Latest / Stuff You Should Know / Large Language Models and You

Transcript

Large Language Models and You · Stuff You Should Know

0:01
Speaker 1: Welcome to Stuff You Should Know, a production of iHeartRadio.
0:11
Speaker 2: Hey, and welcome to the podcast. I'm Josh, and there's
0:14
Speaker 2: Chuck and Jerry's here too, And that makes this a
0:18
Speaker 2: timely topical Not timely is in fork times. I meant
0:22
Speaker 2: to say timely topical episode of Stuff you should Know.
0:27
Speaker 2: That's a great forecast of.
0:29
Speaker 1: How it's going to go to I think fork cast.
0:33
Speaker 1: Oh God, is this really us or is it AI
0:39
Speaker 1: generated Josh and Chuck.
0:42
Speaker 2: Until this year I would have been like, don't be preposterous.
0:48
Speaker 2: Now I'm like, just give it some time.
0:51
Speaker 1: You know how we would know as if it said
0:53
Speaker 1: if one of us said, of course it's the real Us.
0:56
Speaker 1: We met in the office and bonded over our Van
0:59
Speaker 1: Halen denim vests.
1:01
Speaker 2: Yeah, we'd be like, sucker, you just fell for the
1:04
Speaker 2: oldest trap in the book, the Sicilian switcher.
1:08
Speaker 1: Yeah, aka fake Wikipedia entry stuff?
1:12
Speaker 2: Is that still up?
1:13
Speaker 1: I haven't been to our Wikipedia page in years, so
1:16
Speaker 1: I don't know.
1:17
Speaker 2: Well, regardless, we're not talking about Wikipedia, although it does
1:20
Speaker 2: kind of fall into the sure it figures in the
1:22
Speaker 2: rubric of this. Not sure if I use that word correctly,
1:25
Speaker 2: but it felt right. We're talking today about what are
1:30
Speaker 2: in the biz known as large language models, but more
1:34
Speaker 2: colloquially known by basically their their public facing names, things
1:38
Speaker 2: like chat, GPT or barred or being AI. But essentially
1:44
Speaker 2: what they all are are algorithms, artificially intelligent algorithms that
1:50
Speaker 2: are trained on text, tons and tons and tons of
1:54
Speaker 2: text written English language stuff, that are so good at
2:00
Speaker 2: recognizing patterns in those things that they can actually simulate
2:04
Speaker 2: a conversation with you, the person on the other side
2:07
Speaker 2: of the computer, asking them questions.
2:10
Speaker 1: Yeah, this is it's gonna be fun doing this episode
2:13
Speaker 1: over every six months, right until we're replaced totally.
2:19
Speaker 2: So I think we should say, though, like this is
2:21
Speaker 2: we're going to like this is such a huge wide
2:24
Speaker 2: topic that is just we're in the ignition phase, like
2:30
Speaker 2: the fuse just caught, right, Yeah, that we're going to
2:34
Speaker 2: really try to keep it narrow just strictly to large
2:36
Speaker 2: language models and the immediate effect they're planning on having
2:39
Speaker 2: or they're going to have, hopefully not planning on anything yet,
2:43
Speaker 2: but I really would like to do one on how
2:46
Speaker 2: to keep AI friendly and keeping it from running away. Yeah,
2:50
Speaker 2: I say, we just kind of avoid that whole kind
2:52
Speaker 2: of stuff. And really I'm talking to myself right now
2:55
Speaker 2: at least for this episode.
2:56
Speaker 1: Okay, Yeah, you know, we're going to kind of explain
2:59
Speaker 1: how these things were and what the initial applications look
3:03
Speaker 1: like and kind of where we are right now, and
3:05
Speaker 1: then what it could mean for like jobs and the
3:08
Speaker 1: economy and stuff like that. But you're right, it is
3:11
Speaker 1: as a whole ball of wax, as you well know.
3:14
Speaker 1: And this is a great time to plug the End
3:16
Speaker 1: of the World with Josh Clark, which is still out there.
3:19
Speaker 1: You can still listen to it.
3:20
Speaker 2: The truth is out there in the form of the
3:23
Speaker 2: End of the World with Josh Clark.
3:24
Speaker 1: Yeah, it's a great ten part series that you did,
3:29
Speaker 1: and AI is among those existential risks existential that you covered.
3:34
Speaker 2: Yeah, it's episode four, I believe. And Chuck like just
3:38
Speaker 2: from having done that research in forming my own opinions
3:40
Speaker 2: over the years about this. Yeah, Like I'm it's staggering
3:46
Speaker 2: to me that we're like we've just entered like what's
3:50
Speaker 2: going to be the most revolutionary transitional phase in the
3:58
Speaker 2: entire history of humanity. You can argue everything else took
4:02
Speaker 2: place over very long periods of time. We started playing
4:05
Speaker 2: with stone tools and then we started building cities. All
4:07
Speaker 2: this stuff took place over thousands and thousands, hundreds of thousands,
4:10
Speaker 2: millions of years. We just entered a period where stuff's
4:13
Speaker 2: going to start happening within weeks pretty soon. As of
4:17
Speaker 2: twenty twenty three, the whole thing just started.
