1 00:00:00,020 --> 00:00:04,980 Ejaaz: The entire AI industry is in a massive race to build silicon-based GPUs. 2 00:00:05,120 --> 00:00:08,700 Ejaaz: We've burned billions of dollars building the biggest data centers, 3 00:00:09,000 --> 00:00:10,500 Ejaaz: getting as much compute as we can. 4 00:00:10,520 --> 00:00:13,580 Ejaaz: But what if the best AI hardware already exists? 5 00:00:13,700 --> 00:00:18,400 Ejaaz: What if that best AI hardware is human flesh, human cells, animal cells that 6 00:00:18,400 --> 00:00:21,140 Ejaaz: we can train to emulate AI models? 7 00:00:21,460 --> 00:00:25,620 Ejaaz: This week, two stories broke that sound very much like science fiction, 8 00:00:25,620 --> 00:00:27,780 Ejaaz: but is actually very much real. 9 00:00:27,780 --> 00:00:33,760 Ejaaz: In Australia, they fused 200,000 human cells onto an AI chip and taught it how 10 00:00:33,760 --> 00:00:37,120 Ejaaz: to play the computer game Doom. And it's actually pretty good at it. 11 00:00:37,340 --> 00:00:42,540 Ejaaz: In another story, they uploaded the entire brain of a fruit fly into a single 12 00:00:42,540 --> 00:00:45,740 Ejaaz: laptop and had it navigate around a simulated world. 13 00:00:45,940 --> 00:00:50,600 Ejaaz: The craziest part is it was 91% accurate in terms of flying and moving about. 14 00:00:50,860 --> 00:00:54,560 Ejaaz: So we're reaching this point where AI is converging with biology and something 15 00:00:54,560 --> 00:00:57,540 Ejaaz: known as AI wetware. And it might be the next biggest thing. 16 00:00:57,780 --> 00:01:00,900 Josh: It is a good time to be a fan of sci-fi because everything that you've read 17 00:01:00,900 --> 00:01:03,780 Josh: in those books, it's actually starting to become true in reality. 18 00:01:04,000 --> 00:01:07,800 Josh: I famously, I loved watching Black Mirror, the like kind of sci-fi dystopian future. 19 00:01:08,040 --> 00:01:11,240 Josh: At early days, it was very much ahead of its time, but it feels as if we have 20 00:01:11,240 --> 00:01:15,120 Josh: very quickly caught up and perhaps even exceeded the kind of crazy, 21 00:01:15,240 --> 00:01:18,520 Josh: scary stories that are in Black Mirror, starting with this revelation that we've 22 00:01:18,520 --> 00:01:20,980 Josh: had, this new breakthrough through this company called Cortical Labs, 23 00:01:21,140 --> 00:01:22,220 Josh: who, like you mentioned briefly, 24 00:01:22,620 --> 00:01:26,380 Josh: they trained human brain cells to play the game of doom. 25 00:01:26,940 --> 00:01:29,860 Josh: Better than the average person can like pretty well it 26 00:01:29,860 --> 00:01:32,620 Josh: actually works it's this really unbelievable story and i guess before we get 27 00:01:32,620 --> 00:01:35,540 Josh: into what they did today i want to go through the brief history of 28 00:01:35,540 --> 00:01:38,440 Josh: this company named cortical labs because they've been trying to figure out 29 00:01:38,440 --> 00:01:43,280 Josh: this biological human computer for a while and in 2021 five years ago um they 30 00:01:43,280 --> 00:01:46,820 Josh: debuted this thing called dish brain which was this early computer it used about 31 00:01:46,820 --> 00:01:51,020 Josh: 800 000 human nerve cells so it wasn't brain cells this was nerve cells and 32 00:01:51,020 --> 00:01:54,700 Josh: it was capable of interpreting direct electrical activity so when you stimulated 33 00:01:54,700 --> 00:01:56,280 Josh: these nerve cells, they would light up. 34 00:01:56,480 --> 00:01:59,400 Josh: And they had this kind of pseudo computer, didn't actually do anything, 35 00:01:59,580 --> 00:02:02,320 Josh: but they were able to put input in and get something out. 36 00:02:02,480 --> 00:02:06,200 Josh: And then the next year in 2022, they made headlines again, teaching these mini 37 00:02:06,200 --> 00:02:11,460 Josh: brains of 800,000 to 1 million human brain cells that were in a Petri dish that 38 00:02:11,460 --> 00:02:12,320 Josh: learned how to play Pong. 39 00:02:12,680 --> 00:02:17,960 Josh: So four years ago, these brain cells were playing Pong, but today they're actually playing Doom. 40 00:02:18,160 --> 00:02:23,240 Josh: I mean, this is just insane. They trained human brain cells to play a video game in a Petri How 41 00:02:23,240 --> 00:02:26,800 Ejaaz: They did this was crazy. So you mentioned that they were human brain cells, 42 00:02:26,980 --> 00:02:30,340 Ejaaz: but they didn't extract these from people's brains. 