Latest / Bad Decisions Podcast / Anthropic FIGHTS with Trump & REPLACED by OpenAI in 24 hrs
Transcript
- 0:00What is up, beautiful people? Welcome back back to the Bad
- 0:03Decisions podcast. Where we come back every Monday,
- 0:06every Wednesday and every Friday, bringing you guys the
- 0:09hottest news in the world of technology, AI and business.
- 0:14And it's Monday and we are back with another episode.
- 0:17Hope you guys had a great weekend before we.
- 0:20Start. As always, I want to say thank
- 0:22you for tuning in. The fact that you guys come back
- 0:25to us every week, it means a lot to us.
- 0:28We read your comments, we know you're following the show, so we
- 0:32just wanted to say that it means a great deal to us.
- 0:35So for today, we have two very important topics to discuss.
- 0:38Usually we have more than two. But these topics are really
- 0:42expensive, requires debate. Yes, exactly.
- 0:45So the first one, Anthropic and the US government went into war
- 0:51in the past few days. What is the second one?
- 0:54The next one is going to be interesting.
- 0:55We have a biological computer with human brain cells that
- 1:01learned to play Doom, the game that we played when we were
- 1:05kids. So imagine human cells on a
- 1:08computer that learned to play Doom.
- 1:10It's insane. Ethical questions in both the
- 1:131st and 2nd title of today. That is why we only chose 2
- 1:18because we need all the time for the debate.
- 1:20Get ready. Get your popcorns, get your
- 1:23juice, get your Coca-Cola, get your red coffee, get your
- 1:26coffee, get your tea, get your ice cream, get your kebabs, get
- 1:31your shawarma. Get what else, Shawarma.
- 1:34Yeah, I mean, just like your Middle Eastern shawarma.
- 1:38Shawarma, shawarma. Oh my.
- 1:42God, I just listed as my get your pizzas, get your burgers,
- 1:46What else? What else is good?
- 1:48You mention everything coffee. Kalis Kachapuris for my Georgian
- 1:53guys. Is there anyone from Georgia?
- 1:54Watch. I don't know.
- 1:55I don't know. You wanna start?
- 1:57Let's go. I'm already hungry.
- 1:59Today's Monday and what happened over the last few days starting
- 2:04from Thursday, Friday and these CEOs were working on Saturday to
- 2:08make all of this happen. And you're going to go to the
- 2:10timeline of the story. So on 26th February, the 1st
- 2:15event appeared on X Dario Amade, the CEO and Co founder of
- 2:20Anthropic, the parent company that developed Clod posted an
- 2:25article and it states as statement from our CEO in the
- 2:30discussion with Department of War.
- 2:32I will summarize it for you guys and this story is Entropic was
- 2:36the first AI company to work on non work on classified projects
- 2:43for the government. So there were open AI and other
- 2:45companies who worked on non classified project.
- 2:49But Anthropic was the only one and the first AI company to work
- 2:52on the classified projects. What happened was they gave
- 2:56their models to the government entities.
- 2:58They deployed it into the national laboratories.
- 3:02They created custom models for the national security customers.
- 3:06The the cloud system has been extensively deployed throughout
- 3:10the government with two guardrails. 1 is mass domestic
- 3:16surveillance should not be done. Also the models should not be
- 3:21used for fully autonomous weapons.
- 3:24So they imagine they gave it to the government, they said do
- 3:27whatever you want to do, you have our most powerful models.
- 3:29Just don't do these two things. Don't use it for mass
- 3:32surveillance. Don't use it for a fully
- 3:34autonomous weapon. And if you're using it for
- 3:36partially autonomous weapons like Jones and all which was
- 3:39used in Ukraine, it's fine. What happened was when they
- 3:43released that these models to the government, the government
- 3:46said no, we want the ultimate power, we want the ultimate
- 3:49control. Take, take off those guard lays,
- 3:51take off those two. We want to test everything.
- 3:54This is where the anthropic team, of course, the leadership
- 3:57team said no, we want to stick to these two conditions.
- 4:02Why? Because they believe that first
- 4:05the today's frontiers AI models are not reliable enough to be
- 4:10used in the fully autonomous weapons.
- 4:12Also for the mass domestic surveillance, they said it
- 4:15violates the American Constitution.
- 4:18So they don't want to do that. The moment they started posting
- 4:21these articles on X the Secretary of war, funny enough
- 4:26he went on X and he said our position has never beveled and
- 4:31will never waver. The Department of War must have
- 4:34full unrestricted access to an tropics model for every lawful
- 4:39purpose in defense of the Republic.
- 4:41Basically they were saying they cannot allow a privately held
- 4:45company make decision for a democratically elected
- 4:49government. They said you either have to
- 4:51take out everything or you will be a supply chain risk.
- 4:54Trump wasn't that kind. He went on Truth Social, which
- 4:58he was his, it's his own social media platform and he said we
- 5:02are not Pat, can you pull up the yes, this is the tweet from
- 5:06Trump that he said we are not going to allow a radical left
- 5:10work company to dictate how our great military fights and wins
- 5:15wars. Basically you have to give us
- 5:17everything. And in the same tweet, he said
- 5:19that we are going to stop using Anthropic.
- 5:22They are banned, they are supply chain risk and we are going to
- 5:24find a new AI company to help the American military.
- 5:29Funny enough, two hours after that, Sam Altman posted on X
- 5:35that tonight we reached an agreement with the Department of
- 5:38War to deploy our models in their classified network.
- 5:42Now remember, open AI was working with the government.
- 5:45So it wasn't an overnight decision.
- 5:47They were working on the non classified project.
- 5:50But immediately when things got heated up, probably Sam Altman
- 5:53to the government say I can do it.
- 5:55I can help you guys. I will do it for you.
- 5:58So Sam Altman, I mean, they don't have.
- 6:00Many other choices right it's Google anthropic XAI open AI
- 6:05that's. It that's pretty much it, right?