4:19
Speaker 1: Yeah, and none of this was around like this when
4:22
Speaker 1: you did The End of the World and that was
4:24
Speaker 1: what like five years ago.
4:26
Speaker 2: Is Yeah, it was twenty eighteen. All this was being
4:28
Speaker 2: worked on, but we hadn't hit that point. Like all
4:31
Speaker 2: this was pretty much predicted and projected, and it was
4:34
Speaker 2: clear that this was the direction people were going.
4:36
Speaker 1: And it's here, baby, it is.
4:37
Speaker 2: It's nuts, but it's actually here. So we're talking about
4:40
Speaker 2: our large language models, which is a type of neural
4:44
Speaker 2: network that are easiest to think of in terms of
4:48
Speaker 2: like a human brain, where you have neurons that are
4:51
Speaker 2: connected to other neurons, but they're not connected to some
4:54
Speaker 2: other neurons, and all of those neural connections kind of
4:58
Speaker 2: are activated by input. It's that put out something like
5:02
Speaker 2: your conscious experience or you say a sentence or something
5:05
Speaker 2: like that. It's it's very similar in its most basic nature.
5:10
Speaker 1: I guess, yeah, I mean Olivia helped us out with this,
5:13
Speaker 1: and che did a great job. I think and Google
5:17
Speaker 1: themselves basically say, you know what it's really it's sort
5:20
Speaker 1: of like how when you go to search for something
5:22
Speaker 1: on our search engine tool here a weird way I
5:28
Speaker 1: could have said that, I don't think so are handy
5:31
Speaker 1: search for Then you know, basically what we're doing is
5:34
Speaker 1: auto completing, like an analysis of like probability, like like
5:40
Speaker 1: what you're typing. If you type in you know, John Coltrane,
5:44
Speaker 1: or start to chop type in John Cole, that might
5:47
Speaker 1: finish it out as John Coltrane a Love Supreme or
5:51
Speaker 1: John Coltrane jazz and they're saying, you know what, what
5:54
Speaker 1: is happening now with these lms is it's the same thing.
6:00
Speaker 1: It's just it's got way more data, way more calculations
6:03
Speaker 1: in the algorithm. So it's not just completing like a
6:07
Speaker 1: word or two. It's potentially you know, hey, rewrite the
6:11
Speaker 1: Bible or whatever you tell it to do.
6:13
Speaker 2: Yeah, And the big difference is in the amount of
6:16
Speaker 2: info the neural network is capable of taking into consideration. Yeah,
6:23
Speaker 2: so may I for a minute, oh please, So imagine
6:26
Speaker 2: with one of those autocomplete suggestion tools like they have
6:29
Speaker 2: on Google Search. If there's five hundred thousand words in
6:34
Speaker 2: the English language, that means that you have five hundred
6:37
Speaker 2: thousand words that a person could possibly put in. That's
6:40
Speaker 2: the input into the neural network, and then there's five
6:44
Speaker 2: hundred thousand possible words that that network could put out.
6:47
Speaker 2: So you have five hundred thousand connections to five hundred
6:50
Speaker 2: thousand other connections. So it's like, I think two hundred
6:53
Speaker 2: and fifty billion connections you're starting with right there. That's
6:56
Speaker 2: just the autocomplete suggestion because it based on those connections.
7:00
Speaker 2: In studying words in the English language and phrases in
7:03
Speaker 2: the English language, it places emphasis more on some connections
7:07
Speaker 2: than others. So John Coltrane is what's what's his album
7:12
Speaker 2: I Can't Remember is a Love Supreme classic album. So
7:16
Speaker 2: John Coltrane is much more closely related to a Love
7:20
Speaker 2: Supreme in the mind of a neural network than John Coltrane.
7:25
Speaker 2: Charlie Brown disco is right, just to take something off
7:29
Speaker 2: the top of my head. And so based on that
7:32
Speaker 2: analysis and that weight that it gives to some things
7:35
Speaker 2: other than others, it suggests those words what the large
7:38
Speaker 2: language models that like chat GPT, that we're seeing today.
7:42
Speaker 2: They do the same thing. They have all those same connections,
7:45
Speaker 2: but the analysis they do, the weight that they put
7:48
Speaker 2: on the connections is so much more advanced, yes, and
7:52
Speaker 2: exponential that it's it's actually not just capable of suggesting
7:57
Speaker 2: the next word, it's capable of holding a conversation with you.
8:00
Speaker 2: That's how much it understands how the English language works.
8:04
Speaker 1: Yeah, Like if you said, you know, write a story
8:08
Speaker 1: about wintertime, and you know, it got to the word snowy,
8:14
Speaker 1: it would it would go through you know, I mean,
8:17
Speaker 1: and this is like instantaneously it's doing these calculations. It
8:20
Speaker 1: might say like you know, oh, hillside or winter or
8:24
Speaker 1: snowy day, like these are all things that make sense
8:26
Speaker 1: because I've learned that that makes sense I being you know,
8:29
Speaker 1: the chatbot or whatever. But it probably won't be snowy
8:33
Speaker 1: chicken wing because that doesn't seem to fit the algorithm.