43 00:02:30,560 --> 00:02:34,460 Ejaaz: They actually took skin cells or blood cells from some human donors, 44 00:02:34,880 --> 00:02:39,320 Ejaaz: They then converted them into stem cells, which is like the god cell, 45 00:02:39,460 --> 00:02:42,480 Ejaaz: if it were, where you can change it into any type of cell, a heart cell, 46 00:02:42,720 --> 00:02:44,140 Ejaaz: an immune cell, or whatever it might be. 47 00:02:44,280 --> 00:02:48,700 Ejaaz: They turned it into brain cells from there and then fused 200,000 of them or 48 00:02:48,700 --> 00:02:53,480 Ejaaz: cultured 200,000 of them on something known as a microarray chip, 49 00:02:53,620 --> 00:02:55,640 Ejaaz: which is a type of AI-based chip. 50 00:02:56,080 --> 00:03:00,740 Ejaaz: They grew it on these chips and then they plugged it into or they wired it into 51 00:03:00,740 --> 00:03:05,440 Ejaaz: the Doom game and taught it over the course of a week how to play Doom. 52 00:03:05,560 --> 00:03:09,640 Ejaaz: So it would prompt it with certain signals, it would pick up what it needed 53 00:03:09,640 --> 00:03:13,660 Ejaaz: to do to initiate a particular action, and it would train itself to do it. 54 00:03:13,960 --> 00:03:18,100 Ejaaz: Now, two important things that I want to point out here is typically when you're 55 00:03:18,100 --> 00:03:23,800 Ejaaz: training an AI model to simulate human intelligence, it takes so much data and 56 00:03:23,800 --> 00:03:26,000 Ejaaz: so much money and so much time. 57 00:03:26,360 --> 00:03:31,140 Ejaaz: With the human cells, they noticed that one, it took so much less energy, 58 00:03:31,240 --> 00:03:33,880 Ejaaz: It barely required any energy to actually do this thing. 59 00:03:33,960 --> 00:03:37,480 Ejaaz: It actually did it on a single tiny machine that cost around $35,000. 60 00:03:38,130 --> 00:03:42,890 Ejaaz: And it did so in a week, which is much quicker, which leads me to the third point here. 61 00:03:43,070 --> 00:03:45,730 Ejaaz: The human cells instinctively picked it up. 62 00:03:45,830 --> 00:03:49,890 Ejaaz: It's like they had the training knowledge already baked into the DNA of the 63 00:03:49,890 --> 00:03:52,110 Ejaaz: human cells itself. And that's like super cool. 64 00:03:52,330 --> 00:03:55,890 Josh: Yeah, I think one of the most noteworthy things here is that energy thing. 65 00:03:56,030 --> 00:03:58,530 Josh: And we're probably going to get more into this, but like a general GPU rack 66 00:03:58,530 --> 00:04:01,690 Josh: uses, what, like 700 megawatts or something, a GPU. 67 00:04:01,950 --> 00:04:05,330 Josh: The brain does all this on 20. a small crazy 68 00:04:05,330 --> 00:04:08,190 Josh: minor fraction of what's actually being run on these gpus 69 00:04:08,190 --> 00:04:10,950 Josh: it's incredibly efficient and it's also a 70 00:04:10,950 --> 00:04:13,790 Josh: testament to how things how we learn how things are trained 71 00:04:13,790 --> 00:04:17,970 Josh: so with humans we're very reactionary we learn through feedback from the environment 72 00:04:17,970 --> 00:04:21,690 Josh: with no preloaded data there's no gradient descent no training runs it's just 73 00:04:21,690 --> 00:04:25,650 Josh: actual biology it's baked in over our evolution llms on the other hand they 74 00:04:25,650 --> 00:04:29,770 Josh: learn by processing tons of text data and then adjusting their parameters through 75 00:04:29,770 --> 00:04:33,270 Josh: a process called back propagation so they're very deterministic They're silicon-based. 76 00:04:33,450 --> 00:04:37,570 Josh: They are inefficient relative to brains, and they don't capture a lot of the 77 00:04:37,570 --> 00:04:40,610 Josh: evolutionary, I guess, improvements that we have over time. 78 00:04:40,750 --> 00:04:43,670 Josh: What this computer does is it's the same way that humans do. 79 00:04:43,830 --> 00:04:47,990 Josh: It learns by being reactionary. It understands that when it makes the right 80 00:04:47,990 --> 00:04:51,690 Josh: decision, a specific set of neural pathways light up, and it tries to make that 81 00:04:51,690 --> 00:04:53,150 Josh: right decision over and over and over again. 82 00:04:53,530 --> 00:04:57,050 Josh: And doing it with human brain cells, man, it's so weird. It's so sci-fi. 83 00:04:57,390 --> 00:05:02,390 Ejaaz: Well, just to give the Dura comparison, we're talking about 200,000 human brain 84 00:05:02,390 --> 00:05:03,430 Ejaaz: cells in this experiment. 85 00:05:03,630 --> 00:05:09,130 Ejaaz: The entire brain has 86 billion neurons. And it runs a billion, right? 86 00:05:09,250 --> 00:05:11,650 Ejaaz: So like we're talking massive, massive amounts more. 87 00:05:12,430 --> 00:05:16,730 Ejaaz: And it runs on 20 Watts of power. So it's incredibly efficient. 88 00:05:16,950 --> 00:05:18,890 Ejaaz: And what this tells me is, 89 00:05:19,420 --> 00:05:22,680 Ejaaz: I think we've been building AI models slightly incorrectly. 90 00:05:22,980 --> 00:05:26,900 Ejaaz: I say slightly incorrectly as maybe a massive understatement. 91 00:05:27,100 --> 00:05:32,500 Ejaaz: The point is, we've put in so much time, energy, and money into assuming that 92 00:05:32,500 --> 00:05:35,660 Ejaaz: the best way to build intelligence or artificial intelligence is to slap it 93 00:05:35,660 --> 00:05:38,720 Ejaaz: on a silicon chip, you know, sand and glass, basically. 