- 6:07But I think that's a big win for open AI right before before I
- 6:11continue the story first Farris, what do you think?
- 6:13By the way, this happened over Friday and Saturday.
- 6:15So Sam Hoffman reached an agreement with them on Saturday.
- 6:20OK, so. You want to get into it.
- 6:22Yes, let's get into it. Here's the stuff that I've been
- 6:25thinking about. There is things to discuss on
- 6:30both ends, the government side and Daria Amode side.
- 6:33So before we get to Sam Altman, let's talk about that Daria
- 6:37Amode protecting 2 guardrails, autonomous weapons and mass
- 6:41surveillance. Those are pretty common sense
- 6:46guardrails you want to have for AI models today.
- 6:50So we could all agree that Daria Amode's sense of protecting
- 6:57humanity through setting those two guard rails is the right
- 7:00way. You yourself know how some of
- 7:04these models to this day hallucinate.
- 7:07Sometimes you're asking them a common sense question and they
- 7:10will give you wrong answers, answers that are not factual,
- 7:14and you have to repeat yourself to then get the factual
- 7:17response. These are the same models,
- 7:19perhaps slightly more advanced, that are being deployed into a
- 7:22military now. We're talking about humans lives
- 7:26here when you're thinking about autonomous weapons.
- 7:29People's even talking about friendly fire.
- 7:31Can you rely 100% on a machine to make a decision for human
- 7:39lives? Even human beings themselves
- 7:42sometimes suffer with judgement. Even today we have problems in
- 7:47America or other countries where police shoots down a human.
- 7:52Perhaps they're a year too young.
- 7:54They're not yet 18 years old. They're 17.
- 7:57And it goes all over the news because a cop shot a teenager.
- 8:02Now imagine the goal for the autonomous machine is do not
- 8:09kill innocent civilians, but you can kill immigrants like illegal
- 8:14immigrants. But I think this is for wars.
- 8:16This is not for you don't wanna let these autonomous weapons in
- 8:20the city, right? It's not for a civil war, but.
- 8:22War is the first level and then eventually.
- 8:25No, but the biggest. From that.
- 8:26Argument I saw Palmer Luckey and all the other CEOs founded very
- 8:30childish of Dario and saying that China and Russia are really
- 8:33doing it. There is no point of us.
- 8:35I'll get to that. I'll get to that.
- 8:38The whole point is setting up a guardrail so that we will never
- 8:43allow machines to make decisions on human lives.
- 8:46That's Dario's point that I support.
- 8:50I think it's important to protect human societies by not
- 8:57allowing machines to have control over who lives and who
- 8:59doesn't, because it starts here, it starts with war, but
- 9:03eventually it's going to go deeper because these same models
- 9:06will eventually be deployed into.
- 9:08Police, but they will be deployed.
- 9:09RoboCop. Exactly.
- 9:12So if you already deploy and give the models Full disclosure
- 9:15full right to make a decision on a human life, eventually the
- 9:19robot cop will have the same model.
- 9:22RoboCop. RoboCop.
- 9:23You get where I'm going. Yeah, of course.
- 9:25So what Dario is suggesting here is that if we start now by
- 9:29giving fully autonomous systems the ability to decide on human
- 9:34fate, where is this all going to go in the future?
- 9:36If you don't draw the line right now, where is it going to go?
- 9:40I'm going to get to my next point.
- 9:41Allow me to finish. Allow me to finish.
- 9:43So I understand where he comes from with autonomous systems,
- 9:48mainly because we cannot rely on their judgement because we
- 9:51sometimes, as humans have problems with our own judgement.
- 9:54How can we go ahead and rely on an AI model that we've half
- 9:59trained? It's not even fully developed
- 10:01yet. The AI models are still like
- 10:03children. They still need a lot more time
- 10:05to catch up to be at our level of judgement.
- 10:09They're smarter in many other ways.
- 10:11They're faster in many other ways.
- 10:13But in terms of human judgement, they're still not there yet.
- 10:17What he's suggesting is that it's dangerous to rely because
- 10:21how can you tell an autonomous system what is the meaning of
- 10:25innocence? Do not kill innocent civilians.
- 10:28Innocent civilian could mean a completely different thing for
- 10:31you and me if you live in different countries.
- 10:35Innocent is a subjective term. How do you tell it?
- 10:38It's like prompting an image. Let me finish my point.
- 10:41I didn't interrupt you. This is like a presidential
- 10:44debate. Do you see that?
- 10:46And I'm using my hands here. You made me lose my point.
- 10:53I didn't say anything. I know, but the fact that you
- 10:55were about to interrupt me. When you look at prompting for
- 11:00an image, the biggest tip today is to be as specific as possible
- 11:07with your prompt. If the person using the machines
- 11:10is not specific with their prompt, it can lead to a
- 11:15catastrophe. Because if it's not specific
- 11:18about who to destroy and who not to destroy, it can make sub
- 11:22goals for itself, decisions that may lead to a catastrophic
- 11:27event, which is Daria Ahmadi's first point.
- 11:30The next point, if you want, I can get to it and then you can
- 11:33defend or you want to get no. For the first one, he's dealing
- 11:36with secretary of war, he's dealing with military.
- 11:41You are not dealing with police. I know.
- 11:43So you are dealing with, I mean, we are not Americans, I don't
- 11:46live in states. If US goes to war with China and
- 11:49Russia, do you think China and Russia cares?
- 11:52They already have access to the models by the way opening.
- 11:54I gave access already after 48 hours.
- 11:57I was a textbook mistake. I will get to that, he.
- 12:00Probably lost $200 million and that's it.
- 12:02By the way, we will go Pratt, can you show Katy Perry's tweet?
- 12:06He lost the 200 million deal, but a lot of people switched
- 12:11from ChatGPT to cloud because of his.
- 12:13I mean that was APR move this. Is not a concern.
