WEBVTT - Are We At Peak AI?

0:00:02.480 --> 0:00:03.560
<v Speaker 1>Fall Zone Media.

0:00:05.559 --> 0:00:08.280
<v Speaker 2>Hello and welcome to Better Offline. I'm your host ed

0:00:08.360 --> 0:00:22.200
<v Speaker 2>z tron. It's been just under a year and a

0:00:22.239 --> 0:00:25.960
<v Speaker 2>half since chat gpt, an AI powered chatbot launched by

0:00:26.079 --> 0:00:29.360
<v Speaker 2>so called nonprofit open Ai, ushered in a new investor

0:00:29.360 --> 0:00:32.680
<v Speaker 2>in media hype cycle around how AI would change the world.

0:00:33.040 --> 0:00:36.760
<v Speaker 2>Chat GPT's instant success made both open Ai and its CEO,

0:00:36.920 --> 0:00:40.840
<v Speaker 2>Sam Mortman overnight celebrities. As a result of chat GPT's

0:00:40.920 --> 0:00:44.839
<v Speaker 2>alleged intelligence, which seemingly allowed you to do everything from

0:00:44.880 --> 0:00:47.879
<v Speaker 2>generate an entire essay from a simple prompt to writing

0:00:48.360 --> 0:00:51.640
<v Speaker 2>entire reams of software code. You can theoretically ask it

0:00:51.680 --> 0:00:54.680
<v Speaker 2>anything and it will spit out an intelligent sounding response

0:00:54.880 --> 0:00:57.360
<v Speaker 2>thanks to being trained on terabides of text data, like

0:00:57.400 --> 0:01:00.560
<v Speaker 2>a search engine that's able to think for itself. Ah.

0:01:00.920 --> 0:01:04.000
<v Speaker 2>Big problem is chat gpt doesn't think at all. It

0:01:04.040 --> 0:01:09.080
<v Speaker 2>doesn't know anything. Chat gpt is actually probabilistic. It uses

0:01:09.080 --> 0:01:11.720
<v Speaker 2>a statistical model to generate the next piece of information

0:01:11.760 --> 0:01:14.360
<v Speaker 2>in a sequence. If you ask it what an elephant is,

0:01:14.560 --> 0:01:16.880
<v Speaker 2>it'll guess that the most likely answer to that prompt

0:01:17.080 --> 0:01:19.000
<v Speaker 2>is that an elephant is a large mammal, and then

0:01:19.040 --> 0:01:22.480
<v Speaker 2>perhaps describe its features, such as a long trunk. Chat

0:01:22.600 --> 0:01:25.120
<v Speaker 2>GPT doesn't know what an elephant is, or what a

0:01:25.160 --> 0:01:28.880
<v Speaker 2>trunk is, or what the alefant Haantai family is. It's

0:01:28.880 --> 0:01:32.240
<v Speaker 2>simply ingested enough information to reliably guess what an elephant

0:01:32.280 --> 0:01:35.240
<v Speaker 2>is meant to be, or indeed know that that's what

0:01:35.240 --> 0:01:38.880
<v Speaker 2>you're asking it. This is the technology underpinning the latest

0:01:38.959 --> 0:01:43.920
<v Speaker 2>artificial intelligence boom. It's called generative artificial intelligence, and it's

0:01:43.920 --> 0:01:47.840
<v Speaker 2>powered by large language models, and they underpin tools like

0:01:47.880 --> 0:01:52.559
<v Speaker 2>open ais, chat, GPT, Anthropics, claude x dot COM's horrifying

0:01:52.640 --> 0:01:57.200
<v Speaker 2>chatbot Grock, and of course Google's Gemini. Essentially, they're AI

0:01:57.280 --> 0:02:01.200
<v Speaker 2>systems that ingest vast quantities of written text or other data,

0:02:01.760 --> 0:02:05.160
<v Speaker 2>and then, through mathematics, try and identify patterns of relationships

0:02:05.200 --> 0:02:08.720
<v Speaker 2>between words or symbols, or basically any meaning from the

0:02:08.760 --> 0:02:12.600
<v Speaker 2>text or thing they're being fed. And it almost seems

0:02:12.639 --> 0:02:15.760
<v Speaker 2>like magic because it's able to generate this plausible, seeming

0:02:15.919 --> 0:02:19.920
<v Speaker 2>almost human content at this remarkable speed. These models are

0:02:19.960 --> 0:02:23.280
<v Speaker 2>now capable of generating text, images, and even video in

0:02:23.320 --> 0:02:25.959
<v Speaker 2>response to simple chat prompts or by learning the patterns

0:02:26.000 --> 0:02:29.239
<v Speaker 2>and structures of their data. Yet underneath the hood, there's

0:02:29.280 --> 0:02:34.160
<v Speaker 2>always something a bit wrong. Generative AI, at times authoritatively

0:02:34.200 --> 0:02:37.440
<v Speaker 2>spits out incorrect information, which can range from funny like

0:02:37.480 --> 0:02:40.160
<v Speaker 2>telling you that you can melt an egg, to outright dangerous,

0:02:40.280 --> 0:02:42.919
<v Speaker 2>like when the recently launched AI powered New York City

0:02:43.360 --> 0:02:46.480
<v Speaker 2>chatbot for small business owners started telling them that it

0:02:46.520 --> 0:02:49.760
<v Speaker 2>was legal to fire somebody for refusing to cut their dreadlocks.

0:02:50.560 --> 0:02:53.679
<v Speaker 2>This is why you'll see strange glitches in images generated

0:02:53.720 --> 0:02:57.200
<v Speaker 2>by AI hands with too many fingers, horrifying looking people

0:02:57.240 --> 0:02:59.600
<v Speaker 2>in the back of realistic looking photos, and so on

0:02:59.639 --> 0:03:03.160
<v Speaker 2>and so far. Because these models don't actually know what

0:03:03.200 --> 0:03:08.600
<v Speaker 2>anything is. They don't have meaning, they don't have consciousness

0:03:08.800 --> 0:03:14.320
<v Speaker 2>or intelligence. They're guessing, and when they guess, they sometimes hallucinate,

0:03:14.520 --> 0:03:17.919
<v Speaker 2>which I'll get too soon. And while they might be

0:03:18.000 --> 0:03:22.000
<v Speaker 2>really really good at guessing, there are effectively a very

0:03:22.120 --> 0:03:25.120
<v Speaker 2>very powerful version of auto complete. I don't know anything.