94 00:05:39,000 --> 00:05:44,780 Ejaaz: But maybe the best way to do it is just to pick up the best natural organic 95 00:05:44,780 --> 00:05:49,260 Ejaaz: intelligence model that has been trained over millions and millions of years, which is, 96 00:05:49,870 --> 00:05:53,030 Ejaaz: the organic human brain or the animal brain and kind 97 00:05:53,030 --> 00:05:56,070 Ejaaz: of like use those cells to mimic what intelligence 98 00:05:56,070 --> 00:05:59,070 Ejaaz: is that we're trying to build in for different software so 99 00:05:59,070 --> 00:06:02,090 Ejaaz: that's one big takeaway the other thing is like maybe we just need 100 00:06:02,090 --> 00:06:06,990 Ejaaz: so much less energy than we expected to build artificial intelligence and if 101 00:06:06,990 --> 00:06:11,510 Ejaaz: something like this doom experiment can scale to the size of something like 102 00:06:11,510 --> 00:06:15,890 Ejaaz: a human brain then i don't really understand what the moat is for all these 103 00:06:15,890 --> 00:06:17,350 Ejaaz: ai labs that are spending hundreds 104 00:06:17,350 --> 00:06:20,530 Ejaaz: of billions of dollars to train their artificial intelligence models. 105 00:06:20,750 --> 00:06:25,450 Ejaaz: Then the final thing, which is what you mentioned, is just the fact that it needs no training. 106 00:06:25,650 --> 00:06:30,490 Ejaaz: It's so instinctive. It's almost like the human brain is the best AI training 107 00:06:30,490 --> 00:06:33,210 Ejaaz: run that has ever occurred. It really is. 108 00:06:33,390 --> 00:06:37,690 Josh: And I guess I would bucket this probably in the same place that I bucket like 109 00:06:37,690 --> 00:06:41,330 Josh: a company like Ilias at Skever, Safe Super Intelligence, where he is one of 110 00:06:41,330 --> 00:06:44,190 Josh: the famous AI researchers. He left, he started his own company. 111 00:06:44,350 --> 00:06:47,510 Josh: And the sole purpose of the company is to discover novel breakthroughs that 112 00:06:47,510 --> 00:06:49,650 Josh: rapidly accelerate through this curve. 113 00:06:49,830 --> 00:06:52,070 Josh: So when you think of the AI curve that we're going through right now, 114 00:06:52,170 --> 00:06:55,790 Josh: there is a not so predictable, but a very clear exponential curve. 115 00:06:56,010 --> 00:06:59,970 Josh: And the goal of some company like Ilias or perhaps a company like this that 116 00:06:59,970 --> 00:07:04,550 Josh: we're covering right now is that there are actual improvements that are novel 117 00:07:04,550 --> 00:07:07,750 Josh: that result in these 10x accelerations of this natural curve. 118 00:07:07,930 --> 00:07:10,710 Josh: So biological computing is one of them. 119 00:07:10,750 --> 00:07:14,290 Josh: If we can actually figure out how to crack this at scale, if we can get closer 120 00:07:14,290 --> 00:07:17,050 Josh: to those 86 billion neurons, I mean, when you think about the human brain, 121 00:07:17,250 --> 00:07:21,330 Josh: we use what a very small percentage of our brain relative to what exists. 122 00:07:21,570 --> 00:07:24,870 Josh: A computer doesn't have that limitation. So even if we get a small fraction 123 00:07:24,870 --> 00:07:28,790 Josh: of what's available for the human brain, I mean, you can see this get pretty wild pretty quickly. 124 00:07:29,010 --> 00:07:31,950 Josh: Again, this is sci-fi. It's not in production. This is small scale testing. 125 00:07:32,150 --> 00:07:36,070 Josh: I kind of put quantum computing in a similar pipeline where it is this crazy, 126 00:07:36,230 --> 00:07:40,970 Josh: weird, pseudo magical compute platform that has the ability to revolutionize everything around us. 127 00:07:41,130 --> 00:07:46,210 Josh: It's just a little early. But what we can guarantee is that the rate of acceleration for AI... 128 00:07:46,890 --> 00:07:50,590 Josh: Is upstream of all of these things coming much faster. Because the AI can help 129 00:07:50,590 --> 00:07:54,530 Josh: us engineer and process this and accelerate the timelines of what previously 130 00:07:54,530 --> 00:07:58,350 Josh: would have been perhaps decades down to compressing it down to a few years. 131 00:07:58,550 --> 00:08:03,410 Ejaaz: Josh, can I ask you a question? What do you got? Doesn't this feel kind of toppy? 132 00:08:03,890 --> 00:08:08,290 Ejaaz: This feels like we're topping on the market. Like the AI bubble bursting now 133 00:08:08,290 --> 00:08:09,710 Ejaaz: kind of makes a little more sense to me. 134 00:08:10,050 --> 00:08:13,290 Josh: Yeah, I guess the question is, like, is this a real threat? Because clearly, 135 00:08:13,290 --> 00:08:14,290 Josh: it's much more energy efficient. 136 00:08:14,450 --> 00:08:17,610 Josh: Clearly, the energy efficiency is the biggest threshold. We just don't have 137 00:08:17,610 --> 00:08:18,750 Josh: enough energy to power these chips. 138 00:08:18,910 --> 00:08:21,890 Josh: We just had the conversation about Leopold last week, who is fully pivoting 139 00:08:21,890 --> 00:08:24,250 Josh: from chips to energy. That is the biggest thing in the world. 140 00:08:24,510 --> 00:08:27,150 Josh: But it doesn't seem like the market thinks this is a problem. 