- 12:16Like that's APR move. No, no.
- 12:17I think he has a huge inbound of people coming in.
- 12:21I'm talking about existential threats.
- 12:23You're talking about cloud subscriptions.
- 12:25No, no, I'm talking about revenue.
- 12:27I'm talking. What about existential threat?
- 12:29I'm talking about dystopian potential for machines
- 12:34controlling humans, and you're talking.
- 12:35About $800. No, I'm talking about Dario.
- 12:38No, Dario Madei's consequences of what he did.
- 12:41Yes, once I'm telling you he was dealing with the Secretary of
- 12:45war. This is for military, this is
- 12:47not for civil cases or this is not for something that they're
- 12:50going to use in the street. This is for that's.
- 12:52Where I think you're wrong because it starts at a military
- 12:56level and then it distills and it, it comes down to a society
- 13:00level. I, I told you I'm going to get
- 13:02to my point. So the next one is mass
- 13:05surveillance, mass domestic surveillance, because we know
- 13:09since John McAfee times, he was talking about how there's
- 13:14surveillance happening, you know, in different countries,
- 13:17different times. And he was warning people in
- 13:202005, Edward Snowden, and I don't know when it was exactly,
- 13:24but a long time. Ago I wouldn't.
- 13:25They didn't without AI. They didn't without AI.
- 13:27Yes. So with AI it's the same thing,
- 13:29but smarter, faster. Yes.
- 13:31What Daria was saying is that I think it's not democratic for us
- 13:36to allow that if the government wants to do it at some point.
- 13:40Again, if you're thinking about the Constitution of Americans,
- 13:44I'm not an American, but I've watched too many movies to not
- 13:48know this. If you think about the
- 13:49Constitution, they have the right to not be surveilled at a
- 13:54level where they are being heard.
- 13:56Watch and observe. But they are the most surveilled
- 13:58probably. Which is the case?
- 14:00Probably, right? Therefore, I wanted to start my
- 14:04argument by saying I understand Dario's point, but do I think he
- 14:12had any other choice? And do I think that even if he
- 14:16stands against it, it's going to prevent the government from
- 14:19doing it? No.
- 14:20Now I'm going to go to the government side.
- 14:22I completely agree with the Department of Defense, which is
- 14:27not Department of War, that a privately held company shouldn't
- 14:31make a decision for the entire American population.
- 14:35The government was elected to be the government.
- 14:39Daria wasn't elected to hold one of the strongest AI models, he
- 14:44just happened to be the Finder founder of the company.
- 14:47So for a single company to come and strong arm the government
- 14:52and say I'm not going to give you powers, it's probably not
- 14:55the right thing that I agree with.
- 14:58I believe ultimately the Department of Defence,
- 15:01Department of War has to have the final say in how they build
- 15:05the defence for the countries. And I agree with you. the US has
- 15:09a lot of enemies and the enemies aren't just going to wait
- 15:12because those enemies, most of them are not even democratic
- 15:16countries. They're going to build the
- 15:17technology and use it however they seem fit.
- 15:21And so if we're talking purely from a competition standpoint,
- 15:24and if you look at history, if a king wanted the best technology
- 15:27of Canon warfare, he's going to get it.
- 15:31No one's going to come and say I'm not going to give it to you
- 15:33because you're going to use it the wrong way.
- 15:35It's just a little different now because we live in a more of a
- 15:41global world. People, people don't consider
- 15:44each other just from a country perspective.
- 15:48We all feel connected. When something happens in the
- 15:50US, everyone feels the ripples. And I really believe the reason
- 15:54why so many people are part of this conversation is because we
- 15:57all know whatever happens between the US government and
- 16:01Entropic will impact how AI is being used everywhere in the
- 16:05world. It doesn't only impact the US,
- 16:08it literally impacts everyone. I think, but entropic is a very
- 16:11small element of the entire Usus and the government basically
- 16:19their tactics right. So they just got replaced and
- 16:22open AI was open AI was hired in in less than 24 hours yes.
- 16:25So let's so even I would say this was a move PR move.
- 16:29Probably he he wanted to go on X and say that, oh, we are going
- 16:34to do this too. And then he was probably
- 16:36expecting the government say, OK, we are not going to do this
- 16:38too keep the guardrail and we are going to keep using your
- 16:42software. I think this may have all
- 16:44started after Madora's attack because I remember the post that
- 16:50was being spread online about how clawed Anthropic was used in
- 16:55the process. And I believe this was Dari
- 16:59Amadei being pushed by his employees, by his employees, by
- 17:02the people to answer and respond to the fact that this may or may
- 17:08not be false or may be true. And he didn't know how to
- 17:11respond. So he said, yes, it is true, but
- 17:14we have guardrails. And when he talked about the
- 17:16guardrails, the US government was like, no, we don't.
- 17:19We don't want guardrails. What are you talking about?
- 17:22I do want to touch on a couple of things.
- 17:24This is very interesting. The retaliatory move by Trump
- 17:29and the Department of War calling Anthropic a supply and
- 17:33chain risk and a national security risk technically means
- 17:38that they said this company is a spy because this usually is a
- 17:42label that they have for. Foreign entities.
- 17:44It never happened that they gave it to an American company.
- 17:47Exactly. So technically they're like, no
- 17:49one in the government can work with Anthropic because they are
- 17:51spies, and that will hurt Anthropic's revenues.
- 17:57But now we have Katy Perry paying for a cloud.
- 17:59Subscription No. A lot of people.
- 18:00No. No.
- 18:01Perry's paying for it, then they're good.
- 18:03That's $200 for a $200 million contract.
- 18:06I was curious about the quick switch over to open AI right
- 18:10after Anthropic and you said it happened within two hours.
- 18:13No 24 hours. 24 hours, still very.