0:03:25.720 --> 0:03:29.320
<v Speaker 2>I really mean that these things aren't even intelligent. But

0:03:29.400 --> 0:03:31.799
<v Speaker 2>because these models seem like they know stuff, and they

0:03:31.840 --> 0:03:33.760
<v Speaker 2>seem to be able to do stuff, and the things

0:03:33.760 --> 0:03:36.880
<v Speaker 2>that they create almost seem right. The media and the

0:03:36.920 --> 0:03:40.240
<v Speaker 2>Vocal Investor class on Twitter have declared that large language

0:03:40.240 --> 0:03:44.680
<v Speaker 2>models would change everything to them. Llms like chat GPT

0:03:44.960 --> 0:03:49.760
<v Speaker 2>would upend entire business models, render once unassailable tech giants nunerable,

0:03:49.960 --> 0:03:53.680
<v Speaker 2>and rewrite our entire economic playbook by turning entire industries

0:03:53.840 --> 0:03:56.240
<v Speaker 2>into something you tell a chatbot to do. In a sentence,

0:03:57.280 --> 0:03:59.240
<v Speaker 2>you know, it doesn't really matter that genera ive AA

0:03:59.320 --> 0:04:02.480
<v Speaker 2>is mostly good at reams of generic slop, and that

0:04:02.520 --> 0:04:05.840
<v Speaker 2>it's also clogging services like Amazon's Kindle eBookstore. And I

0:04:05.880 --> 0:04:08.400
<v Speaker 2>guess the rest of the Internet with generative content that

0:04:08.480 --> 0:04:13.520
<v Speaker 2>doesn't matter at all, because it's kind of good. Obviously,

0:04:13.560 --> 0:04:17.520
<v Speaker 2>I'm being sarcastic. This is all very, very bad. In

0:04:17.560 --> 0:04:20.760
<v Speaker 2>the last year, there have been hundreds of muling articles

0:04:20.760 --> 0:04:23.960
<v Speaker 2>about how AI will replace everything from drive through workers

0:04:23.960 --> 0:04:27.680
<v Speaker 2>to medical professionals. Theoretically, you could just feed whatever information

0:04:27.720 --> 0:04:30.640
<v Speaker 2>a potential customer could ask for into a vast database

0:04:31.040 --> 0:04:33.920
<v Speaker 2>and have an AI chew it up, and then they

0:04:34.200 --> 0:04:37.240
<v Speaker 2>could just generate exactly the answer you'd need. AI would

0:04:37.320 --> 0:04:40.880
<v Speaker 2>just naturally slip into areas of disorganization and inefficiency and

0:04:40.960 --> 0:04:44.760
<v Speaker 2>spit out remarkable new ideas, all with minimum human input.

0:04:46.360 --> 0:04:48.840
<v Speaker 2>In every one of these stories carries with them a

0:04:48.920 --> 0:04:52.200
<v Speaker 2>shared belief one might even call it a shared hallucination.

0:04:53.560 --> 0:04:56.240
<v Speaker 2>And they all believe that generative AI will actually be

0:04:56.279 --> 0:04:58.560
<v Speaker 2>able to do these things, that it will actually be

0:04:58.560 --> 0:05:01.080
<v Speaker 2>able to replace people. And what we're seeing today is

0:05:01.160 --> 0:05:05.080
<v Speaker 2>just the beginning of our glorious automated future. But what

0:05:05.160 --> 0:05:09.560
<v Speaker 2>if it's not. What if generative AI can't actually do

0:05:09.760 --> 0:05:13.680
<v Speaker 2>much more than it can today? What if we're actually

0:05:13.720 --> 0:05:18.320
<v Speaker 2>at peak AI. In the next two episodes, I'm going

0:05:18.360 --> 0:05:20.200
<v Speaker 2>to tell you why I think that is. I'm going

0:05:20.279 --> 0:05:22.320
<v Speaker 2>to tell you how I think this whole thing falls apart.

0:05:33.200 --> 0:05:36.560
<v Speaker 2>AI's media hype train has been fairly relentless since November

0:05:36.560 --> 0:05:40.160
<v Speaker 2>twenty twenty two, when chat GBT launched, and AI champions

0:05:40.200 --> 0:05:43.479
<v Speaker 2>like open Ai CEO Sam Mortman have proven all too

0:05:43.520 --> 0:05:46.400
<v Speaker 2>willing to grease its wheels, making bold promises about what

0:05:46.440 --> 0:05:50.000
<v Speaker 2>AI could do and how today's problems are so easily overcome.

0:05:50.640 --> 0:05:53.400
<v Speaker 2>It's also helped that the tech media has largely accepted

0:05:53.400 --> 0:05:57.520
<v Speaker 2>these promises without asking basic questions like how and when

0:05:57.560 --> 0:06:00.320
<v Speaker 2>will it do this stuff? And can it do this stuff?

0:06:00.880 --> 0:06:03.960
<v Speaker 2>Don't believe me, Go and look at any interview with

0:06:04.080 --> 0:06:07.719
<v Speaker 2>Sam Moltman from the last few years, watch any of them.

0:06:07.760 --> 0:06:10.520
<v Speaker 2>In fact, just look at any AI figurehead getting interviewed

0:06:10.680 --> 0:06:13.080
<v Speaker 2>and count the amount of times they've actually received any

0:06:13.120 --> 0:06:17.200
<v Speaker 2>pushback or been asked to elaborate on any specific issue.

0:06:17.279 --> 0:06:20.000
<v Speaker 2>It's actually very rare. Let me play you one of

0:06:20.000 --> 0:06:23.280
<v Speaker 2>the few times that anyone's actually interrogated an AI person.

0:06:23.720 --> 0:06:26.400
<v Speaker 2>Specifically Joanna Stern of The Wall Street Journal, who you

0:06:26.480 --> 0:06:29.160
<v Speaker 2>might remember from the Vision Pro episode A Better Offline.

0:06:29.520 --> 0:06:33.800
<v Speaker 2>She interviewed open AI's Chief Technology Officer, Mirror Marathi about Sora,

0:06:33.880 --> 0:06:37.159
<v Speaker 2>which is open aiy's video based version of chat GPT

0:06:37.320 --> 0:06:40.640
<v Speaker 2>where you can theoretically ask it to generate videos. Just

0:06:40.640 --> 0:06:42.920
<v Speaker 2>to be clear, it's unreleased and unclear whether it'll ever

0:06:42.960 --> 0:06:45.040
<v Speaker 2>actually get released, and the videos look good at first,

0:06:45.279 --> 0:06:47.880
<v Speaker 2>then they look really weird. But just listen to this

0:06:47.960 --> 0:06:52.600
<v Speaker 2>particular question. It's Joanna asking Mirror, the CTO of open

0:06:52.640 --> 0:06:56.400
<v Speaker 2>ai and eighty billion dollar AI company, Hey, did you

0:06:56.520 --> 0:06:59.040
<v Speaker 2>train on YouTube? What data was.

0:06:59.040 --> 0:06:59.640
<v Speaker 3>Used to train?

0:07:01.200 --> 0:07:07.479
<v Speaker 1>We used publicly available data and licensed data, so videos

0:07:07.520 --> 0:07:08.000
<v Speaker 1>on YouTube?

0:07:09.720 --> 0:07:11.520
<v Speaker 2>Now, I encourage you to go and look up this

0:07:11.600 --> 0:07:16.000
<v Speaker 2>clip because at this point Maurti makes the strangest face

0:07:16.280 --> 0:07:18.000
<v Speaker 2>I've ever seen in a tech interview.

0:07:19.040 --> 0:07:19.840
<v Speaker 1>I'm actually not.

0:07:19.760 --> 0:07:26.520
<v Speaker 3>Sure about that okay, videos from Facebook, Instagram.

0:07:26.640 --> 0:07:32.360
<v Speaker 1>You know, if they were publicly available available, publicly available

0:07:32.400 --> 0:07:38.200
<v Speaker 1>to use, there might be the data, but I'm not sure.