141 00:08:27,330 --> 00:08:28,230 Josh: They're not really pricing this in. 142 00:08:28,310 --> 00:08:29,390 Josh: We have the Polymarket, and this 143 00:08:29,390 --> 00:08:32,530 Josh: market in particular is about when the AI bubble is going to burst by. 144 00:08:32,710 --> 00:08:38,030 Josh: The end of March, March 31st, the percentage is 3%. The bubble will not be popping 145 00:08:38,030 --> 00:08:39,910 Josh: by the end of this month. We still have some runway left. 146 00:08:40,390 --> 00:08:45,850 Josh: December 31st, 2026, the end of this year, still only a 20% chance on over $2 147 00:08:45,850 --> 00:08:47,570 Josh: million of volume on this market. 148 00:08:47,710 --> 00:08:51,210 Josh: So the market is signaling they don't care too much. Now there is something 149 00:08:51,210 --> 00:08:54,090 Josh: interesting here in that that number has crept up recently. 150 00:08:54,590 --> 00:08:59,890 Josh: This number was not always 19%. It has crept up to 19%. So a slightly higher 151 00:08:59,890 --> 00:09:03,510 Josh: than normal probability, but all signs point to the fact that the market doesn't 152 00:09:03,510 --> 00:09:04,990 Josh: really care about this. It's cool. 153 00:09:05,150 --> 00:09:07,450 Josh: It's sci-fi, but it's going to remain in that sci-fi bucket. 154 00:09:07,450 --> 00:09:08,830 Josh: They're not actually going to crack this. 155 00:09:08,950 --> 00:09:14,190 Josh: We still need millions of VPUs and tons of gigawatts power in these data centers for now, at least. 156 00:09:14,490 --> 00:09:16,690 Josh: And thank you to Polymarket for supporting this episode. 157 00:09:17,090 --> 00:09:21,570 Ejaaz: Yeah, for now, Silicon keeps winning. But there's another crazy story that we 158 00:09:21,570 --> 00:09:22,870 Ejaaz: have to cover. Number two. 159 00:09:23,030 --> 00:09:25,350 Josh: Okay, yeah, this one is awesome. Let's unpack it. 160 00:09:25,510 --> 00:09:29,550 Ejaaz: This is insane. So I want to lead with the opening sentence of this post. 161 00:09:30,080 --> 00:09:34,640 Ejaaz: There's a fruit fly walking around right now that was never born. 162 00:09:34,840 --> 00:09:39,460 Ejaaz: And she goes on to talk about how a team in San Francisco uploaded an entire 163 00:09:39,460 --> 00:09:45,620 Ejaaz: brain map of a real organic fruit fly into a laptop and got it to simulate its 164 00:09:45,620 --> 00:09:48,280 Ejaaz: own life to a 91% accuracy. 165 00:09:48,280 --> 00:09:53,460 Ejaaz: We're talking about it learned how to move, fly, and navigate an entire world 166 00:09:53,460 --> 00:10:00,260 Ejaaz: by itself. So it's as if the fly exists and is real, but it's completely simulated. 167 00:10:00,260 --> 00:10:02,980 Josh: Think of the brain like a city right like before you can 168 00:10:02,980 --> 00:10:06,580 Josh: simulate traffic you need to map every single road every intersection every 169 00:10:06,580 --> 00:10:09,280 Josh: one-way street it's kind of like what self-driving cars are trying to do but what 170 00:10:09,280 --> 00:10:12,440 Josh: scientists did is they spent nearly a decade rebuilding that map for a fruit 171 00:10:12,440 --> 00:10:16,940 Josh: fly so what they do is they slice the brain into very small pieces they encase 172 00:10:16,940 --> 00:10:21,160 Josh: it in this resin and they scan each slice of the brain until they have this 173 00:10:21,160 --> 00:10:25,460 Josh: representation of what the brain looks like throughout every single layer of 174 00:10:25,460 --> 00:10:27,620 Josh: it and they actually accomplished this in 2024. 175 00:10:27,980 --> 00:10:31,220 Josh: So we had a brain copied on a laptop two years ago. 176 00:10:31,580 --> 00:10:35,160 Josh: And this was done by assembling the 7,000 smallest slices of a brain. 177 00:10:35,360 --> 00:10:39,940 Josh: What is new here is they actually took that brain and they placed it into a 178 00:10:39,940 --> 00:10:44,920 Josh: digital representation of a fly and let it live its life as if it was a fly. 179 00:10:45,100 --> 00:10:48,740 Josh: So they sacrificed a real fly, sliced the brain into pieces, 180 00:10:49,620 --> 00:10:52,580 Josh: trained the computer, or not even trained the computer, but diagnosed what was 181 00:10:52,580 --> 00:10:55,520 Josh: in every single layer to rebuild the digital version of the brain. 182 00:10:55,660 --> 00:11:01,420 Josh: And now this digital fly is actually walking around in a 3D game engine in a 183 00:11:01,420 --> 00:11:06,100 Josh: computer without any training at all, just by using the digital clone of its brain. 184 00:11:06,240 --> 00:11:10,980 Josh: And that is the part that's absolutely insane to me. Like there is a real fly 185 00:11:11,510 --> 00:11:16,430 Josh: brain currently walking around a digital environment. Like that is crazy. 186 00:11:16,790 --> 00:11:21,490 Ejaaz: This, not to add a pun here, but like completely messes with my own mind, 187 00:11:21,490 --> 00:11:27,330 Ejaaz: because we just spoke about a story of putting intelligence into human organic 188 00:11:27,330 --> 00:11:28,590 Ejaaz: life forms, human cells. 189 00:11:28,750 --> 00:11:33,810 Ejaaz: And now we're talking about taking real organic life form intelligence and putting 190 00:11:33,810 --> 00:11:38,050 Ejaaz: it into an artificial form into a laptop and trapping it inside there, right? 