8:38
Speaker 1: And it learns all this stuff by reading the internet.
8:43
Speaker 1: And you know, put a pin in that, because that's
8:46
Speaker 1: pretty thorny for a whole lot of reasons, but not
8:51
Speaker 1: the least of which is the fact that some companies,
8:53
Speaker 1: and again we'll get to it, are starting to say
8:55
Speaker 1: like wait a minute, Like, we created this content and
8:57
Speaker 1: now you're just scrubbing it and then using it and
9:02
Speaker 1: charging people to use it, and we're not getting a
9:05
Speaker 1: piece of it. So that's just one tiny little thorn.
9:08
Speaker 1: But in order to do this, like you said, it's
9:10
Speaker 1: like it needs to know more, and you came up
9:13
Speaker 1: with a great example, like the word lemon. In a
9:17
Speaker 1: very basic way, it might understand that a lemon is
9:22
Speaker 1: roundish and sour and yellow. But if it needs to
9:26
Speaker 1: get smart enough to really write as if it were
9:29
Speaker 1: a human, it needs to know that it can make lemonade,
9:31
Speaker 1: and that it grows on a tree in these agricultural zones,
9:35
Speaker 1: and that it's a citrus fruit, because it has to
9:38
Speaker 1: be able to group lemon together with like things. And
9:43
Speaker 1: those groups are either like, you know, hey, it's super
9:46
Speaker 1: similar to this, like maybe other citrus fruits, or it's
9:50
Speaker 1: you know, sort of similar to this but not as
9:51
Speaker 1: similar as citrus fruits, like desserts. And then you get
9:55
Speaker 1: to chicken wings, although actually that's not true because lemon
9:58
Speaker 1: chicken wings.
10:00
Speaker 2: You could have lemon pepper chicken wings.
10:01
Speaker 1: Right, Yeah, that's what I'm saying, So yeah, Kau, But
10:04
Speaker 1: the instance you use is like greenland, which I guess
10:06
Speaker 1: doesn't grow lemons.
10:08
Speaker 2: No, but I mean, I'm sure they import lemons, so
10:10
Speaker 2: there's some connection there. But based on how connected, how
10:14
Speaker 2: often these words show up together, and the billions and
10:17
Speaker 2: billions of lines of texts that these language large language
10:21
Speaker 2: models are trained on, it starts to get more and
10:25
Speaker 2: more dimensions and making more and more connections. Right. So
10:30
Speaker 2: as that happens, words start to cluster together, like lemon
10:34
Speaker 2: and pie and ice box all kind of cluster together.
10:37
Speaker 2: And by taking words and understanding how they connect to
10:42
Speaker 2: other words, you can take the English language, just the
10:46
Speaker 2: words of the English language, and make meaning out of it.
10:48
Speaker 2: That's what that's all we do. And large language models
10:51
Speaker 2: are capable of doing the same thing. But it's really
10:55
Speaker 2: really important for you to understand that the large language
10:59
Speaker 2: model doesn't understand what it's doing. It doesn't have any
11:04
Speaker 2: meaning to the word lemon whatsoever. All of these dimensions
11:10
Speaker 2: that it waits to decide whether what word it should
11:13
Speaker 2: use next, they're called embeddings. They're just numerical representations. Yeah,
11:18
Speaker 2: So the higher the number, the likelier it is it
11:21
Speaker 2: goes with the word that the user just put in
11:23
Speaker 2: or that the large language model. Just use the lower
11:26
Speaker 2: the number, the further away it is in the cluster, right,
11:30
Speaker 2: it doesn't understand what it's saying to you. And as
11:33
Speaker 2: we'll see later, that accounts for a phenomenon that we're
11:36
Speaker 2: going to have to overcome for them to get smarter,
11:38
Speaker 2: which is called hallucinations. But that's a really critically important
11:42
Speaker 2: thing to remember.
11:43
Speaker 1: Yeah, Another critically important thing to remember is and you
11:48
Speaker 1: probably get this from what we said so far if
11:49
Speaker 1: you already know about a little bit about it, But
11:51
Speaker 1: there's no programmer that's teaching these things and typing in
11:56
Speaker 1: inputs and then saying here's how you learn things. Like
12:00
Speaker 1: it's doing this on its own, and it's learning things
12:03
Speaker 1: on its own, and what we're talking about eventually like that.
12:07
Speaker 1: You know, where it could get super scary is when
12:09
Speaker 1: it gets to what's called emergent abilities, where it's so
12:12
Speaker 1: powerful and there's so much data that the nuance that's
12:18
Speaker 1: missing now will be there right exactly.
12:21
Speaker 2: So, yeah, that's when things are going to get even
12:25
Speaker 2: harder to understand, you know, to remind yourself that you're
12:29
Speaker 2: talking to a machine, you know.