0:07:38.280 --> 0:07:39.480
<v Speaker 1>I'm not confident about it.

0:07:39.560 --> 0:07:41.880
<v Speaker 3>What about Shutterstock, I know you guys have a deal

0:07:41.960 --> 0:07:44.200
<v Speaker 3>with them.

0:07:44.280 --> 0:07:46.080
<v Speaker 1>I'm just not going to go into the details of

0:07:46.400 --> 0:07:49.360
<v Speaker 1>the data that was that was used, but it was

0:07:49.400 --> 0:07:51.240
<v Speaker 1>publicly available or licensed data.

0:07:52.160 --> 0:07:54.640
<v Speaker 2>The remarkable part about this interview is that it's a

0:07:54.760 --> 0:07:58.120
<v Speaker 2>relatively simple question. You, as the CTO of an eighty

0:07:58.160 --> 0:08:01.960
<v Speaker 2>billion dollar AI company, what training data did you use

0:08:02.000 --> 0:08:05.280
<v Speaker 2>to train your model? Did you use YouTube? So yes

0:08:05.400 --> 0:08:09.560
<v Speaker 2>or no? Question? Mirror mirror, answer the bloody question, mirror

0:08:11.280 --> 0:08:13.880
<v Speaker 2>all right, right? The answer is, of course Open Ai

0:08:14.400 --> 0:08:17.560
<v Speaker 2>likely trained it's video generating model, Sora on YouTube videos,

0:08:18.240 --> 0:08:21.680
<v Speaker 2>which might be why they're yet to launch it. And

0:08:21.920 --> 0:08:26.560
<v Speaker 2>the videos generated by Soora also feature some remarkably similar

0:08:26.560 --> 0:08:30.440
<v Speaker 2>images to say, SpongeBob SquarePants, and I wouldn't be surprised

0:08:30.480 --> 0:08:33.960
<v Speaker 2>if they carry with them multiple weird biases about race

0:08:34.000 --> 0:08:37.559
<v Speaker 2>and gender that we'll see in the future. But also

0:08:37.640 --> 0:08:41.360
<v Speaker 2>when you watch these videos, much like most generative AI content,

0:08:41.600 --> 0:08:44.120
<v Speaker 2>there's something a bit off about them. In The Wall

0:08:44.120 --> 0:08:46.760
<v Speaker 2>Street Journal's interview. You get to see some of the

0:08:46.960 --> 0:08:48.920
<v Speaker 2>prompts that were used and some of the videos that

0:08:48.960 --> 0:08:51.920
<v Speaker 2>came out, and you see crazy things happening, like a

0:08:52.000 --> 0:08:56.440
<v Speaker 2>robot completely changing shape as it turns, cars disappearing and

0:08:56.520 --> 0:09:00.560
<v Speaker 2>appearing behind the robot. It's not very good. It seems

0:09:00.559 --> 0:09:02.960
<v Speaker 2>cool at first. If you squint really hard, it looks real,

0:09:03.320 --> 0:09:06.160
<v Speaker 2>but there's always something off. And that's because, as I've

0:09:06.160 --> 0:09:08.360
<v Speaker 2>said before, these models don't know anything and don't know

0:09:08.360 --> 0:09:10.040
<v Speaker 2>what at robot looks. They can make a really good

0:09:10.080 --> 0:09:16.400
<v Speaker 2>guess though. Anyway, Sterne's interview with Marati of Open AI

0:09:16.760 --> 0:09:19.480
<v Speaker 2>is a great example of how the entire AI artifst

0:09:19.520 --> 0:09:23.199
<v Speaker 2>falls apart at the slightest touch, because it's fundamentally flawed

0:09:23.480 --> 0:09:27.000
<v Speaker 2>and not actually able to deliver the society defining promises

0:09:27.160 --> 0:09:29.680
<v Speaker 2>that Sam Morltman and the venture capital sect would have

0:09:29.760 --> 0:09:33.400
<v Speaker 2>you believe. In a year and a half, despite billions

0:09:33.400 --> 0:09:37.320
<v Speaker 2>of dollars of investment, despite every major media outlet claiming otherwise,

0:09:37.559 --> 0:09:42.400
<v Speaker 2>generative artificial intelligence has proven itself incapable of replacing or

0:09:42.440 --> 0:09:46.920
<v Speaker 2>even meaningfully enhancing human work. And the thing is, all

0:09:46.920 --> 0:09:49.600
<v Speaker 2>of these problems I'm talking about with generative AI all

0:09:49.640 --> 0:09:52.800
<v Speaker 2>of these hallucinations, all of these weird artifacts that are

0:09:52.800 --> 0:09:56.080
<v Speaker 2>popping up throughout these videos, the weird mistakes that the

0:09:56.120 --> 0:09:59.640
<v Speaker 2>texts that are popped out by chat GPT have, all

0:09:59.679 --> 0:10:03.679
<v Speaker 2>of these. The problems are problems that aren't necessarily just technological.

0:10:04.120 --> 0:10:10.080
<v Speaker 2>Their physics, they're mathematics. These aren't things you can just outrun.

0:10:11.040 --> 0:10:14.760
<v Speaker 2>And I believe that there are four intractable problems that

0:10:14.840 --> 0:10:18.080
<v Speaker 2>will stop generative AI from progressing much further than it

0:10:18.160 --> 0:10:21.440
<v Speaker 2>is today. The first is, of course, its energy demands,

0:10:21.480 --> 0:10:25.120
<v Speaker 2>the massive amounts of power requires. The second are its

0:10:25.120 --> 0:10:29.640
<v Speaker 2>computational demands, the amount of compute power it requires to

0:10:29.679 --> 0:10:34.280
<v Speaker 2>even crunch the simplest things out of chat GPT, It's hallucinations,

0:10:34.320 --> 0:10:37.680
<v Speaker 2>the authoritative failures it makes when it spits out nonsense

0:10:37.760 --> 0:10:41.520
<v Speaker 2>or creates a human hand with eighteen fingers, and of

0:10:41.559 --> 0:10:44.160
<v Speaker 2>course the fact that these large language models have an

0:10:44.160 --> 0:10:49.240
<v Speaker 2>insatiable hunger for more training data. Now, let me break

0:10:49.240 --> 0:10:54.560
<v Speaker 2>that down. Large language models are extremely technologically and environmentally demanding.

0:10:55.040 --> 0:10:57.160
<v Speaker 2>The New York Are reported in March twenty twenty four

0:10:57.320 --> 0:11:00.280
<v Speaker 2>the CHATGBT uses more than half a million kis what

0:11:00.360 --> 0:11:03.240
<v Speaker 2>hours of electricity to respond to the two hundred million

0:11:03.280 --> 0:11:07.000
<v Speaker 2>requests it receives in a day, or seventeen thousand times

0:11:07.040 --> 0:11:09.839
<v Speaker 2>the amount that the average American household uses in a day,

0:11:10.080 --> 0:11:12.600
<v Speaker 2>and others have suggested it might be as high as

0:11:12.679 --> 0:11:18.079
<v Speaker 2>thirty three thousand households worth. Generative AI models demand specialist

0:11:18.120 --> 0:11:21.760
<v Speaker 2>chips called graphics processing units, typically a souped up version

0:11:21.800 --> 0:11:23.959
<v Speaker 2>of the technology used to drive the graphics in a

0:11:24.000 --> 0:11:27.480
<v Speaker 2>gaming console, albeit at a much higher cost, with each

0:11:27.640 --> 0:11:31.199
<v Speaker 2>one costing tens of thousands of dollars each. They do

0:11:31.240 --> 0:11:34.600
<v Speaker 2>this because large language models like chat GPT are highly

0:11:34.640 --> 0:11:38.400
<v Speaker 2>computationally intensive. I'm going to break that down. Don't worry.