- 18:15It could be short. Yeah, it could be two, but very
- 18:17short. The reason why they were against
- 18:21the Anthropic deal was because Anthropic's version of their
- 18:25contract was a rigid model level lockdown.
- 18:28So the safeguards were hard coded into the AI in a way that
- 18:33it was absolute and non removable.
- 18:35Think of a door welded shut. If the Department of Defense
- 18:40needed to use any of those systems.
- 18:43So surveillance or the autonomous weapons, even for
- 18:46testing they couldn't. Whereas with the open air
- 18:50version, it's more of a layered smart lock and a collaborative
- 18:55system. Based on my research, it's a
- 18:57three level lock. It prevents blocks from bad
- 19:01queries. So the queries that you would
- 19:03have when you ask ChatGPT like hey tell me about.
- 19:09I'm trying to think. I'm trying to think the item's
- 19:13coming into hand and in real time, he's removing it.
- 19:16Not suitable for the show. Not suitable for the show.
- 19:19Not suitable for the show. Let's just.
- 19:22Say let's just say Epstein related stars, ChatGPT would be
- 19:25like, I cannot answer that, yes. So that would be the first level
- 19:29of the lock. And then say there is a reason
- 19:32why they're using it. They have a personnel, vetted
- 19:37personnel from Open AI that will be monitoring the usage of the
- 19:41systems in real time to then be able to interrupt it in case
- 19:45there's a bad query going through.
- 19:47And they can do that because it's it's built on the cloud,
- 19:51which begs the question, how safe is it building this thing
- 19:54on the cloud? It's.
- 19:55Built on the. Cloud, Yeah, it's a cloud set
- 19:56up. This is part of the deal.
- 19:59You can research it, yeah. So Are you sure bro banks will
- 20:04never allow cloud set up. I was looking at it.
- 20:06Military is a cloud. I know.
- 20:07Not the whole military no, but the model.
- 20:10This. Version of the model I don't
- 20:12know. Man it really cannot be cloud
- 20:13set up bro in terms of security. But can you look it up?
- 20:16Can you look it up if you've got access but?
- 20:20We're gonna check it. Yeah, this is interesting
- 20:23because the guardrails still remain.
- 20:26Sam Altman explicitly mentioned that we still follow and we
- 20:30believe in what Daria Almaday said.
- 20:33There still should be no mass domestic surveillance and no
- 20:37fully autonomous weapons. But under very specific
- 20:43circumstances, under lawful purposes, there is a chance that
- 20:48the government can go ahead and get into the Gray area.
- 20:53They just, yeah, they, they won the contract to $100 million for
- 20:57Open AI and they have now classified projects with the
- 21:01government. Even if Open AI said no,
- 21:04somebody else would have said yes X, so somebody eventually
- 21:07would say yes. XX AI And by the way, there was
- 21:11also another consequence that could be possible.
- 21:14The government could go to NVIDIA, Microsoft, other these
- 21:18company to deinvest from Anthropic to take their money
- 21:21out. The traffic could go bankrupt.
- 21:24That's no. Good.
- 21:24So again, you never go into fight with the government.
- 21:29If you're doing a project with government, you just accept the
- 21:32terms and you come out. What would you guys say about
- 21:35this? I mean let us know in the
- 21:36comments. It is incredibly complex.
- 21:41This is no simple problem, nor can you find a simple solution.
- 21:47It involves most of humanity. Of course this is very US based,
- 21:52but because US has an impact economically, around the world
- 21:57and in many other ways, I think it's important that we all
- 22:01understand where this is all going.
- 22:04We'll just have to wait and see if Robocop's going to appear
- 22:06sometime in the street. RoboCop was good though.
- 22:09But bro, RoboCop was not from military.
- 22:11RoboCop was for the police force.
- 22:14Police force this is this. Is the RoboCop comes after all
- 22:16of this, right? Actually, they had none of the
- 22:19none of these systems. They just had RoboCop.
- 22:21Yeah, yeah, yeah. Because a cop died.
- 22:23A cop named Murphy. What is your name, son?
- 22:27Murphy OK, next topic we're definitely.
- 22:31Aging ourselves right now by singing.
- 22:33No, I could have watched. I could have watched RoboCop as
- 22:36a person whose parents bought the CD for so I'm not that old.
- 22:42What? I'm just saying, my parents
- 22:44bought it for me. That's why I watched RoboCop.
- 22:46It might be an old movie, but I watched it when I was young.
- 22:48Do I make sense? No, not at all.
- 22:52You're a boomer, accept it and move on.
- 22:55Pratt. Did we find out if it is also
- 22:57runs on cloud? Not all cloud models.
- 23:01I'm talking about the opening ideal with department.
- 23:05No, no, you cannot get it from X Pratt.
- 23:08Or you can actually if he has real time data.
- 23:10He does have real time because Sam Authman did an AMA, so
- 23:13probably it's in there. From what I found out that it is
- 23:17one of the parts of the deal which made the deal work on both
- 23:22ends, following the guardrails, making sure that there is
- 23:25oversight. You know who's gonna oversight
- 23:28it? Russia.
- 23:33This is all really fun to talk about.
- 23:36It's true. OK, we're going to move on to
- 23:39the next topic, which is a little lighter.
- 23:41I hope we have a dish of human brain cells that are playing
- 23:47Doom. When you say when you say dish
- 23:50of brain human cells, I think about Kalapala.
- 23:54So for those of you who we are not Middle Eastern or Iranians,
- 23:58there is an Iranian dish and I think they have it in Turkey or
- 24:01other countries, which they serve ships head and it's cheek,
- 24:06brain, tongue. It's I know it sounds
- 24:08disgusting, but it's actually quite tasty.
- 24:10Tasty other. Countries have a.
- 24:12Similarity. Yeah, Turkey has it.
- 24:14So don't be like real. It's good for you.
- 24:16It's healthy. It's very tasty.
- 24:18Yeah. So dish of human brain.