12:31
Speaker 1: Yeah, And the other thing too, though, even though I
12:32
Speaker 1: said humans aren't in putting this data. One of the
12:36
Speaker 1: big things that is allowing this stuff to get smarter
12:40
Speaker 1: is human feedback. It's called r l HF, which is
12:45
Speaker 1: reinforcement learning on human feedback. So at the end of
12:49
Speaker 1: your whatever you've told it to create, you can go
12:53
Speaker 1: back in and say, well, you got this wrong and
12:55
Speaker 1: this wrong, this is what that really is, and it says,
12:58
Speaker 1: thank you, I have now just gotten smart.
13:01
Speaker 2: Exactly. So one of the reasons why these things are
13:04
Speaker 2: suddenly just so smart and can say thank you, I've
13:07
Speaker 2: just gotten so much smarter is because of a paper
13:10
Speaker 2: that Google engineers published openly in twenty seventeen describing what's
13:16
Speaker 2: now like the essential ingredient for a large language model
13:20
Speaker 2: or probably any neural network from now on. It's called
13:22
Speaker 2: a transformer. And rather than analyzing each bit of text,
13:27
Speaker 2: let's say you say one of the very famous things.
13:30
Speaker 2: Marvin Minsky was one of the founders of the field
13:33
Speaker 2: of AI, and his son Henry prompted chet Gpt to
13:38
Speaker 2: describe what losing a sock in the dryer is like
13:41
Speaker 2: in the style of the Declaration of Independence. Right, So
13:45
Speaker 2: depending on how Henry Minsky type that in before transformers
13:51
Speaker 2: the neural network would analyze each word and do it
13:55
Speaker 2: one increment and a time, maybe not even words, sometimes
13:57
Speaker 2: strings of just letters together, phone names even if you
14:02
Speaker 2: can believe it, phone names even And what the transformer
14:07
Speaker 2: does is it changes that. It allows it to analyze
14:09
Speaker 2: everything all at once. So it's so much faster, not
14:13
Speaker 2: just in putting out a coherent answer to your question
14:16
Speaker 2: or request, but in also training itself on that text.
14:20
Speaker 2: So you just feed it the internet and it starts
14:23
Speaker 2: analyzing it and self correcting. It trains itself, It learns
14:27
Speaker 2: on its own, and that unfortunately also makes AI, including
14:32
Speaker 2: large language models what are known as black boxes. Yeah,
14:35
Speaker 2: we don't know how they're doing what they're doing. We
14:38
Speaker 2: have a good idea how to make them do the
14:40
Speaker 2: things we want, but the in between stuff, we cannot
14:43
Speaker 2: one hundred percent say what they're doing. How they come
14:47
Speaker 2: up with these conclusions, which also explains hallucinations in them
14:51
Speaker 2: not really making sense to us.
14:53
Speaker 1: Yeah, and you know, the T in GPT stands for transformer.
14:57
Speaker 1: It's generative pre trained transformer. And the reason they call
15:02
Speaker 1: it GPT for short is because if they call it
15:05
Speaker 1: generative pre trained transformer, everybody would be scared out of
15:10
Speaker 1: their minds.
15:10
Speaker 2: We just start running around to nowhere in particular.
15:14
Speaker 1: Yeah, should we take a break, I say, we do.
15:17
Speaker 2: I think that we kind of explained that fairly well.
15:20
Speaker 1: Yeah, fairly robust beginning, my friend. All right, So open
15:50
Speaker 1: Ai launched their chat GPT and very recently in November
15:54
Speaker 1: of twenty twenty two, and just in that brief window
15:59
Speaker 1: was like six or eight months ago. Things are kind
16:02
Speaker 1: of flying high, and all kinds of companies are launching
16:06
Speaker 1: their own stuff. Some of it is well. First of all,
16:10
Speaker 1: open ai is now at chat GPT four, yes, and
16:14
Speaker 1: I'm sure you know more will be coming in quick succession.
16:18
Speaker 1: But companies are launching, and we're going to talk about
16:20
Speaker 1: all of them kind of like broad stuff like chat,
16:23
Speaker 1: BT GPT and really specific stuff like well, hey, I'm
16:28
Speaker 1: in the banking business. Can we just design something for
16:31
Speaker 1: banking or just something for real estate? So they're also
16:35
Speaker 1: getting specific on a smaller level in addition to these
16:38
Speaker 1: large like Google and Microsoft and Being and all that stuff.
16:41
Speaker 2: Yeah, And to get specific, all you have to do
16:44
Speaker 2: is take an existing GPT, a large language model, and
16:50
Speaker 2: add some software that helps guide it a little more. Yeah,
16:54
Speaker 2: and there you go or just train it on specific
16:57
Speaker 2: stuff like medical notes. That's another one one of the
17:01
Speaker 2: other things that's that's changed very quickly between November of
17:05
Speaker 2: twenty twenty two in March of twenty twenty three, when
17:08
Speaker 2: I think GPT four became available. Just think about that.