0:11:38.880 --> 0:11:41.760
<v Speaker 2>When you ask chat GPT a question, it tokenizes it,

0:11:42.240 --> 0:11:45.359
<v Speaker 2>breaking it down into smaller parts for the model to understand.

0:11:46.760 --> 0:11:50.120
<v Speaker 2>It then feeds these tokens into various mechanisms that help

0:11:50.160 --> 0:11:52.480
<v Speaker 2>it understand the meaning of the thing you asked it

0:11:52.520 --> 0:11:56.040
<v Speaker 2>to do based on the parameters that it learned in training.

0:11:56.240 --> 0:11:58.920
<v Speaker 2>Chat GPT generates a response by predicting the most likely

0:11:58.960 --> 0:12:00.800
<v Speaker 2>sequence of things that you might want it to do.

0:12:01.160 --> 0:12:03.720
<v Speaker 2>An answer to a question, an image, so on, and

0:12:03.760 --> 0:12:07.560
<v Speaker 2>so forth. Each one of these steps is extremely demanding,

0:12:07.720 --> 0:12:11.360
<v Speaker 2>processing hundreds of billions of these parameters learned patterns from

0:12:11.520 --> 0:12:14.440
<v Speaker 2>ingesting training data such as how the English language works

0:12:14.480 --> 0:12:16.800
<v Speaker 2>or what a dog looks like to produce even the

0:12:16.840 --> 0:12:21.280
<v Speaker 2>simplest thing. Training these models is equally intensive, requiring chat

0:12:21.280 --> 0:12:24.440
<v Speaker 2>GPT to process massive amounts of data. Another problem I'll

0:12:24.440 --> 0:12:27.320
<v Speaker 2>get to in a bit adjusting those hundreds of billions

0:12:27.320 --> 0:12:30.000
<v Speaker 2>of parameters and developing new ones based on what the

0:12:30.120 --> 0:12:33.439
<v Speaker 2>data says as it quote unquote learns more. Though as

0:12:33.440 --> 0:12:37.120
<v Speaker 2>we're clear, chap GPT doesn't learn anything. It just makes

0:12:37.160 --> 0:12:40.960
<v Speaker 2>new parameters to read things. A model like chat GBT grows,

0:12:41.000 --> 0:12:44.199
<v Speaker 2>making it more complex, which in turn requires more data

0:12:44.240 --> 0:12:47.520
<v Speaker 2>to train on and more compute power to both ingest

0:12:47.559 --> 0:12:50.439
<v Speaker 2>the data, create more parameters and turn it into something

0:12:50.480 --> 0:12:54.240
<v Speaker 2>resembling an answer, And because it doesn't know anything, it's

0:12:54.280 --> 0:12:57.200
<v Speaker 2>suggesting the most likely to be correct answer, which leads

0:12:57.200 --> 0:13:00.360
<v Speaker 2>it to hallucinating and correct things that, based on proper ability,

0:13:00.600 --> 0:13:03.920
<v Speaker 2>kind of seem like the right thing to say. These

0:13:03.960 --> 0:13:07.200
<v Speaker 2>hallucinations are the dirty little secret of generative AI, and

0:13:07.240 --> 0:13:09.640
<v Speaker 2>are impossible to avoid thanks to the fact that every

0:13:09.720 --> 0:13:13.480
<v Speaker 2>single thing these models say is a mathematical equation rather

0:13:13.559 --> 0:13:17.520
<v Speaker 2>than any kind of intellectual exercise. If you ask CHATGPT

0:13:17.840 --> 0:13:19.560
<v Speaker 2>how many days there are in a week. It doesn't

0:13:19.600 --> 0:13:21.720
<v Speaker 2>know that there are seven days, but it's been trained

0:13:21.720 --> 0:13:24.360
<v Speaker 2>on patterns of language and generates a result based on

0:13:24.400 --> 0:13:27.600
<v Speaker 2>those patterns, which at times can be correct and can

0:13:27.640 --> 0:13:32.240
<v Speaker 2>also be wrong. There's no way of fixing this problem.

0:13:33.000 --> 0:13:36.200
<v Speaker 2>You can mitigate it, you can make it less likely

0:13:36.200 --> 0:13:39.640
<v Speaker 2>it will mess up, but hallucinations will happen because there

0:13:39.679 --> 0:13:43.160
<v Speaker 2>is no consciousness. It is not learning anything. This thing

0:13:43.320 --> 0:13:48.120
<v Speaker 2>has no knowledge. More computing power would allow it more

0:13:48.160 --> 0:13:50.720
<v Speaker 2>parameters to give it more rules, so that a generative

0:13:50.720 --> 0:13:53.200
<v Speaker 2>AI will be more likely to give a correct answer,

0:13:53.679 --> 0:13:56.880
<v Speaker 2>but there's no eliminating them, and doing so may require

0:13:56.920 --> 0:14:01.160
<v Speaker 2>more computing power than actually exists or is possible without

0:14:01.160 --> 0:14:04.400
<v Speaker 2>an AI of consciousness, an impossible dream known as average

0:14:04.440 --> 0:14:07.760
<v Speaker 2>generalized intelligence that Sam Mortman would have you believe is imminent.

0:14:08.280 --> 0:14:13.280
<v Speaker 2>There's really no solving hallucinations. When you answer questions using probability,

0:14:13.920 --> 0:14:16.880
<v Speaker 2>you're always going to have mistakes because you're not actually

0:14:16.960 --> 0:14:21.560
<v Speaker 2>answering them using knowledge, intellect, or experience. You're using dice rolls.

