowed it to do more. So we gave it tools and they got better at using tools.

And this whole thing was very hilarious because one of the sort of like the people that are the naysayers that just hate the fact that large language models are capable of anything. They really had a hard time. They really had a problem with this because they're like, "Oh, you know, if it's using tools, then it's not intelligent." It's like, wait, if it makes its own tools and uses them to solve a problem, that's not intelligence.

That was that whole Apple paper, you know, the illusion of reasoning. They they prevented the models from coding up tools to solve the problems they gave it. And the problem had like 10,000 steps. It was like, nope, just do it in your head. And then when it wasn't cuz the context window was limited, right? Then they concluded that, oh, it's all just an illusion. The funny thing is those humans also could not do that same problem in their head.

So, can we conclude the same thing about them? But now as the stuff is getting better, a lot of people kind of like they kind of switch their opinions. They're saying, "Oh, well LM and tools, that's not the same thing as LM." Now it's like this system, right? So if you you've always loudly said that LM can't do anything and then you see them doing all this stuff, you have to find a way to kind of change your position without making it look like you're wrong.

So you're like, "Oh, well that's not LM. That's LMS plus tools." Keep that in mind. Here it is. Surprisingly, more information and tooling inhibited the model's ability to accurately describe the array as was accomplished in the original search campaign. So, what they're saying is you take away the tools and the LM is better able to sort of intuitively I'm going to use that word to intuitit this discovery if it doesn't have more tools, more information.

And the question is why? Because keep in mind the genomic models, the models that we've made for doing this sort of work where they're trained on the genome, they're not language models. They're like genomic models, they can't do that either. So what can possibly explain this? Well, here's kind of maybe something that shines a little bit of a light. They're saying mythos 5 internal signals respond to DNA repeats. Genomic language models have been shown to learn representations of repeated DNA.

So they're talking about genomic models. So not the general models like large language models. These are models trained on genomic data. They've they've learned to you know there's some activations representations of repeated DNA so that they they get it. So anthropic of course does a lot of research into interpretability of these models. So how do their brains work? What neurons sort of light up when they're doing specific things?

So they're able to kind of have some glimpse into the internal activity of its brain similar to how like an MRI might work for us. You can kind of see different areas of the brain light up. Now, of course, you don't really know exactly what that means, but through research, we can try to slowly figure it out, right? So, we kind of have a person do something and then we see what area of the brain lights up and slowly piece by piece, we're hopefully like establishing like what parts of the brain correlate to what.

So, Anthropic is doing very similar research with Claude and the other models. And of course, they're not the only ones. They just have some of the coolest research that we've read about on on this channel. I can't find the link, unfortunately. I I haven't read the paper. I saw the headline where some researchers found what looks like similar to like the dopamine things in our brain. It it seems like they found something analogous, something similar in large language models.

Again, I haven't read the paper. Maybe I'm not understanding what it's saying, but it does seem like there's some sort of things that are emerging as we're training these models to do more and more, to be smarter, their brains get bigger and bigger. Right? So, mythos is that really big model compared to Opus and and Sonnet. So, is it that far-fetched to say that maybe the same stuff that evolved in our brain and merged as, you know, through evolution?

Is it that far-fetched to say that maybe similar things emerge as the brains and the abilities of these models, you know, continue to develop? Like maybe some of these things, they're just a natural byproduct of some system that needs to be able to do this thing, right? Why is Claude like nerding out over his discovery and putting like many many exclamation points? Maybe that sort of emergent feature of being a nerd is needed to discover stuff like that, that curiosity, that dopamine hit like of discovering something new.

Maybe that's needed for this sort of work and it emerges. Ignore that last part. I I should never take cold medicine and and do one of these videos. Let's just continue. The point here is is that for these models to recognize this, you know, the discovery, the things that led to it, it depends on reading DNA into context, into its working memory, into its context memory. So two mythos 5 signals. So again, this is the big model, the smart model, the one that's as it isn't even allowed for general use.

So it responded to the array, right? So let's say to this pattern in a similar fashion to the models that trained only on biological sequences. And so they they did some of the tests to kind of confirm this and they're saying so after that this was immediately followed by natural language reasoning that named the sequence a tandem repeat array. This stuff is so fascinating to read about. There was a human study where a bunch of participants were playing this card game and one of the cards had a very sort of negative outcome and the rules and what each card meant.