17:13
Speaker 2: That's that's such a short amount of time. Yeah, all
17:16
Speaker 2: of a sudden. Now you can take a picture and
17:19
Speaker 2: feed it into a large language model and it will
17:22
Speaker 2: describe the picture. It will look at the picture essentially
17:26
Speaker 2: and describe what's going on. There's a there's a demonstration
17:31
Speaker 2: from one of the guys from open Ai who doodles
17:34
Speaker 2: like on a little like scrapbook piece of paper some
17:37
Speaker 2: ideas for a website. He takes a picture of that
17:41
Speaker 2: paper that he's written on, feeds it into chat GPT four,
17:46
Speaker 2: and it builds a website for him in a couple
17:49
Speaker 2: of minutes. That functions the way he was thinking of
17:51
Speaker 2: the on the doodle scratch pad.
17:54
Speaker 1: I wonder if the only way to slow this stuff
17:55
Speaker 1: down is to literally slow down the Internet again. Go
17:59
Speaker 1: back to like the old days when a picture would
18:02
Speaker 1: load like three lines at a time, right, and they
18:05
Speaker 1: describe a picture will be like someone's hair, someone's nose,
18:10
Speaker 1: someone's gin. Don't forget between Yeah, an hour later you
18:14
Speaker 1: have a complete picture.
18:15
Speaker 2: Right. I don't think there's any way to slow this
18:17
Speaker 2: down because we're in not to be alarmist, but we're
18:21
Speaker 2: in a second worst case scenario for introducing AI to
18:26
Speaker 2: the world, which is, rather than state actors doing this,
18:30
Speaker 2: which would be really bad, we have private companies doing it,
18:34
Speaker 2: which is just slightly less bad. But they're competing in
18:38
Speaker 2: an arms race to get the best, brightest, smartest AI
18:42
Speaker 2: out there as fast as they can, and they're not
18:45
Speaker 2: taking into account like all of the downsides to it.
18:48
Speaker 2: They're just throwing it out there as much as they can.
18:51
Speaker 2: Because one of the ways that these things get smarter
18:53
Speaker 2: is by interacting with the public. They get better and
18:56
Speaker 2: better at what they do from getting feedback from people
19:01
Speaker 2: using them.
19:02
Speaker 1: Yeah, even if it's for just some goofy fun thing
19:05
Speaker 1: you're doing, it's learning from that. And you talked about
19:08
Speaker 1: the advancements made between the launch of three point five
19:12
Speaker 1: and GPT four, and three point five scored in the
19:17
Speaker 1: tenth percentile when it took the uniform bar exam, and
19:22
Speaker 1: four has already scored in the ninetieth percentile, and they
19:26
Speaker 1: found that chet GPT four is really it's great at
19:29
Speaker 1: taking tests, and it's scoring really well on tests, particularly
19:34
Speaker 1: you know, standardized tests. All. I think it basically aced
19:38
Speaker 1: all of the AP tests that you would take to
19:41
Speaker 1: get into AP classes except well, it took a couple
19:45
Speaker 1: of AP class but the max score is five, and
19:49
Speaker 1: I think it got five's kind of on everything except
19:52
Speaker 1: for math. It got a four and math it's kind
19:59
Speaker 1: of it's we it's kind of weirdly counterintuitive because it's
20:02
Speaker 1: a number space thing, but it has more trouble with math,
20:06
Speaker 1: like rudimentary math, than it does with like constructing a
20:10
Speaker 1: paragraph on you know, Shakespeare or something, or as Shakespeare
20:14
Speaker 1: does better with like math word problems and more advanced
20:18
Speaker 1: math than it does just at basic math apparently, or.
20:20
Speaker 2: Like describing how a formula functions using you know, writing.
20:26
Speaker 2: The thing is though, and this is another great example
20:28
Speaker 2: of how fast this is moving. They've already figured out
20:31
Speaker 2: that all you have to do is do what's called
20:33
Speaker 2: prompting where you where you basically take the answer that
20:38
Speaker 2: the the incorrect answer that the large language model gives you,
20:43
Speaker 2: and then basically re explain it by breaking it down
20:45
Speaker 2: into different parts, and it learns as you're doing that,
20:48
Speaker 2: and then all of a sudden it comes up with
20:50
Speaker 2: it gets better at math. So they've figured out tools,
20:53
Speaker 2: extra software you can lay over a GPT that basically
20:58
Speaker 2: teach it to do math or prompt it in the
21:01
Speaker 2: correct way so that you get the answer you're looking
21:03
Speaker 2: for that's based on math.
21:05
Speaker 1: Yeah. I mean, every time I read something that said, well,
21:08
Speaker 1: right now, it's not so great at this, I just
21:11
Speaker 1: assume that meant and we'll have that worked out in
21:14
Speaker 1: the next few weeks.
21:15
Speaker 2: Yeah, pretty much. I mean, because as these things get
21:18
Speaker 2: like bigger and smarter, in the data sets that they're
21:21
Speaker 2: trained on get wider, they're just going to get better
21:24
Speaker 2: and better at this because they again they learned from
21:26
Speaker 2: their mistakes.
21:29
Speaker 1: Yeah, just like humans, right, exactly like humans. So you
21:34
Speaker 1: mentioned these hallucinations kind of briefly, and this this is
21:38
Speaker 1: one of the big problems with them so far that
21:40
Speaker 1: again I'm sure they will figure this out in due time.