0:14:22.200 --> 0:14:25.520
<v Speaker 2>It's a bloody game of dungeons and dragons. We turned

0:14:25.520 --> 0:14:30.200
<v Speaker 2>in Carter into dungeons and dragons anyway. A newly published

0:14:30.240 --> 0:14:33.280
<v Speaker 2>paper by Tepo Felon and Matthias Holweg of the University

0:14:33.280 --> 0:14:36.440
<v Speaker 2>of Oxford agrees funding that large language models like chat

0:14:36.480 --> 0:14:40.680
<v Speaker 2>GPT are incapable of generating new knowledge. It's a remarkably

0:14:40.760 --> 0:14:44.280
<v Speaker 2>in depth rundown of the fundamental differences between a large

0:14:44.320 --> 0:14:47.160
<v Speaker 2>language model and a human brain, and it combines both

0:14:47.200 --> 0:14:51.080
<v Speaker 2>psychological and mathematical research going back to child psychology as

0:14:51.120 --> 0:14:54.480
<v Speaker 2>well the basic building blocks of how we consume and

0:14:54.560 --> 0:14:57.080
<v Speaker 2>learn things and how we make decisions. As a result,

0:14:57.720 --> 0:15:01.440
<v Speaker 2>the paper title theory is or You Need AI Human

0:15:01.480 --> 0:15:05.360
<v Speaker 2>Cognition and Decision Making argues that AI's data and prediction

0:15:05.440 --> 0:15:09.920
<v Speaker 2>based orientation is an incomplete view of human cognition and

0:15:10.160 --> 0:15:13.560
<v Speaker 2>the forward thinking theorizing of the human mind. In Layman's terms,

0:15:13.880 --> 0:15:16.200
<v Speaker 2>the mess of the information we've learned over our lives,

0:15:16.200 --> 0:15:19.560
<v Speaker 2>our experiences, and our ability to look forward and consider

0:15:19.880 --> 0:15:23.560
<v Speaker 2>the future is just fundamentally different to a model that

0:15:23.640 --> 0:15:27.240
<v Speaker 2>predicts things only based off of past data. Think of

0:15:27.280 --> 0:15:30.480
<v Speaker 2>it like this, If you've read a book and you

0:15:30.560 --> 0:15:33.160
<v Speaker 2>might think about writing a new book based on those ideas,

0:15:33.600 --> 0:15:36.320
<v Speaker 2>you're not remembering every part of the book. You don't

0:15:36.360 --> 0:15:40.560
<v Speaker 2>have a perfect memory, and you're also constantly thinking about

0:15:40.600 --> 0:15:43.040
<v Speaker 2>things as your day goes on. The human brain is

0:15:43.040 --> 0:15:47.080
<v Speaker 2>a goddamn mess. Generative AI is in some level stuck

0:15:47.120 --> 0:15:50.960
<v Speaker 2>in amber. Though the billions of parameters might change, the

0:15:51.080 --> 0:15:54.080
<v Speaker 2>data never does the way it consumes the data. May

0:15:54.160 --> 0:15:58.800
<v Speaker 2>be that the data doesn't change. In essence, generative AI

0:15:58.920 --> 0:16:01.240
<v Speaker 2>is held back by the fact that it can't consider

0:16:01.280 --> 0:16:04.760
<v Speaker 2>the future and is actually permanently mired in the data

0:16:04.800 --> 0:16:09.480
<v Speaker 2>of the past. Their largest problem might be a fast,

0:16:09.520 --> 0:16:12.760
<v Speaker 2>simpler one, a farcillio, or a kind of an ironic one.

0:16:13.920 --> 0:16:16.240
<v Speaker 2>There might not be enough data for these bloody things

0:16:16.280 --> 0:16:31.080
<v Speaker 2>to actually train on. While the Internet may at times

0:16:31.120 --> 0:16:34.440
<v Speaker 2>feel limitless, a researcher recently told The Wall Street Journal

0:16:35.120 --> 0:16:37.840
<v Speaker 2>that only a tenth of the most commonly used web

0:16:37.920 --> 0:16:40.800
<v Speaker 2>data set, the common Crawl, are freely available. Two hundred

0:16:40.800 --> 0:16:43.880
<v Speaker 2>and fifty billion page dump of the web's information is

0:16:43.920 --> 0:16:47.520
<v Speaker 2>actually of high enough quality data for large language models

0:16:47.520 --> 0:16:51.080
<v Speaker 2>like CHATGBT to actually train on. Putting aside the fact

0:16:51.080 --> 0:16:52.880
<v Speaker 2>that I can't find a single definition of what high

0:16:52.960 --> 0:16:58.400
<v Speaker 2>quality actually means, the researcher pat Blow Vilobos suggested that

0:16:58.440 --> 0:17:01.360
<v Speaker 2>the next version of chat GBT would require more than

0:17:01.480 --> 0:17:04.400
<v Speaker 2>five times the amount of data it took to train

0:17:04.400 --> 0:17:07.159
<v Speaker 2>in its previous version, GPT four. The new one is

0:17:07.200 --> 0:17:10.240
<v Speaker 2>called GPT five. By the way, and other researchers have

0:17:10.320 --> 0:17:12.520
<v Speaker 2>suggested that AI companies are going to run out of

0:17:12.560 --> 0:17:17.960
<v Speaker 2>training data in the next two years. Now. That sounds dire,

0:17:18.040 --> 0:17:20.280
<v Speaker 2>but don't worry. They've come up with a very funny

0:17:20.359 --> 0:17:24.440
<v Speaker 2>and extremely stupid idea to fix it. One specifically posed

0:17:24.480 --> 0:17:27.200
<v Speaker 2>by The Wall Street Journal is that the AR companies

0:17:27.200 --> 0:17:29.800
<v Speaker 2>are going to create their own synthetic data to train

0:17:29.880 --> 0:17:33.440
<v Speaker 2>their models, a computer science version of inbreeding. The researcher

0:17:33.680 --> 0:17:38.920
<v Speaker 2>Jason Stadowski calls habsburg AI. This is, of course, an

0:17:38.960 --> 0:17:43.560
<v Speaker 2>absolutely terrible idea. A research paper from last year found

0:17:43.600 --> 0:17:47.200
<v Speaker 2>that feeding model generated data into models to train them

0:17:47.240 --> 0:17:51.000
<v Speaker 2>creates something called model collapse, a degenerative learning process where

0:17:51.040 --> 0:17:54.040
<v Speaker 2>models start forgetting improbable events over time as the model

0:17:54.040 --> 0:17:57.560
<v Speaker 2>becomes poison with its own projection of reality. The paper,

0:17:57.800 --> 0:18:00.640
<v Speaker 2>called the Curse of recursion. Training on generated data makes

0:18:00.680 --> 0:18:03.960
<v Speaker 2>models forgam highlights an example where feeding a generative AI

0:18:04.040 --> 0:18:07.400
<v Speaker 2>its own data eventually destroys its ability to answer questions,

0:18:07.720 --> 0:18:10.840
<v Speaker 2>and within nine generations. One answered a simple prompt about

0:18:10.920 --> 0:18:14.560
<v Speaker 2>architecture with an insane screed about jack rabbits full of

0:18:14.840 --> 0:18:19.800
<v Speaker 2>at symbols and weird characters. So not to worry again.