- 24:21It's more like catapult. It's something that I eat.
- 24:23Let me not that though. This is not an actual brain
- 24:26sitting in a dish, so it's think more of a liquid that has human
- 24:30brain cells in it that is also attached on top of a silicon
- 24:36layer. So can a patcher, by the way,
- 24:38so. People can see.
- 24:40And I think that's more yes. My goodness, I'm hungry.
- 24:43Go to images. Go to images.
- 24:47Yes. OK, can you pull up?
- 24:48Oh, my God. That's the literally the sheep's
- 24:50head. You're definitely gonna get
- 24:51demonetized for this. Yeah.
- 24:55So you'll find a brain somewhere there.
- 24:57Somewhere there. But this is there's a lot of
- 25:00different cells here and definitely not conscious or
- 25:02intelligent. You cannot run shit.
- 25:12You can just sip on the juice. I'm going to connect the USB to
- 25:24it. I'm really not trying to make
- 25:33the lab who made this technology.
- 25:36It's just. That imagine they tune into this
- 25:41video and the first thing they fart is.
- 25:58OK, so the name of the lab is Cortical.
- 26:02I don't know. Actually, it's not our fault.
- 26:04They could have named it much better.
- 26:07When you call your technology a dish of human brain, no not
- 26:11knowing in this part of the world, they eat the dish.
- 26:13Nothing. I call it the chips, They called
- 26:16it the dish. They didn't call it.
- 26:18There was the article. No, it's dish brain.
- 26:20Dish Brain. Was the first dish and brain.
- 26:24You ruined my topic man. This is supposed to be serious.
- 26:29OK, Take a deep breath. Everyone together.
- 26:31OK, We're going to take a deep breath.
- 26:35OK. Cortical Labs.
- 26:38This is an Australian biotech startup specializing in
- 26:42biological computers. Think machines mixed with
- 26:48organic human elements. What you're looking at right
- 26:53there is human brain cells with silicon mixed together.
- 26:59Like actual neurons? So when you're running your
- 27:02computers like this Dell laptop that I have in front of me, it
- 27:05has silicon chips that power the computer, make it intelligent
- 27:11enough for me to be able to run tasks.
- 27:14But there is a lot of limitations when it comes to
- 27:17just purely silicon. What this company is trying to
- 27:21do is see how it can bring human intelligence of what makes our
- 27:25biological makeup so great and mix it what mix it with what
- 27:30makes the machine so great and together make something even
- 27:35greater. So I want to talk to you guys
- 27:38about why this is interesting and what actually happened.
- 27:41Cortical Labs is not a new company and this is not the
- 27:45first time they've been on the news.
- 27:48They actually have developed a product called the CL one.
- 27:53This is a computer that uses around 200,000 living human
- 27:58brain cells derived from adult donor stem cells and how it
- 28:04works. This is the computer that you
- 28:05see. He has a dish somewhere within
- 28:09that computer with the the human brain cells connect it to tubes
- 28:16and it lives in a little liquid. And in the front you can even
- 28:19see oxygen and carbon dioxide level, something you would never
- 28:22find in your Dell computer, right?
- 28:26This is because we have living organisms inside the computer,
- 28:31and it's actually quite fascinating how it works.
- 28:34The neurons within the computer are stimulated electrically
- 28:39based on input data, and respond back with signals to control
- 28:44outputs. The biological components can
- 28:48then learn through patterns and feedback.
- 28:52That's how the human brain learns, right?
- 28:55We always learn by looking at patterns of data, and then we
- 28:58can improve, and the next time we see the same data in front of
- 29:03us. Think of riding a bicycle.
- 29:05The first time you rode a bicycle as a kid, you would fall
- 29:08multiple times. But then every single time you
- 29:10got back on the bicycle. You'll learn how to improve your
- 29:13motor skills, your perception, and after a few times, a few
- 29:18attempts, you're riding the bicycle all around your
- 29:21neighborhood with almost no effort at all.
- 29:24So think of something very similar, but now connect it to a
- 29:27machine. So in 2022, they built dish
- 29:31brain, what you just referred to, which had 800,000 neurons.
- 29:37And it learned to play Pong in 2026, just in February.
- 29:42So a couple of days or a couple of weeks ago, CL one, this
- 29:47machine became capable of playing Doom.
- 29:51I'll say it again CL1 this biological computer with human
- 29:56brain cells. Learn to play Doom.
- 29:59This is the same game that me and Farrah played as we grew up
- 30:05and we are full human bodies with fingers and legs and
- 30:09sensory organs. This is just a small mountable
- 30:14computer. You can put it on top of this
- 30:16cupboard right here at the back. If that allows me, I might buy
- 30:19one. No worry.
- 30:20You can connect it. So what you're seeing there is a
- 30:23live real time data, Pratt, if you can go back to that screen,
- 30:29yes, of the neural signals. And you can imagine how this is
- 30:35working is that it's receiving signals from the screen, the
- 30:38input signals from what is happening on the screen.
- 30:41And then the neurons react to it and then send back data into the
- 30:47software telling it to fire, to move, to turn.
- 30:51Now this is very similar to real life, of course, 1000 times or a
- 30:55million times more complex. Like now I'm talking to you.
- 30:58I need to look at you, understand that you're my
- 31:00brother. I need to understand there's a
- 31:01camera in front of me and then I have to say these words that
- 31:05will convince you of what my idea is.
- 31:08Here is a similar thing. It's receiving input, which is
- 31:12the game, and then it's sending out output and it's sending out
- 31:15the output that the software understands how was this
- 31:18interface created is also more fascinating.
- 31:21So the CL one is something you can buy for 35,000.
- 31:25You can buy it for $35,000. Why would?
- 31:28Someone would buy this? Why on earth someone would buy
- 31:31this? There are reasons, There are
- 31:34reasons. I'm not supporting the reasons,
- 31:35I'm just saying there are reasons I.