21:43
Speaker 1: But one example that Livia found was to prompt it
21:47
Speaker 1: with what mammal lays the largest eggs. And one of
21:51
Speaker 1: the problems is when it gives hallucinations or wrong answers it.
21:55
Speaker 1: You know, it's not saying like, well, I'm not so
21:57
Speaker 1: sure about this. It's saying this is true, just like
22:00
Speaker 1: anything else. I'm spitting out right with a lot of confidence.
22:03
Speaker 1: So the answer there was the mammal that lays the
22:06
Speaker 1: largest eggs is the elephant, and elephant's eggs are so
22:08
Speaker 1: small that they are often visible to the naked eye,
22:11
Speaker 1: so they're not commonly known to lay eggs at all. However,
22:14
Speaker 1: in terms of sheer size, and elephant's eggs are the
22:17
Speaker 1: largest of any mammal.
22:18
Speaker 2: Which makes sense in a really weird way if you
22:20
Speaker 2: think about it.
22:21
Speaker 1: Sure, those little invisible eggs.
22:23
Speaker 2: Yeah, because mammals don't lay eggs obviously, But the way
22:26
Speaker 2: that it put it was if you didn't know that
22:28
Speaker 2: mammals don't lay eggs, or you didn't know anything about elephants,
22:31
Speaker 2: you'd be like, oh, that's interesting, and take that as
22:33
Speaker 2: a fact, because it's saying this confidently. And I saw
22:36
Speaker 2: written somewhere that one GPT actually argued with the user
22:40
Speaker 2: and told them they were wrong when they told the
22:41
Speaker 2: GPRIT that it was wrong, Yeah, which is not a
22:46
Speaker 2: behavior you want at all. But that's what's termed as
22:50
Speaker 2: a hallucination, and a hallucination is a good way to
22:53
Speaker 2: understand it. That I saw is that again, this GPT,
23:00
Speaker 2: this large language model, doesn't have any idea what it's
23:04
Speaker 2: saying means. It's just picked up it's noticed patterns that
23:09
Speaker 2: we've not noticed before, and it's putting them together in
23:13
Speaker 2: nonsensical ways. But they're still sensible if you read them.
23:17
Speaker 2: It's just factually they're not sensible because it doesn't have
23:20
Speaker 2: any fact checking necessarily, it just knows what it's finding
23:25
Speaker 2: kind of correlates with other things. So there's some sensible
23:28
Speaker 2: stuff in there, like the phrase invisible to the naked eye,
23:32
Speaker 2: or laying eggs or elephants in mammals. Like this stuff
23:36
Speaker 2: all makes sense. It's not like these are just strings
23:38
Speaker 2: of letters. Yeah, it's just putting them together in ways
23:42
Speaker 2: that are not true. They're factually incorrect, and that's a hallucination.
23:46
Speaker 2: It's not like the computer is thinking that this is true.
23:50
Speaker 2: It doesn't understand things like truth in falsehood. It just creates,
23:58
Speaker 2: and some of the time it gets it really wrong.
24:00
Speaker 1: Yeah, I didn't know what an elephant is.
24:03
Speaker 2: No, it just knows that it correlates to in some
24:06
Speaker 2: really small way. That we've never noticed before the word eggs.
24:11
Speaker 1: Yeah, and this is uh that that's a problem if
24:15
Speaker 1: if it's just like, oh, well, this thing isn't quite
24:17
Speaker 1: where it needs to be at because it thinks elephants
24:19
Speaker 1: lay eggs. But there have already been plenty of real
24:22
Speaker 1: world examples where people are using this and it's screwing
24:25
Speaker 1: things up for their business or for commerce or something
24:28
Speaker 1: where the yeah, or their client, well, that's that's one.
24:31
Speaker 1: There was an attorney who was representing a passenger who
24:35
Speaker 1: was suing an airline and used chat bt to do research,
24:40
Speaker 1: and it came up with a bunch of fake cases
24:42
Speaker 1: that this attorney didn't bother to fact check, I guess.
24:47
Speaker 1: And there were like a dozen fake cases that this
24:50
Speaker 1: attorney submitted in his brief.
24:52
Speaker 2: And it wasn't like so like from what I understand,
24:54
Speaker 2: like the the brief was largely compiled from what the
24:57
Speaker 2: GPT spit out. The It wasn't like the GPT just
25:01
Speaker 2: made up the names of cases. It made up the
25:03
Speaker 2: names of cases and then described the background of the
25:06
Speaker 2: case and how they related to the case at hand, right,
25:09
Speaker 2: So it just completely made these up out of out
25:12
Speaker 2: of the blue. And yeah, that lawyer had no idea.
25:15
Speaker 2: He said in a brief later that he had no
25:17
Speaker 2: idea that this thing was capable of being incorrect. So
25:20
Speaker 2: it was like one of the first times he used it,
25:22
Speaker 2: and he threw himself on the mercy of the court.
25:24
Speaker 2: And I'm not quite sure exactly what happened. I think
25:26
Speaker 2: they're still figuring out what to do about it.
25:29
Speaker 1: Maybe just go spend some quality time with your little
25:31
Speaker 1: chat butt exactly.