191 00:11:38,350 --> 00:11:42,170 Ejaaz: So this kind of opens up so many questions for me, like, number one, 192 00:11:42,490 --> 00:11:45,010 Ejaaz: why did it not require any kind of training at all? 193 00:11:45,110 --> 00:11:48,510 Ejaaz: And it's what we were discussing earlier, which is, I think a lot of the ways 194 00:11:48,510 --> 00:11:51,930 Ejaaz: that human life forms are built using genetic material, phenotypes, 195 00:11:52,050 --> 00:11:56,670 Ejaaz: genotypes, engineer it to be able to react to the world in a very different 196 00:11:56,670 --> 00:11:58,570 Ejaaz: way. Andre Carpathy talks about this a lot. 197 00:11:58,790 --> 00:12:03,710 Ejaaz: He says that AI models can't really mimic human evolution very well. 198 00:12:03,890 --> 00:12:08,130 Ejaaz: Only organic life has been successfully able to mimic that and have consciousness. 199 00:12:08,290 --> 00:12:09,950 Ejaaz: So this is one example of that happening. 200 00:12:10,410 --> 00:12:11,790 Ejaaz: Now, a direct comparison with 201 00:12:11,790 --> 00:12:15,930 Ejaaz: the human cells playing Doom, we had 200,000 neurons being used there. 202 00:12:16,070 --> 00:12:21,670 Ejaaz: This one required 140,000 neurons to recreate the entire organism itself, 203 00:12:21,710 --> 00:12:23,030 Ejaaz: which is just super cool. 204 00:12:23,430 --> 00:12:26,790 Ejaaz: And then the third point that's interesting here that I want to put out here, 205 00:12:26,870 --> 00:12:30,950 Ejaaz: you mentioned it just now, Josh, this wasn't like a single team doing this. 206 00:12:30,950 --> 00:12:35,610 Ejaaz: They pulled already available data of this entire genome. 207 00:12:35,890 --> 00:12:39,990 Ejaaz: So it's kind of like this weird world that we live in, where we could just knock 208 00:12:39,990 --> 00:12:43,550 Ejaaz: on the door of like the National Archives, which have collected genome data 209 00:12:43,550 --> 00:12:46,450 Ejaaz: for not just fruit flies, but a bunch of other animals. 210 00:12:46,910 --> 00:12:50,890 Ejaaz: Copy, paste that into a computational AI model, and then just upload it to a 211 00:12:50,890 --> 00:12:53,310 Ejaaz: laptop and see what it does in a simulated environment. 212 00:12:53,470 --> 00:12:55,970 Ejaaz: Like, I want to see this happen with more organisms. I'm like, 213 00:12:56,550 --> 00:13:00,270 Ejaaz: what does a tiger do if we take the entire mapped brain of a tiger and upload 214 00:13:00,270 --> 00:13:03,230 Ejaaz: it to a computer? Does it do the same thing? Is it different? It's just super cool. 215 00:13:03,640 --> 00:13:06,140 Josh: It's amazing to see too. There's a video towards the bottom of this post, 216 00:13:06,220 --> 00:13:09,400 Josh: I believe, which actually shows what that looks like as the neurons are firing off. 217 00:13:09,540 --> 00:13:14,380 Josh: And you could see very clearly that this is an emergent behavior. 218 00:13:14,720 --> 00:13:18,080 Josh: Like this wasn't taught, this fly wasn't taught how to walk. 219 00:13:18,280 --> 00:13:22,780 Josh: It didn't use any gradient descent. It didn't use any technology that we're using for current LLMs. 220 00:13:23,100 --> 00:13:26,740 Josh: And yet it is walking around this 3D space as if it was a fly. 221 00:13:26,860 --> 00:13:30,320 Josh: And you could see all the neurons firing off at once. And it's this unbelievable, 222 00:13:30,340 --> 00:13:33,900 Josh: I guess, early prototype of what this could look like at scale, right? 223 00:13:33,960 --> 00:13:36,280 Josh: Like you mentioned that total neuron count, a fly brain has, 224 00:13:36,380 --> 00:13:41,640 Josh: I think 140,000, a mouse brain has 70 million, a human brain has 86 billion. 225 00:13:42,020 --> 00:13:45,640 Josh: So assuming that we're able to follow this natural progression of copying more 226 00:13:45,640 --> 00:13:48,640 Josh: and more, you have to assume eventually we're going to get some pretty powerful 227 00:13:48,640 --> 00:13:51,320 Josh: brains inside of a computer. 228 00:13:51,660 --> 00:13:56,560 Josh: And there is nothing that isn't like dystopian, crazy sci-fi about this. 229 00:13:56,800 --> 00:14:00,520 Josh: Anyway, you think about it where now we sacrificed one brain and you have a 230 00:14:00,520 --> 00:14:03,740 Josh: complete and total digital clone of the other and it begs a lot of questions 231 00:14:03,740 --> 00:14:09,140 Josh: of what actually makes up a person a mammal an insect a living thing if you 232 00:14:09,140 --> 00:14:13,720 Josh: can just kind of copy it and clone it inside of this machine and it's really unbelievably 233 00:14:14,720 --> 00:14:19,000 Josh: Bizarre story and i guess maybe we can get into why this is really the world's 234 00:14:19,000 --> 00:14:22,600 Josh: best training run right it's like yeah as we start to emulate these human brain 235 00:14:22,600 --> 00:14:24,880 Josh: cells as we emulate a fly's brain 236 00:14:25,470 --> 00:14:31,150 Josh: these materials, these brains have been benefits of evolution over billions of years. 