0:18:20.280 --> 0:18:22.560
<v Speaker 2>The tech overlords have come up with a great idea

0:18:22.600 --> 0:18:25.680
<v Speaker 2>to fix this problem. Their common retort to the problem

0:18:25.760 --> 0:18:29.400
<v Speaker 2>of synthetic data is that you could use another generative

0:18:29.400 --> 0:18:32.600
<v Speaker 2>AI to monitor the synthetic data being fed into all

0:18:32.680 --> 0:18:36.400
<v Speaker 2>model to make sure it's right. At this point, I'd

0:18:36.440 --> 0:18:39.080
<v Speaker 2>like to get slightly angry. Are you kidding me? Are

0:18:39.119 --> 0:18:41.320
<v Speaker 2>you fucking kidding me? You're saying that the way to

0:18:41.359 --> 0:18:44.720
<v Speaker 2>make sure the data generated by an unreliable generative AI

0:18:45.200 --> 0:18:48.560
<v Speaker 2>is to use another generative AI, one with the same

0:18:48.640 --> 0:18:53.240
<v Speaker 2>goddamn problems, which also hallucinates information that knows nothing. You're

0:18:53.240 --> 0:18:55.359
<v Speaker 2>going to use that AI to monitor whether the data

0:18:55.640 --> 0:18:58.000
<v Speaker 2>that is created by an AI is any good? Are

0:18:58.000 --> 0:19:01.800
<v Speaker 2>you completely insane? You insane? You're going to feed the

0:19:01.840 --> 0:19:04.240
<v Speaker 2>crap from the crap machine into another crap machine to

0:19:04.280 --> 0:19:07.480
<v Speaker 2>make it not make crap? Why am I reading journalists

0:19:07.520 --> 0:19:10.960
<v Speaker 2>credulously printing this ridiculous solution in the New York Goddamn

0:19:11.000 --> 0:19:14.440
<v Speaker 2>Times every time? Every time, these bubbles are inflated because

0:19:14.480 --> 0:19:17.200
<v Speaker 2>tech executives are able to get their half ass, half

0:19:17.240 --> 0:19:20.399
<v Speaker 2>baked solutions parroted by reporters who should know better. You

0:19:20.440 --> 0:19:22.199
<v Speaker 2>don't have to give them the better affair of the

0:19:22.200 --> 0:19:24.840
<v Speaker 2>goddamn fucking doubt. This is how we got the bloody

0:19:24.840 --> 0:19:31.520
<v Speaker 2>meta us. Pardon me, I've calmed down now. Anyway. Anyway,

0:19:31.920 --> 0:19:34.920
<v Speaker 2>if you're worried about model collapse, you're already too late,

0:19:35.200 --> 0:19:37.960
<v Speaker 2>as these models are likely already being fed their own data.

0:19:38.840 --> 0:19:41.679
<v Speaker 2>You see, these models are trained on the web, as

0:19:41.720 --> 0:19:45.119
<v Speaker 2>I previously told you, and they're desperate. They need data.

0:19:45.200 --> 0:19:47.960
<v Speaker 2>They need more stuff. They need more stuff to ingest

0:19:48.040 --> 0:19:51.439
<v Speaker 2>so they can spit out more stuff. The problem is

0:19:51.440 --> 0:19:54.240
<v Speaker 2>that these machines are purpose built to make a lot

0:19:54.240 --> 0:19:58.600
<v Speaker 2>of content, and so the Web's already being filled with

0:19:58.680 --> 0:20:03.560
<v Speaker 2>generative AI. Generative AI is already spamming the Internet. A

0:20:03.680 --> 0:20:06.840
<v Speaker 2>report from four oh four Media from last week said

0:20:06.880 --> 0:20:11.280
<v Speaker 2>that Google Books has already started to index several different

0:20:11.280 --> 0:20:14.360
<v Speaker 2>works that were potentially written by AI, featuring the hallmark

0:20:14.400 --> 0:20:17.720
<v Speaker 2>generic writing tropes of these models. Four or four Media

0:20:17.760 --> 0:20:19.879
<v Speaker 2>also reports that the same thing is happening over at

0:20:19.880 --> 0:20:22.879
<v Speaker 2>Google Scholar their index of scientific papers with one hundred

0:20:22.880 --> 0:20:25.840
<v Speaker 2>and fifteen different articles featuring the phrase as of my

0:20:25.920 --> 0:20:30.280
<v Speaker 2>last knowledge update a specific phrase spat out by generative models.

0:20:32.080 --> 0:20:34.560
<v Speaker 2>This is really bad, by the way, and this is

0:20:34.640 --> 0:20:36.719
<v Speaker 2>only going to get worse. When you have an Internet

0:20:36.800 --> 0:20:39.960
<v Speaker 2>economy that is built so that the people that can

0:20:40.000 --> 0:20:42.280
<v Speaker 2>put the most out there will probably get the most traffic.

0:20:42.840 --> 0:20:44.800
<v Speaker 2>They're going to use these tools. These tools are great

0:20:44.800 --> 0:20:46.720
<v Speaker 2>for that. If you don't give a rap fuck about

0:20:46.720 --> 0:20:48.680
<v Speaker 2>the quality, this is the best thing in the world

0:20:48.680 --> 0:20:54.160
<v Speaker 2>for you. And that's the thing. This is a problem

0:20:54.920 --> 0:20:58.119
<v Speaker 2>both created and caused by these models. You see. The

0:20:58.160 --> 0:21:00.880
<v Speaker 2>other dirty little secret of generative II is that these

0:21:00.920 --> 0:21:04.920
<v Speaker 2>models unashamedly plagiarize the entire web, leading outlets like The

0:21:04.960 --> 0:21:07.480
<v Speaker 2>New York Times and authors like John Grisham to sue

0:21:07.520 --> 0:21:11.719
<v Speaker 2>open ai for plagiarism. While open ai won't reveal exactly

0:21:11.720 --> 0:21:14.000
<v Speaker 2>what their training data is, the New York Times is

0:21:14.080 --> 0:21:18.400
<v Speaker 2>able to successfully make chat gpt reproduce content from the newspaper,

0:21:18.640 --> 0:21:21.119
<v Speaker 2>and the company has repeatedly said that it trains on

0:21:21.160 --> 0:21:24.960
<v Speaker 2>publicly available data from the Internet, which will naturally include

0:21:25.040 --> 0:21:29.080
<v Speaker 2>things like Google scholar and Google Books. The Times also

0:21:29.160 --> 0:21:31.840
<v Speaker 2>reports that open ai has become so desperate for data

0:21:31.920 --> 0:21:35.040
<v Speaker 2>that they've used their whisper tool to transcribe YouTube videos

0:21:35.240 --> 0:21:39.600
<v Speaker 2>into texts to feed into chat GPT's training data. Pretty sure,

0:21:39.640 --> 0:21:44.840
<v Speaker 2>that's plagiarism, but who am I to tell you? And

0:21:44.920 --> 0:21:47.840
<v Speaker 2>as the web gets increasingly pumped full of this generative content,

0:21:47.920 --> 0:21:50.639
<v Speaker 2>these models are going to just start eating their own swill,

0:21:51.119 --> 0:21:56.080
<v Speaker 2>slowly corrupting themselves in a kind of ironic death. According

0:21:56.080 --> 0:21:58.640
<v Speaker 2>to Zakhar Schumelov, one of the authors of the model

0:21:58.680 --> 0:22:01.639
<v Speaker 2>collapsed paper at the University of Cambridge, the unique problem

0:22:01.640 --> 0:22:05.000
<v Speaker 2>that synthetic data creates is that it lacks human errors.

0:22:05.760 --> 0:22:08.080
<v Speaker 2>Human made training data, by the nature of it being

0:22:08.080 --> 0:22:10.879
<v Speaker 2>written by a human, includes errors and imperfections, and models

0:22:10.960 --> 0:22:13.919
<v Speaker 2>need to be robust to such errors. So what do

0:22:13.960 --> 0:22:15.960
<v Speaker 2>we do if models are trained off of content created

0:22:16.000 --> 0:22:20.000
<v Speaker 2>without them? Do we introduce the errors ourselves? How many

0:22:20.080 --> 0:22:22.680
<v Speaker 2>errors are there? How do we can introduce them more?