25:34
Speaker 2: Similarly, Meta had a large language model that basically got
25:39
Speaker 2: laughed off of the Internet because it was very science
25:42
Speaker 2: focused and it would make up things that just didn't exist,
25:45
Speaker 2: like mathematical formula, like there was one that called the
25:50
Speaker 2: yoko or no, the lenin Ono correlation or something like
25:54
Speaker 2: that completely made up this thing that I read, and
25:57
Speaker 2: I was like, oh, that's interesting. I had no idea.
26:00
Speaker 2: I have never heard this stuff before, and I would
26:02
Speaker 2: have just thought that it was real had I not
26:04
Speaker 2: realized and known ahead of time that it was a
26:07
Speaker 2: hallucination that this math thing does not exist anywhere. And
26:10
Speaker 2: it even attributed it to a live mathematician said that
26:15
Speaker 2: this was the guy who discovered it. So like, it
26:18
Speaker 2: really can get hard to discern what's true and what's not,
26:22
Speaker 2: which again is a really big problem if we haven't
26:25
Speaker 2: gotten that cross yet.
26:26
Speaker 1: Did they say that mathematician's name was math be calculus. Yeah.
26:33
Speaker 1: Another example, and this is, you know, we're going to
26:35
Speaker 1: talk a little bit about you know, replacing jobs in
26:38
Speaker 1: the various ways that can and already is happening. But
26:42
Speaker 1: seeing that, for instance, said oh, you know what, let
26:46
Speaker 1: me try this thing out and see if we can
26:47
Speaker 1: get it to write an actual story. And so they
26:51
Speaker 1: got an AI tool to write one on what is
26:54
Speaker 1: compound interest, and it was just there was a lot
26:57
Speaker 1: of stuff wrong in it. There was some plagiarism, you know,
27:01
Speaker 1: directly lifted. So there's you know, these things aren't fool
27:04
Speaker 1: proof yet, and it's definitely not something that should be
27:08
Speaker 1: utilized for like a public facing website that's supposed to
27:12
Speaker 1: have like really solid vetted articles about uh well, especially
27:17
Speaker 1: seen it about a tech right of all things.
27:21
Speaker 2: That's something that the National Eating Disorder Association found out
27:24
Speaker 2: the hard way. They apparently replaced entirely it's human staffed
27:30
Speaker 2: hotline with the chatbot, and supposedly they were accused of
27:34
Speaker 2: doing this to bust the union that had formed there
27:38
Speaker 2: and so when they released the chatpot into the world
27:41
Speaker 2: and it started offering advice to people suffering from eating disorders.
27:44
Speaker 2: It gave standard, you know, weight loss advice, which you
27:49
Speaker 2: probably get from your doctor who didn't realize you had
27:51
Speaker 2: an eating disorder, but in the context of an eating disorder,
27:54
Speaker 2: it was all like trigger, trigger, trigger, one right after
27:57
Speaker 2: the other. Right, Like, it was telling these people with
28:00
Speaker 2: eating disorders to like, weigh yourself every week and try
28:03
Speaker 2: to cut out five hundred to one thousand calories a
28:05
Speaker 2: day and you'll lose some weight, and just stuff that
28:08
Speaker 2: that would set everybody off. And very quickly they took
28:11
Speaker 2: it offline and I guess brought their humans back, hopefully
28:15
Speaker 2: at double the pay.
28:16
Speaker 1: Yeah. But I mean this stuff is that's already being
28:19
Speaker 1: solved as well, because they point out that GPT four
28:23
Speaker 1: has already scored forty percent higher than three point five,
28:27
Speaker 1: again just a handful of months ago on these accuracy tests,
28:32
Speaker 1: so that that is even getting better. And you know,
28:36
Speaker 1: where where I guess people want it to get to
28:39
Speaker 1: is to the point where it doesn't need human supervision
28:42
Speaker 1: to spit out really really accurate stuff exactly.
28:45
Speaker 2: That's pretty much where they're hoping to get it. And
28:48
Speaker 2: I mean it's just they have the model, they have
28:51
Speaker 2: everything they need. They just it just has to be
28:53
Speaker 2: tinkered with.
28:54
Speaker 1: Now, should we take another break? I think so, all right,
28:58
Speaker 1: we'll take another break and then get into sort of
29:00
Speaker 1: the the economics of it and whether or not your
29:04
Speaker 1: job may be at risk right after this.
29:29
Speaker 2: So one of the astounding things about this that it
29:33
Speaker 2: really caught everybody off guard is that these large language models,
29:37
Speaker 2: the jobs they're coming after are white collar knowledge jobs.
29:42
Speaker 2: They're so good at things like writing, they're good at researching,
29:45
Speaker 2: they're good at analyzing photos. Now and that's a huge
29:50
Speaker 2: sea change from what it's been like traditionally. Right wherever
29:54
Speaker 2: whenever we've automated things, it's usually replaced manual labor. Now
29:57
Speaker 2: it's the manual labor that's safe in this Yeah, this
30:00
Speaker 2: generation of automation, it's the white collar knowledge shops that
30:05
Speaker 2: are at risk. And not just white collar job but
30:08
Speaker 2: artists in yeah, like just who have nothing to do
30:12
Speaker 2: with white collar or jobs, they're at risk as well.