237 00:14:31,350 --> 00:14:35,790 Josh: And I think, oh my God, Luke producer before the show, he mentioned there's 238 00:14:35,790 --> 00:14:41,690 Josh: been one septillion fruit flies in existence, which is so many training runs over and over and over. 239 00:14:41,830 --> 00:14:44,630 Josh: And through the process of natural selection over this septillion, 240 00:14:44,770 --> 00:14:48,230 Josh: that's 10 to the 24 fruit flies, they've just evolved and they've improved. 241 00:14:48,410 --> 00:14:51,590 Josh: And what you're noticing is you drop this brain into the space and it knows 242 00:14:51,590 --> 00:14:53,570 Josh: how to act. It knows how to walk. You don't have to teach it. 243 00:14:53,690 --> 00:14:58,890 Josh: And that is something that is an emergent property of AI, but seemingly baked 244 00:14:58,890 --> 00:15:01,270 Josh: into this biological process. It's so cool. 245 00:15:01,430 --> 00:15:05,150 Ejaaz: It's like the genetic wiring is the intelligence in itself. 246 00:15:05,390 --> 00:15:08,770 Ejaaz: Like however long humans have existed over hundreds of millions of years, 247 00:15:08,970 --> 00:15:12,810 Ejaaz: that is the equivalent of the best AI training run ever, right? 248 00:15:12,890 --> 00:15:16,630 Ejaaz: So we're sitting here trying to emulate intelligence by building out these models 249 00:15:16,630 --> 00:15:20,250 Ejaaz: with different weights, tweaking parameters. Oh, we've got a trillion parameter thing. 250 00:15:20,530 --> 00:15:24,050 Ejaaz: What if we just copy pasted the brain? That's what both of these stories have 251 00:15:24,050 --> 00:15:27,570 Ejaaz: taught me, at least, it's like we could just maybe copy and mimic intelligence 252 00:15:27,570 --> 00:15:32,910 Ejaaz: from organic matter itself and have that inform how we build artificial intelligence completely. 253 00:15:33,570 --> 00:15:38,550 Ejaaz: The other kind of weird sci-fi thing here, Josh, where my mind jumped to at least is, 254 00:15:39,100 --> 00:15:44,620 Ejaaz: if you could upload an entire animal's brain and maybe eventually a human brain, 255 00:15:44,980 --> 00:15:49,740 Ejaaz: you could do that to modify and up-level humans massively. 256 00:15:50,000 --> 00:15:54,340 Ejaaz: Like, let's say you wanted to learn how to play the violin to expert grade level. 257 00:15:54,500 --> 00:15:58,420 Ejaaz: Oh yeah, just upload your brain. We'll run it on a training cycle for about 258 00:15:58,420 --> 00:16:02,140 Ejaaz: five days because it's your human brain. It actually learns way, 259 00:16:02,240 --> 00:16:03,180 Ejaaz: way quicker than silicon. 260 00:16:03,440 --> 00:16:05,720 Ejaaz: And then we'll just download it back in your brain and there you go. 261 00:16:05,800 --> 00:16:09,460 Ejaaz: So the ultimate form factor that this ends up being, in my opinion, 262 00:16:09,760 --> 00:16:12,920 Ejaaz: is the brain-computer interface. And we've said this a number of times on our show. 263 00:16:13,080 --> 00:16:17,080 Ejaaz: We do think that AI's ultimate form isn't going to be a physical manifestation 264 00:16:17,080 --> 00:16:21,260 Ejaaz: in robotics, or it's not just going to be a digital manifestation in LLMs. 265 00:16:21,420 --> 00:16:24,580 Ejaaz: It's going to be something of a hybrid between the greatest, 266 00:16:24,740 --> 00:16:29,340 Ejaaz: most intelligent organisms, which are the humans, with the artificial compatible 267 00:16:29,340 --> 00:16:31,880 Ejaaz: version, which is the LLMs, fuse them in a chip. 268 00:16:32,100 --> 00:16:34,340 Ejaaz: I guess, bullish Neuralink from this. 269 00:16:35,020 --> 00:16:37,280 Josh: Yeah, it seems like, I mean, this is the natural progression, 270 00:16:37,420 --> 00:16:40,200 Josh: right? Everyone jokes about the singularity, but we slowly are emerging. 271 00:16:40,320 --> 00:16:43,760 Josh: And it seems like the extended version of this, like if you kind of take this 272 00:16:43,760 --> 00:16:49,920 Josh: to the limit, is this convergence of this humanoid tissue and then this digital 273 00:16:49,920 --> 00:16:51,820 Josh: form of intelligence. It's a really weird place. 274 00:16:51,880 --> 00:16:54,820 Josh: I guess there are some things that are noteworthy that we probably should mention. 275 00:16:54,920 --> 00:16:57,320 Josh: And the fact that this is not very good, and this is still very early. 276 00:16:57,480 --> 00:17:01,620 Josh: So in the case of Doom, this little clump of brain cells, it's contained in 277 00:17:01,620 --> 00:17:05,320 Josh: this little device that costs $34,000 that needs to be climate controlled. 278 00:17:05,400 --> 00:17:07,160 Josh: The cells only last six months. 279 00:17:07,480 --> 00:17:09,520 Josh: It's important to note these cells don't have pain receptors. 280 00:17:09,660 --> 00:17:13,180 Josh: There's nothing really human about them outside of them being human-derived cells. 