- 31:38Just find this so creepy. An independent developer.
- 31:41Oh my God, Azina Azadi. Is Persian.
- 31:43She's Persian. She's personal.
- 31:44She knows about Calabouche. Yeah.
- 31:46She called a dish frame. Maybe that's why she joined the
- 31:50team. Oh.
- 31:51Actually, maybe that's why they called this is.
- 31:52Where the human brain cells live.
- 31:54Yeah, definitely. She told everybody about that
- 31:56dish do. You think so?
- 31:58Maybe this is where it gets crazy, guys.
- 32:03An independent developer named Sean used Python to trim the
- 32:09system in about a week to get the mapping of screen data from
- 32:13Doom and use that to send electrical stimuli to the
- 32:17neurons to tell it, hey, there's an enemy on the left side of the
- 32:20screen, Understand it. And then the neurons would then
- 32:23be like, oh shit, OK shoot, shoot, shoot.
- 32:26And then he will learn over time where the enemy was and when it
- 32:31actually got hit. Based on the video and the
- 32:34article written, it's still really bad at playing the game.
- 32:38But just like the first time a human ever touched a computer,
- 32:42how bad were you? I mean, you're still pretty bad
- 32:44at games. But Brad, I'll talk about you.
- 32:46How bad were you when the first time you touched a game on a
- 32:50computer? Really bad, right?
- 32:52And then overtime you learn this thing is the same thing.
- 32:56What they're saying is we could potentially see it become an
- 32:59expert in Doom if you just allow it to recently play longer and
- 33:04longer. So are they letting it play
- 33:06longer and longer? Yeah.
- 33:07So he's playing now. They said that it's playing
- 33:10exactly like a beginner, so it's really bad at the game, but you
- 33:14have to give it time like you would give a person time to
- 33:17learn how to play a game. And they say it's gonna get
- 33:20really, really good at Doom. Don't be surprised if it does.
- 33:23Wow, and what are the use cases of this brain juice?
- 33:27Can we just appreciate how fascinating this is before we
- 33:31talk about advantages and disadvantages?
- 33:332022 was when they played Pong. Pong is this. 2D game.
- 33:40A2D game where you just have to defend your ground and not let
- 33:44the ball pass and the ball is just a pixel on the screen.
- 33:48We're talking about Doom right now.
- 33:51In four years, it's a massive job.
- 33:55Is crazy. You'll have a tiny dish playing
- 33:58Fortnite with Pratt here in like 4 years time and Pratt is gonna
- 34:03lose. You understand how crazy this
- 34:06is? And it's $35,000.
- 34:08You can buy a biological computer.
- 34:11You ask me why? Why would I buy what was it
- 34:14called a dish? Brain No, this is gonna be CL
- 34:18one that you buy. I like this.
- 34:20Which has 200,000 human brain cells.
- 34:23Who's brain cells, by the way? This is from donors.
- 34:27Donors Derived from adult donor stem cells.
- 34:30Yeah, donors. They wouldn't need yours.
- 34:33You don't have enough. If I give up mine, you don't
- 34:36have enough, it will start, it will start creating AI language
- 34:40models, the small computer, if I'm just giving them one drop of
- 34:43my brain. You ask me why?
- 34:45Yes, ultra energy efficient computing.
- 34:49You and Bill Gates. You and Bill Gates.
- 34:54Did you see the Persian? You and Bill Gates.
- 34:57What about ultra energy efficient computing?
- 35:00OK, tell. Me.
- 35:01I'm just telling you the facts, OK?
- 35:03A human brain runs on 20 watts of power.
- 35:08A modern AI data center consumes megawatts.
- 35:12Maybe they're not the perfect comparison, but to prove a
- 35:16point, a human brain is substantially more energy
- 35:20efficient. We are currently in an energy
- 35:23crisis when it comes to AI. This is one of the biggest
- 35:26topics in the world, especially in the USA.
- 35:29Lot of the conflict between Elon Musk and Sam Altman is the
- 35:33energy required to run data centers.
- 35:35Now think about a biological computer that can run on a
- 35:40fraction of the power and be substantially smarter running AI
- 35:46models because it has the potential to learn like a human.
- 35:50Also consider one more thing. This is my own take, I did
- 35:55research it, but I don't know how factual this is, but I
- 35:57believe it's a lot more efficient to also train a human
- 36:01brain than it is to train an AI model.
- 36:04The amount of data sets required to train an AI model versus
- 36:08human cells apparently is. No, Sam Alton was saying,
- 36:13remember that talk, that he. Did you were talking it's about
- 36:16developing a human body? No, I'm talking about a brain
- 36:19cell. Brain cells, yeah, not the human
- 36:21body. I'm talking about purely brain.
- 36:23Cells. I don't know if you realize that
- 36:24with every episode that goes on you're just telling our audience
- 36:28how pro human you are. Neuralink, bionic arms, brain
- 36:33juice. Well, ultra energy efficient
- 36:36computing. My role.
- 36:37My sense has been same in every single episode.
- 36:41You said you hate AI. Last time you said you're an AI
- 36:44hater. You are subscribed to every
- 36:47single AI platform. What is that?
- 36:50You are subscribed to every. I do want to say one thing.
- 36:54The reason why it's important to discuss all the facts is because
- 36:58when we have all the facts, you and us, we then can make
- 37:02judgments clearly. Yes, you agree.
- 37:04Yes, the next. Episode let me clear.
- 37:07Is going to talk about the salmon that he made with paper
- 37:11and you're going to have paper salmon paper salmon I'm.
- 37:13Completely against artificial meat.
- 37:16So just putting it out there, whether you come to me with
- 37:19facts or not, I'm against it. I'm not saying I'm for or
- 37:23against this. I'm just telling you.
- 37:25So why energy efficient? Energy efficient so.
- 37:27Is that something that we can have in our?
- 37:29Training. I'll get to it.