0:22:22.800 --> 0:22:25.959
<v Speaker 2>And indeed, what are the errors? What do they look like?

0:22:26.040 --> 0:22:28.680
<v Speaker 2>Do we even know? Are we conscious of the errors

0:22:28.680 --> 0:22:33.960
<v Speaker 2>in the human language that make us human? The models aren't? Well,

0:22:33.960 --> 0:22:37.520
<v Speaker 2>maybe they are. It's kind of unclear. It's kind of

0:22:37.560 --> 0:22:41.399
<v Speaker 2>tough to express how deeply dangerous the synthetic data idea

0:22:41.600 --> 0:22:45.520
<v Speaker 2>is for AI models like chat, GPT and Claude are

0:22:45.520 --> 0:22:48.919
<v Speaker 2>deeply dependent on training data to improve their outputs, and

0:22:48.960 --> 0:22:52.040
<v Speaker 2>their very existence is actively impeding the creation of the

0:22:52.160 --> 0:22:55.080
<v Speaker 2>very thing they need to survive. While publishers like Axel

0:22:55.160 --> 0:22:57.600
<v Speaker 2>Springer have cut deals to license their companies data to

0:22:57.640 --> 0:23:00.480
<v Speaker 2>chat GPT for training purposes, this money is flowing to

0:23:00.480 --> 0:23:02.960
<v Speaker 2>the writers that create the content that open Ai and

0:23:03.040 --> 0:23:07.000
<v Speaker 2>Anthropic need to grow their models much further. In fact,

0:23:07.440 --> 0:23:10.000
<v Speaker 2>I don't think you're going to see more journalists get

0:23:10.080 --> 0:23:12.879
<v Speaker 2>hired as a result of these deals, which kind of

0:23:12.920 --> 0:23:17.919
<v Speaker 2>makes them a little bit stupid. This puts AI companies

0:23:17.920 --> 0:23:20.359
<v Speaker 2>in a kind of Kafka esque bind, but they can't

0:23:20.359 --> 0:23:23.040
<v Speaker 2>really improve a tool for automating the creation of content

0:23:23.240 --> 0:23:26.639
<v Speaker 2>without human beings creating more content than they've ever created before,

0:23:28.080 --> 0:23:31.359
<v Speaker 2>just as said tool actively crowds out human made data.

0:23:32.320 --> 0:23:36.719
<v Speaker 2>It's a little silly. The solution to these problems, if

0:23:36.760 --> 0:23:39.679
<v Speaker 2>you ask open AI's Sam Altman, is always more money

0:23:39.680 --> 0:23:42.480
<v Speaker 2>and power, which is why the information reports he is

0:23:42.480 --> 0:23:45.600
<v Speaker 2>trying to convince open ai investor Microsoft to build him

0:23:45.800 --> 0:23:49.120
<v Speaker 2>and I'm not kidding, and one hundred billion dollars supercomputer

0:23:49.440 --> 0:23:54.760
<v Speaker 2>called Stargate. This massive series of interconnected machines will require

0:23:55.119 --> 0:23:58.480
<v Speaker 2>entirely new ways to mountain cool processing units and is

0:23:58.640 --> 0:24:03.159
<v Speaker 2>entirely contingent on pen AI's ability to meaningfully improve chet GPT,

0:24:04.080 --> 0:24:08.280
<v Speaker 2>something Sam Mortman claims isn't possible without more computing power.

0:24:09.720 --> 0:24:12.840
<v Speaker 2>To be clear, open Air already failed to build a

0:24:12.880 --> 0:24:20.800
<v Speaker 2>more efficient model, dubbed Arakis, which ended up getting mothballed

0:24:20.800 --> 0:24:24.800
<v Speaker 2>because it wasn't more efficient. It's also important to note

0:24:24.800 --> 0:24:28.560
<v Speaker 2>that every major cloud company now has inextricably tied themselves

0:24:28.720 --> 0:24:32.480
<v Speaker 2>to the generative AI movement. Google and Amazon have invested

0:24:32.560 --> 0:24:36.359
<v Speaker 2>billions into chat GPT competitor Anthropic, and both claim to

0:24:36.359 --> 0:24:40.480
<v Speaker 2>be Anthropic's primary cloud provider, though isn't really obvious which

0:24:40.520 --> 0:24:44.000
<v Speaker 2>one is. In doing so, they've guaranteed, according to a

0:24:44.000 --> 0:24:47.000
<v Speaker 2>source of mind, about seven hundred and fifty million dollars

0:24:47.040 --> 0:24:49.879
<v Speaker 2>a year of revenue for Google's cloud and eight hundred

0:24:49.960 --> 0:24:53.520
<v Speaker 2>million dollars a year of revenue for Amazon Web Services

0:24:53.560 --> 0:24:57.040
<v Speaker 2>the cloud service from Amazon by mandating that Anthropic uses

0:24:57.119 --> 0:25:00.639
<v Speaker 2>their services to power their clawed model. This similar to

0:25:00.680 --> 0:25:04.040
<v Speaker 2>the thirteen billion dollar investment that Microsoft gave open AI

0:25:04.480 --> 0:25:06.840
<v Speaker 2>last year, most of which was made up of credits

0:25:06.840 --> 0:25:09.960
<v Speaker 2>for Microsoft's cy or cloud, And I somehow doubt that

0:25:10.000 --> 0:25:13.280
<v Speaker 2>Microsoft is going to be the noble party that goes

0:25:13.280 --> 0:25:15.240
<v Speaker 2>on the earnings and says, well, we don't want to

0:25:15.320 --> 0:25:17.960
<v Speaker 2>count the credits that we gave open out want to

0:25:17.960 --> 0:25:19.600
<v Speaker 2>be fair. No, they're going to mash that shit right

0:25:19.640 --> 0:25:23.840
<v Speaker 2>back into their revenue. Kind of a con kind of

0:25:23.880 --> 0:25:25.639
<v Speaker 2>makes me angry when I think about it too. Anyway,

0:25:25.720 --> 0:25:27.520
<v Speaker 2>let me just put that aside and not going to

0:25:27.560 --> 0:25:30.959
<v Speaker 2>get pissed off again. Look, I'm surprised more people aren't

0:25:31.359 --> 0:25:35.040
<v Speaker 2>really upset about this very incestuous relationship between big tech

0:25:35.119 --> 0:25:40.280
<v Speaker 2>and this supposedly independent generative AI movement. Microsoft, Google, and

0:25:40.320 --> 0:25:44.359
<v Speaker 2>Amazon have effectively handed cash to one or two companies

0:25:44.800 --> 0:25:46.800
<v Speaker 2>that will eventually hand the cash back to them in

0:25:46.840 --> 0:25:49.240
<v Speaker 2>exchange for cloud services that are necessary to make their

0:25:49.240 --> 0:25:52.560
<v Speaker 2>companies work, and all three big tech firms are spending