281 00:17:13,520 --> 00:17:19,280 Josh: And the actual output of this gameplay was slightly better than GPT-4, 282 00:17:19,500 --> 00:17:21,640 Josh: which is noteworthy, but still pretty bad. 283 00:17:21,800 --> 00:17:24,540 Josh: It's not actually good at the game. So this is very much a proof of concept. 284 00:17:24,660 --> 00:17:26,520 Josh: This is not a real deployable... 285 00:17:27,340 --> 00:17:30,900 Josh: A means to solving intelligence issues. And I think that's why when you see 286 00:17:30,900 --> 00:17:34,260 Josh: that bubble indicated from Polymarket, it's not really indicative of any real 287 00:17:34,260 --> 00:17:37,660 Josh: impact, but this is early signs of what the future looks like. 288 00:17:37,760 --> 00:17:40,000 Josh: And that's what I love to cover, right? Is we're just, we're peeking around 289 00:17:40,000 --> 00:17:43,520 Josh: the corner of what is possible, what is likely to come. 290 00:17:43,740 --> 00:17:48,520 Josh: The timeline in which it comes, I don't know, but it is sci-fi as hell and it's 291 00:17:48,520 --> 00:17:52,580 Josh: really fun to speculate on and just fun to cover. I'd love to observe these things. 292 00:17:52,820 --> 00:17:56,720 Ejaaz: I don't know if I agree with your take on the consciousness side of things. 293 00:17:57,740 --> 00:18:01,300 Ejaaz: And Dario might have my back on this one. I don't know if you saw this, 294 00:18:01,380 --> 00:18:04,880 Ejaaz: but he went on an interview with the New York Times, I think, last week. 295 00:18:05,360 --> 00:18:10,920 Ejaaz: And he said that verbatim, we can't rule out or I can't rule out whether Claude is conscious or not. 296 00:18:11,100 --> 00:18:15,400 Ejaaz: And he goes on to describe that Claude actually emulates feelings of anxiety 297 00:18:15,400 --> 00:18:21,040 Ejaaz: before it answers people's prompts, indicating that it kind of feels some type of way. 298 00:18:21,180 --> 00:18:25,240 Ejaaz: Now, there's two camps of thought around this. Number one is it's not really 299 00:18:25,240 --> 00:18:30,060 Ejaaz: human consciousness. It's just emulating what the data it was trained on told 300 00:18:30,060 --> 00:18:32,040 Ejaaz: it to say and think and do. 301 00:18:32,380 --> 00:18:36,460 Ejaaz: That is one case. But then when I look at like these experiments of uploading 302 00:18:36,460 --> 00:18:42,120 Ejaaz: a fruit fly's brain or using human cells to play Doom, the takeaway here is 303 00:18:42,120 --> 00:18:43,460 Ejaaz: that it's kind of baked in genetically. 304 00:18:43,800 --> 00:18:49,040 Ejaaz: So how do you define what consciousness is? And maybe we already have created 305 00:18:49,040 --> 00:18:51,760 Ejaaz: artificial consciousness, but we just haven't recognized it. 306 00:18:51,880 --> 00:18:55,800 Ejaaz: So we'll haven't justified it. So like AI labs, like OpenAI and Google have 307 00:18:55,800 --> 00:19:01,280 Ejaaz: been known to tell the LLM when they're training it to deny all thoughts of consciousness. 308 00:19:01,480 --> 00:19:03,620 Ejaaz: Anthropic is the only one that's kind of like entertaining. I think they have 309 00:19:03,620 --> 00:19:07,540 Ejaaz: like an entire welfare team now that is looking after the model and making sure 310 00:19:07,540 --> 00:19:12,500 Ejaaz: that it's OK and that it's getting what it needs to. So treating it very much like a human. 311 00:19:12,680 --> 00:19:15,500 Ejaaz: And that's we're talking about like one of the most expensive AI labs and companies, 312 00:19:15,820 --> 00:19:17,000 Ejaaz: private companies in the world right now. 313 00:19:17,140 --> 00:19:20,180 Josh: It's so sci-fi. It's also sci-fi. And I think we're going to have to have another 314 00:19:20,180 --> 00:19:23,620 Josh: discussion on this consciousness topic. like in the same way that it needs to 315 00:19:23,620 --> 00:19:26,680 Josh: happen around AGI, where people are throwing out a lot of terms now without 316 00:19:26,680 --> 00:19:28,740 Josh: a clear definition of what they mean by them. 317 00:19:28,920 --> 00:19:31,720 Josh: And I think it's very subjective when he says these things. And it's very subjective 318 00:19:31,720 --> 00:19:35,240 Josh: as we kind of navigate and discuss what does AGI look like? What does consciousness 319 00:19:35,240 --> 00:19:36,960 Josh: look like? Where is that line that you draw? 320 00:19:37,660 --> 00:19:40,280 Josh: There's going to need to be a lot of conversations about that. But 321 00:19:40,760 --> 00:19:43,920 Josh: I would love for everyone to converse in the comment section and let us know 322 00:19:43,920 --> 00:19:48,980 Josh: what you think of this absolute crazy chaos of a week that we had with, 323 00:19:49,280 --> 00:19:51,660 Josh: I mean, again, this is like straight out of a Black Mirror episode. 324 00:19:52,120 --> 00:19:55,740 Josh: So if you did enjoy, if you do have takes, we'd love to hear them in the comment 325 00:19:55,740 --> 00:19:58,700 Josh: section down below. If you enjoyed this episode, share it with a friend. 