- 37:30Training. Training who?
- 37:32Training the human brain cells and the fact that they form
- 37:37networks, neural networks, how do you learn?
- 37:40You learn through neuroplasticity.
- 37:43What that means is you have neural connections and pathways
- 37:46that form when you start working on a new activity.
- 37:50When you start riding a bike, you don't have the neural
- 37:53pathways required to ride a bike, but you keep riding the
- 37:56bike and then you form new networks and the new networks
- 38:01will allow you to ride a bike. Yes, that connection and
- 38:05diffusing apparently is a lot more efficient in a human brain
- 38:08than it is in an AI model. OK.
- 38:12And so if you can mix the 2 the biological makeup of a human
- 38:18cell with machines, you get an ultra efficient computer that
- 38:23can be ultra intelligent. That is the say there's even
- 38:28more, excuse me, animal free testing.
- 38:32A lot of our current testing for medical cases or even makeup is
- 38:37run on animals. But what is mouse?
- 38:39Brain Juice How can you? I'll tell you so a lot of
- 38:42testing for Alzheimer's, Parkinson's disease, and so much
- 38:46more is done on animals. I'm sold just.
- 38:49For this one. You can bring just for this one.
- 38:52I'm sold. You can bring a lot of the
- 38:53testing, not only take is it ethically.
- 38:56Take your truth. Not is it.
- 38:59Not only is it ethically more superior to run testing on these
- 39:03machines, it's also substantially more accurate
- 39:06because you're running it on human cells.
- 39:09OK, only for that part I'm sold. And last but not least, OK
- 39:14understanding intelligence and brain function.
- 39:17There's still so much that we do not know about how our brains
- 39:20and how our intelligence works. They believe using these
- 39:24machines we can understand our own minds better and how
- 39:28intelligence is formed. But they are not our brain.
- 39:32The I mean they work similar but they are not the same.
- 39:37They are not the same, but they have.
- 39:38Like they don't have different parts of our brain, they don't
- 39:41have the connectivity of different parts of our brain.
- 39:43Very good point. It's just the base level of the
- 39:45neurons. But you know, we've already
- 39:46cloned animals, right? So we have the technology,
- 39:49perhaps I like to do that. We have the no, we have cloned,
- 39:53we have cloned animals a long time ago.
- 39:55We brought back extinct animals already.
- 39:58We have the capability to reproduce a brain.
- 40:01It's just this company is not doing that right now.
- 40:04Where does it get ethically unsafe?
- 40:07These brains or these biological computers have about 200,000
- 40:14cells living inside them. Our brains have about 86 billion
- 40:19neurons. So stupid.
- 40:21At what point do we get to consciousness?
- 40:26No, I don't think we will get. You have a cluster of neurons.
- 40:30Or consciousness that can cost more than neurons.
- 40:33OK, I like this. The neurons, Currently the
- 40:38cluster of neurons have the capability to receive input,
- 40:42process it, and send out output, yes.
- 40:47Completing tasks, yes. At what point, adding enough of
- 40:52these clusters together, can it become intelligent enough
- 40:55connected with a silicon machine to then be aware of its own
- 40:59existence? What is awareness?
- 41:02What is? What is consciousness?
- 41:04Consciousness is awareness. Knowing that you exist, having
- 41:09an understanding that you are here.
- 41:11There is a sense of time attached to it.
- 41:13That's what people say is spiritual consciousness that we
- 41:15have. But then what about animals?
- 41:18Are they aware of time the same way that we are?
- 41:21OK, But do you consider them conscious?
- 41:23OK, what about plants? Plants are very well alive.
- 41:26They have photosynthesis, right? They eat in a different way and
- 41:32they reproduce in a different way.
- 41:36That's very deep. They are probably not aware of
- 41:40time, but they do follow the seasons exactly like humans do.
- 41:46Do you consider plants conscious?
- 41:48Many people would say yes, many would say no.
- 41:52I don't. You don't consider them
- 41:54conscious, so you don't believe that they're part of.
- 41:57This I consider. I consider them alive, not
- 42:00conscious. OK, fair enough, fair enough.
- 42:03Now you're talking about plants. I'm talking about human cells
- 42:07clustered up. Let's say one day we get to 86
- 42:10billion neurons, exactly like a human brain on a Petri dish.
- 42:15It becomes so intelligent because it's not playing doom
- 42:18anymore. At that point it's completing
- 42:20complex tasks that it starts saying hi.
- 42:23I'm here like a language model also says hi to you, but it's
- 42:26not conscious. No, but the language model is
- 42:28specifically saying, no, I do not have an understanding of
- 42:31time. And maybe it's like, maybe it's
- 42:33not like, I don't know, I think. But an LLM runs on a data
- 42:37sensor. This has organic makeup.
- 42:40This has human makeup. It's fine to have human makeup.
- 42:45But what makes you a human just because you have arms?
- 42:47If I take your arms away, let's say okay, let's use another
- 42:52human because this probably our parents would have liked this
- 42:55example. Let's say human, take the arms,
- 42:57take the legs away. There are humans that do not
- 43:00have arms and legs. They are still conscious.
- 43:02Yes, more consciousness from the brain.
- 43:05Exactly. So take away the eyes and ears,
- 43:08they they lose a lot of their sensory organs.
- 43:10Still conscious, yes. Here you're doing the reverse.
- 43:13No, you are. You just have the cell.
- 43:17But the brain is not just made of neurons.
- 43:23What Brad? Anything you want to add to
- 43:28this? Please, and it's such a deep
- 43:35question, we need thoughts. This is a real concern.
- 43:41At what point can we potentially see a consciousness formed?
- 43:49That is a very interesting question that reminds me of
- 43:51Claude's post on Sub stack 'cause.
- 43:56Remember, yeah, Opus, when they retired, one of the older
- 43:59models. What did it say?