0:25:52.640 --> 0:25:56.719
<v Speaker 2>billions to expand their data center operations to capture this

0:25:57.160 --> 0:26:02.440
<v Speaker 2>theoretical demand from generative AI. Every penny the open AI

0:26:02.640 --> 0:26:05.080
<v Speaker 2>or anthropic makes will now flow back to one of

0:26:05.200 --> 0:26:08.280
<v Speaker 2>three big tech firms, even more so in the case

0:26:08.280 --> 0:26:12.080
<v Speaker 2>of open AI, because Microsoft's investment entitles Microsoft to a

0:26:12.119 --> 0:26:18.040
<v Speaker 2>share of any future profits from open ai and chat GBT. Yeah,

0:26:18.080 --> 0:26:20.120
<v Speaker 2>it doesn't even really matter if they make one, because

0:26:20.119 --> 0:26:23.520
<v Speaker 2>Big tech wins either way. Anthropic has to use Google

0:26:23.560 --> 0:26:28.600
<v Speaker 2>Cloud and Amazon Web services, open Ai has to use

0:26:28.800 --> 0:26:32.080
<v Speaker 2>Microsoft's z your Cloud, and Microsoft is actively selling open

0:26:32.119 --> 0:26:35.160
<v Speaker 2>AI's models to their zero cloud customers. And every time

0:26:35.640 --> 0:26:38.560
<v Speaker 2>somebody uses open AI's models, that model is being run

0:26:38.600 --> 0:26:43.080
<v Speaker 2>on a zero cloud, generating revenue for Microsoft. This is

0:26:43.200 --> 0:26:45.760
<v Speaker 2>the rot economy in action, by the way, Big tech

0:26:45.800 --> 0:26:49.879
<v Speaker 2>has funded its biggest customers for their next growth revenue stream,

0:26:49.960 --> 0:26:53.680
<v Speaker 2>justifying this massive expansion of their data center operations because

0:26:53.760 --> 0:26:57.800
<v Speaker 2>ai is the future and they're telegraphing growth to these

0:26:57.840 --> 0:27:00.280
<v Speaker 2>brainless drones in the market who will buy any thing,

0:27:00.280 --> 0:27:04.280
<v Speaker 2>who never think too hard about what they're actually investing in. Hey,

0:27:04.280 --> 0:27:08.199
<v Speaker 2>i's this big, sexy, exciting and theoretically powerful way to

0:27:08.280 --> 0:27:13.479
<v Speaker 2>centralized labor. And it's innovative sounding enough that it allows

0:27:13.480 --> 0:27:16.360
<v Speaker 2>people to dream big about how it might change their lives. Now,

0:27:16.359 --> 0:27:21.600
<v Speaker 2>it might help them not pay real people to do shit. Yeah,

0:27:21.600 --> 0:27:24.320
<v Speaker 2>here's the biggest worry I have. Here's the real pickle,

0:27:24.359 --> 0:27:27.840
<v Speaker 2>here's the thing that keeps me up at night. None

0:27:27.840 --> 0:27:30.480
<v Speaker 2>of these companies seem to have appeared to consider something.

0:27:31.800 --> 0:27:34.119
<v Speaker 2>What if generative AI can't actually do any of the

0:27:34.160 --> 0:27:39.040
<v Speaker 2>things they're excited about. What if genera avii's content, as

0:27:39.040 --> 0:27:42.680
<v Speaker 2>you probably seen from anything chet GBT spits out, isn't

0:27:42.760 --> 0:27:48.840
<v Speaker 2>really good enough. Hey, is anyone checked if anyone's actually

0:27:48.920 --> 0:27:53.080
<v Speaker 2>using these tools, if they're helpful to anyone? Is this

0:27:53.160 --> 0:28:00.120
<v Speaker 2>actually replacing anyone's work? Huh, that's a bit worrying, mate,

0:28:00.680 --> 0:28:03.480
<v Speaker 2>I didn't think about that before. Just kidding, I've been

0:28:03.480 --> 0:28:07.679
<v Speaker 2>thinking about it for months. Look, here's the thing. I

0:28:07.800 --> 0:28:10.919
<v Speaker 2>think that the big problem here is that Sam Altman

0:28:11.000 --> 0:28:13.879
<v Speaker 2>and his cronies have allowed the media, the markets, and

0:28:13.920 --> 0:28:16.720
<v Speaker 2>big tech to fill in the gaps of their specious messaging.

0:28:17.240 --> 0:28:21.080
<v Speaker 2>They've allowed everybody to think that open AI can do

0:28:21.240 --> 0:28:27.920
<v Speaker 2>whatever anyone dreams. Yeah, I don't think that generative AI

0:28:28.000 --> 0:28:30.800
<v Speaker 2>can do much more than it is today. And also,

0:28:31.720 --> 0:28:34.560
<v Speaker 2>from what I've seen, none of these generative AI companies

0:28:34.560 --> 0:28:38.280
<v Speaker 2>actually make a profit, and with each new model they

0:28:38.360 --> 0:28:42.640
<v Speaker 2>become less profitable, and I don't see that changing in

0:28:42.680 --> 0:28:45.960
<v Speaker 2>the future. And so I've dug in a little more

0:28:46.560 --> 0:28:51.160
<v Speaker 2>looking under the hood. All the demand that's spurring Microsoft, Google,

0:28:51.200 --> 0:28:55.360
<v Speaker 2>and Amazon's data center operations might not actually be there.

0:28:57.120 --> 0:28:59.480
<v Speaker 2>My friends, I think we're in the next tech bubble,

0:28:59.680 --> 0:29:02.239
<v Speaker 2>and the next episode, I'm gonna walk you through how

0:29:02.280 --> 0:29:14.480
<v Speaker 2>I think it might pop. Thank you for listening to

0:29:14.480 --> 0:29:15.320
<v Speaker 2>Better Offline.

0:29:15.480 --> 0:29:17.880
<v Speaker 4>The editor and composer of the Better Offline theme song

0:29:17.960 --> 0:29:20.640
<v Speaker 4>is Matasowski. You can check out more of his music

0:29:20.640 --> 0:29:24.280
<v Speaker 4>and audio projects at Matasowski dot com, M A T

0:29:24.280 --> 0:29:28.560
<v Speaker 4>T O, S O W s ki dot com. You

0:29:28.560 --> 0:29:31.080
<v Speaker 4>can email me at easy at Better offline dot com,

0:29:31.160 --> 0:29:33.120
<v Speaker 4>or check out Better Offline dot com to find my

0:29:33.160 --> 0:29:35.800
<v Speaker 4>newsletter and more links to this podcast. Thank you so

0:29:35.880 --> 0:29:36.560
<v Speaker 4>much for listening.

0:29:37.720 --> 0:29:40.200
<v Speaker 2>Better Offline is a production of cool Zone Media.

0:29:40.320 --> 0:29:43.160
<v Speaker 4>For more from cool Zone Media, visit our website cool

0:29:43.240 --> 0:29:46.360
<v Speaker 4>Zonemedia dot com, or check us out on the iHeartRadio

0:29:46.360 --> 0:30:07.200
<v Speaker 3>App, Apple Podcasts, or wherever you get your podcasts.