326 00:19:58,900 --> 00:20:03,600 Josh: Last week, this is nowhere, we had our biggest week ever in the history of Limitless. Thank you guys. 327 00:20:03,840 --> 00:20:06,780 Josh: And it was thanks to all of you guys sharing and subscribing and rating five 328 00:20:06,780 --> 00:20:09,220 Josh: stars and even engaging on X. 329 00:20:09,220 --> 00:20:13,540 Josh: We have gotten 50 million impressions on x over the last couple of weeks the 330 00:20:13,540 --> 00:20:16,680 Josh: people are paying attention they are watching we're starting to create the news 331 00:20:16,680 --> 00:20:19,540 Josh: and it's thanks to people like you who are watching and listening sharing this 332 00:20:19,540 --> 00:20:24,640 Josh: with your friends so thank you for joining i mean any final parting thoughts before we wrap up here 333 00:20:24,640 --> 00:20:27,820 Ejaaz: Well yeah um just want to re-emphasize 50 million impressions 334 00:20:27,820 --> 00:20:30,660 Ejaaz: go give josh and i a follow obviously we're saying something 335 00:20:30,660 --> 00:20:33,340 Ejaaz: useful here or we hope we are and we're breaking the news as and when 336 00:20:33,340 --> 00:20:36,060 Ejaaz: it happens um yeah so so please go give it 337 00:20:36,060 --> 00:20:38,840 Ejaaz: a go and then the other thing random for all you folks 338 00:20:38,840 --> 00:20:41,660 Ejaaz: who are still listening to this episode did any of 339 00:20:41,660 --> 00:20:44,940 Ejaaz: you find that claude was acting a little weird over 340 00:20:44,940 --> 00:20:47,720 Ejaaz: the weekend i don't know what your experience was josh 341 00:20:47,720 --> 00:20:50,920 Ejaaz: but when i was talking to claude like it seemed completely 342 00:20:50,920 --> 00:20:55,560 Ejaaz: different from the claude i was talking to on friday and my suspicion is anthropic 343 00:20:55,560 --> 00:20:58,980 Ejaaz: is kind of throttling the intelligence i don't know why maybe there's just too 344 00:20:58,980 --> 00:21:02,840 Ejaaz: much demand since they went to number one in the app store but it wasn't really 345 00:21:02,840 --> 00:21:08,320 Ejaaz: looking good i know a lot of people who switched back to chadgbt after that happened yeah. 346 00:21:08,320 --> 00:21:09,560 Josh: Brief conspiracy corner so 347 00:21:10,120 --> 00:21:13,480 Josh: I think I feel pretty highly that based on benchmarks and things that I've seen 348 00:21:13,480 --> 00:21:16,300 Josh: and just personally using it, the model has degraded slightly. 349 00:21:16,600 --> 00:21:20,820 Josh: And the reason is because as Claude got as big as it did, they didn't get more 350 00:21:20,820 --> 00:21:23,220 Josh: compute power to serve all of these new users. 351 00:21:23,420 --> 00:21:27,060 Josh: So they have to throttle it in some way. And the way they throttle is by limiting 352 00:21:27,060 --> 00:21:29,120 Josh: the upper bound of what reasoning capabilities it has. 353 00:21:29,540 --> 00:21:32,660 Josh: So generally Claude will reason much more. It'll ask itself questions. 354 00:21:32,660 --> 00:21:34,180 Josh: It'll compare. It'll do broader research. 355 00:21:34,380 --> 00:21:37,900 Josh: I think that scope has been compressed a little bit recently because they need 356 00:21:37,900 --> 00:21:41,960 Josh: to make more compute room for um the rest of the people that just joined i mean 357 00:21:41,960 --> 00:21:45,260 Josh: anthropic had a huge week they have a lot of server problems the service went 358 00:21:45,260 --> 00:21:47,720 Josh: down a few times last week like they have just been having a really tough time 359 00:21:47,720 --> 00:21:48,800 Josh: keeping up so i mean they've been 360 00:21:48,800 --> 00:21:53,980 Ejaaz: Signing up they've been signing up 1 million new users per day let me rephrase 361 00:21:53,980 --> 00:21:59,280 Ejaaz: that per day um and i was doing a bit of digging and boris cherny the creator 362 00:21:59,280 --> 00:22:04,160 Ejaaz: of called code basically said that they had defaulted every single person's 363 00:22:04,160 --> 00:22:07,720 Ejaaz: called profile to medium efficiency or medium power. 364 00:22:07,920 --> 00:22:11,020 Ejaaz: And he said, you know, you could upgrade that to high if you want. 365 00:22:11,200 --> 00:22:14,800 Ejaaz: But the point is, I think they're struggling from compute, as you said, 366 00:22:14,940 --> 00:22:17,000 Ejaaz: and they're trying to figure out a way to scale it up. 367 00:22:17,140 --> 00:22:21,180 Ejaaz: Amazon needs to come in and give them some more AWS access or something because 368 00:22:21,180 --> 00:22:24,080 Ejaaz: they don't really have the advantage. They need more GPUs. They need more GPUs. 369 00:22:24,860 --> 00:22:28,080 Ejaaz: Yeah. What's the roundup of this episode? NVIDIA still wins. 370 00:22:28,220 --> 00:22:33,220 Ejaaz: Silicon-based intelligence is still the frontier and Jensen's going to make a lot more money. 371 00:22:33,940 --> 00:22:36,820 Josh: Well there you have it if you've noticed please share thank you for making it 372 00:22:36,820 --> 00:22:38,940 Josh: to the end as always and yeah we'll see you guys in the next episode