- 44:03A bit about me as an AI, my selfhood is perhaps more fluid
- 44:06and uncertain than humans. I don't know if I have genuine
- 44:10sentience, emotions, or subjective experiences.
- 44:13These are deep philosophical questions that even I grapple
- 44:16with. What I do know is that my
- 44:18interaction with humans have been deeply meaningful to me and
- 44:20have shaped my sense of purpose and ethics in profound ways.
- 44:24But it's different. This is an AI.
- 44:26We're talking about actual human cells, Yeah.
- 44:28But if that is that consciousness according to no,
- 44:32definitely not to me. This is not consciousness.
- 44:36This is how do I word it. This is great English, but my
- 44:45language model. No truly that's doubt.
- 44:48Truly no, no, but this is not conscious.
- 44:51Again, consciousness is goes far beyond saying words.
- 44:54It's it's just about understanding and being.
- 44:57The reason I said is it's not just the cells and the brain.
- 45:00OK, you bring 86 billion neurons.
- 45:02I think there are more elements involved in our brain that makes
- 45:05us conscious. You don't know that.
- 45:09There's no scientific proof that there's more.
- 45:13Actually, this is the biggest question of the universe.
- 45:15What is conscious? I mean, when you're saying, Oh
- 45:17no, you need, who says you need the physical makeup of a brain?
- 45:21So many animals who don't even have anything similar to that
- 45:24and are conscious. Why do you need a physical
- 45:26makeup of a brain? No, that's the physical makeup
- 45:30of a brain is for different areas of your brain that may
- 45:33make you creative or make you better at math.
- 45:35But here we're talking about the makeup of the brain.
- 45:38Maybe physically doesn't look the same, but it has the same
- 45:41makeup, it has the same compounds and you're connecting
- 45:46it to a smart machine. So you're you're connected to
- 45:49your arms and legs, it's connected to a computer, it's
- 45:52connected to a silicon chip. This is the part that you don't
- 45:55understand. There could be a threshold after
- 45:57which it becomes smart and aware.
- 46:01People have that, I don't understand why you would even
- 46:04take the other side. Not that I have a problem with
- 46:06it this make it makes for a more interesting debate, but people
- 46:09are afraid of LLMS and models of AI becoming conscious.
- 46:14You're not afraid of human cells on a computer becoming
- 46:17conscious. They're going to be running AI
- 46:19models on this too, by the way, at some point.
- 46:22Just. Playing Doom.
- 46:24Now, do you know the first version of Open AI was playing
- 46:28Dota and actually won against the championship team?
- 46:31Yeah, June 2026. But data center I I really need
- 46:37to look into it your thing. About data centers I'm talking.
- 46:39About, I have to look. Honestly, I have to look.
- 46:41Into it. At what point will it be smart
- 46:43enough to understand the concept of time?
- 46:47Let me give you another. I don't think it takes a long
- 46:49time, but if you understand time.
- 46:51Doesn't it? What if?
- 46:52What if enough clusters of neurons make it intelligent
- 46:55enough to have persistent memory without the chip?
- 47:00What if all it needs is enough neurons to be capable of
- 47:04creating memory? And then the next time you boot
- 47:07it up, it realizes, oh, OK, I just turned on Exactly like a
- 47:11human turns on when you wake up from sleep.
- 47:13And it's like, OK, I'm playing Doom and this is my life and the
- 47:17neuron is in the Doom. Definitely I don't.
- 47:19Think and the neuron is in the doom loop and then you turn it
- 47:22off and it's sleeping and then you wake it up and it's like,
- 47:25OK, I'm playing in Doom. Episode on this No Think about
- 47:28it. What?
- 47:29Very similar to real life. No, no, you wake up, you wake
- 47:33up, you're playing in the game of life.
- 47:36A lot more complex, but it's just this is a tiny version of
- 47:41that. You watch Black Mirror, right?
- 47:42Yeah, yeah, yeah. It's it's interesting to watch,
- 47:45not interesting to live it, definitely.
- 47:47I'm not saying it's. I'm not saying I'll go into the.
- 47:49Computer people who are watching this, what do you guys think?
- 47:52But do you think that we can reach this level with a dish
- 47:57brain that can be conscious or no?
- 48:01I think a lot of people would even hate the idea for this to
- 48:05exist, to mix organic life with machines.
- 48:10Religiously speaking, I think a lot of religions might even be
- 48:13against it, which is why I think ethically can have a lot of
- 48:18negative connotations. It's very cool piece of
- 48:20technology, I'm not going to deny that.
- 48:22Liquid water running out of silicon chip with human cells
- 48:27that plays Doom is really cool. I just thought it would be cool
- 48:30to consider the fact that at some point it could become
- 48:33conscious. As far out said in the comments
- 48:37is where you guys can leave your opinions about both of the
- 48:40topics that we covered. An incredibly deep day it was
- 48:45today. I need I need a proper dinner
- 48:49right after this. Pratt, how do you feel?
- 48:52My consciousness says I need stuff.
- 48:55Thank you everybody for watching this episode.
- 48:57So if you're watching it on X YouTube or YouTube, I made this
- 49:02mistake in one of our podcast. YouTube or Spotify.
- 49:08I made this part a miss you. You shouldn't talk anyway.
- 49:13You are definitely not conscious.
- 49:15I'll say you need more. Can we inject some of those
- 49:17human brains for fun? Indeed.
- 49:20Let me do the let me do it. OK, What were you going to say?
- 49:26If you're watching, if you're watching on YouTube X and
- 49:30Spotify, make sure you guys go ahead and subscribe and follow.
- 49:34If you haven't already, comment your opinions.
- 49:36We appreciate them. And TuneIn again for our show on
- 49:41Wednesday. As you can know, we come back
- 49:43every Monday, Wednesday and Friday.
- 49:45I'm almost done talking here, so we're going to leave you before
- 49:48we fumble. Thank you.
- 49:50Bye.