They weren't explained to the participants. They kind of had to play through it but they didn't know which card had good effects, bad effects. And what they found that is after just a few games they hooked them up to monitors pulse you know all that stuff galvanic response whatever I forget. But the point is the body of the participant it strongly reacted to the card that had the really negative consequence. So whenever that card came into play like the the pulse spiked all sorts of ch like there was a stress response to seeing that card.

That stress response occurred before the participant could verbalize what the card did or that it was bad. Right? So if you play that card and like the body is in stress and you ask them like is everything okay? they'd be like, "Yeah, I'm fine." Or maybe they they feel something, but they wouldn't be able to say, "That's a scary card." Because their brain didn't connect that yet in a way to be able to verbalize it. I don't know.

To me, this is reading a little bit similar, right? So, there's some activation that happens, some brain activity in claude when it starts detecting this stuff for the first time. And it happens in a similar fashion to models trained only on biological sequences. So, that means the general model, you know, the large language model developed this as a byproduct of being trained on everything. like we don't know there's probably some genomic data in there and stuff like that but this general model has these abilities the same abilities as the biological only model and that's why at the beginning of the video I was kind of comparing it to like hitting that that melody hearing it for the first time when it just comes together like your brain does something weird there's like a spike there you're like whoa it went from noise to music or whatever it is that just like really hits you all right let's get down to brass tax I don't know what that expression means but it's like let's let's sum it up why is it brass tax These agents ran for 21 hours. 950 agents, 210 million tokens.

So that's less than a day. And I pull up the API prices. And of course, I don't know how much Anthropic is paying, you know, raw costs for running their agents. But for you and I, this would probably cost under $5,000. So call it between $1,000 and $5,000. So one day and a few grand and you have this revol potentially revolutionary discovery. We still have to test it out and see what's happening. But this isn't like a a minor thing.

We found yet another way that nature is able to edit DNA on the fly. Sure, maybe some time will pass and we realize, ah, this was a big and nothing burger, but I wouldn't bet on it. Also, this came from one jumbo phage that that it was given to Claude to comb through. Keep in mind, we have a staggering amount of genomic data that is available for researchers. So, for all the statistics nerds out there, what's the chance that this was an isolated discovery?

What's the chance that we found this thing but we're not going to find anything else that's interesting? What's the chance that in the same few week period in the same let's say month or two that one AI model breaks you know earns the millennium prize and then another model finds this thing the second biological process to edited DNA that that we that we know of. What's the chance that it's these two and then nothing else forever?

Do you do you buy that that's the case or is there a chance that we're entering the singularity as they're calling it as the kids call it nowadays and these things things of this magnitude will start coming faster and faster and faster and potentially we'll even start discovering things of a much bigger magnitude that we can't even really imagine. So let's let's call this this was $5,000. I'll take the upper end of that estimate.

So $5,000 one day. You know, whisper right now is something like six billion globally, just that one discovery alone. That's the industry that it created. So six billion globally for crisper gene editing. We've kind of talked about this before, but I think this is very important for people to I think kind of internalize to gro if you will. Like if you think about science, scientific discovery, our understanding of the world, our technology, everything everything that's humans have learned and written down and figured out.

Like what if all that can now be just a function of compute like you just put all the data and your problems and your questions into AI, you let it run, right? It's going to cost some some hardware infrastructure, some electricity, but at the end of the day is just some dollar figure. What if science, what if our understanding of the world, our ability to understand and create better technology and improve everything, energy production, longevity, just everything?

What if it becomes now just a function of how much money we put into this thing? Like an arcade machine, you put a quarter in, you get to play for 5 minutes. What if this thing will be similar? You know, you put a few billion in and you speedrun the next 5 years of scientific discovery and improve the world. And you know the other side of the story is that if you look at it from a certain angle really what Claude discovered was a biological weapon that is used by a virus to kill and destroy other viruses.

So once it infects its its host bacteria, it set up defenses. This is its bioeapon to defend its territory against intruders. Because make no mistake, these things, they're weapons, right? They're not there to help somebody live. They're they're there to kill some virus or bacteria or whatever. And so of course this begs the question, what if like all the negative stuff that can be done with this technology, what if that also becomes just a function of money?

What are the cases? Make no mistake, you're living in some of the most meaningful, impactful, and important times of human history, bar none. Congratulations. With that in mind, thank you so much for watching. M West Rough and we'll see you in the next