30:17
Speaker 1: Yeah, I'm sure the farmers are all sitting around going,
30:21
Speaker 1: how's that going for you?
30:22
Speaker 2: Yeah, how's that taste? Uh?
30:24
Speaker 1: So? Yeah, art art is when when d A L L. E.
30:28
Speaker 1: Doll E came out, that was an art tool where
30:31
Speaker 1: a lot of people, a lot of people I know,
30:33
Speaker 1: would input. I guess I never did it. I never
30:36
Speaker 1: do anything like that, not because I'm afraid or anything,
30:39
Speaker 1: but I just just not interested basically. But I guess
30:44
Speaker 1: you would submit, like a photograph of yourself and then
30:46
Speaker 1: it would say, well, here's you as a superhero or
30:48
Speaker 1: here's you as a Renaissance painting or whatever. And you know,
30:52
Speaker 1: it's sourcing images from real artists throughout history, from Getty
30:57
Speaker 1: Images and places like that, and there are are ready
31:00
Speaker 1: artists that are suing for infringement. Getty Images is suing
31:04
Speaker 1: for infringement and saying you can't even if you're mixing
31:09
Speaker 1: up things and it's not like a Rembrandt. Let's say
31:13
Speaker 1: you're using all of the artists from that era and
31:16
Speaker 1: mashing it up together in a way that, like we
31:19
Speaker 1: think basically is illegal.
31:21
Speaker 2: Yeah, they say this doesn't count as transformative use, which
31:25
Speaker 2: is typically protected under the law. Right, this is instead
31:31
Speaker 2: just some sort of mash up that that a machine
31:34
Speaker 2: is doing. It's to me, it's almost splitting hairs. But
31:39
Speaker 2: I also very much get where they're coming from not
31:41
Speaker 2: just a place of panic, but like they're a real
31:46
Speaker 2: like they have a basis in fact that these things
31:48
Speaker 2: are not transforming because they don't understand what they're doing.
31:52
Speaker 1: Yeah, and companies are taking notice very quickly. There are
31:56
Speaker 1: some companies I'm sure everyone's going to kind of fall
31:59
Speaker 1: in line, that are already saying, well, no, you got
32:01
Speaker 1: to start paying us for access to this stuff. We
32:05
Speaker 1: paid human beings to create this content for lack of
32:09
Speaker 1: a better word, and put it online for people to access.
32:13
Speaker 1: But you can't come in here now and access it
32:15
Speaker 1: with a bot and use it and charge for it
32:19
Speaker 1: without giving us a little juice. And there are a
32:21
Speaker 1: lot of companies that are already saying like, you can't
32:25
Speaker 1: use this. If you're an employee of our company, you
32:27
Speaker 1: can't use chatbots at all because some of our company's
32:32
Speaker 1: secrets might end up being spilled somehow, or you know,
32:37
Speaker 1: our databases are all of a sudden exposed. So companies
32:40
Speaker 1: are really moving fast to trying to protect their ip
32:44
Speaker 1: I guess.
32:44
Speaker 2: Well, yeah, and one of the I mean some of
32:48
Speaker 2: the companies that are behind the GPTs that are out
32:51
Speaker 2: right now, the large language models that are out right
32:54
Speaker 2: now are well known for not only not protecting their
32:58
Speaker 2: users information, but for rating it for its own use. Like,
33:03
Speaker 2: for example, Meta is one of the ones with They
33:06
Speaker 2: have their large language models called Lama, and there's a
33:11
Speaker 2: chatbot called Alpaca. And it makes total sense that you
33:15
Speaker 2: are probably signing away your right to protect your information
33:19
Speaker 2: when you use those things on whatever computer you're using
33:21
Speaker 2: it on or whatever network you're using it on. I
33:24
Speaker 2: don't understand exactly. I haven't seen anything that says this
33:27
Speaker 2: is how they're doing it, or even that they are
33:29
Speaker 2: definitely doing this. I think it's just that the powers
33:32
Speaker 2: that be no, like they would totally do this if
33:35
Speaker 2: they can, and they probably are, so we should just
33:37
Speaker 2: stay keep our employees away from it, you know, as
33:40
Speaker 2: much as we can.
33:42
Speaker 1: Yeah, it's like we said, it's being used on smaller
33:46
Speaker 1: levels by One of the uses that Livia dug up
33:50
Speaker 1: was like, let's say a real estate agent, instead of
33:53
Speaker 1: taking time to write up listings, has a chatbot to it,
33:57
Speaker 1: and then they can go through afterward and make adjustments
33:59
Speaker 1: to it as needed.
34:01
Speaker 2: Well, in exchange, that database now knows exactly what you
34:05
Speaker 2: think of that one ugly bathroom.
34:07
Speaker 1: That's right, or doctors may be using it to compile
34:12
Speaker 1: lists of possible diseases or conditions that someone might have
34:17
Speaker 1: based on symptoms. These all sound like uses that are like, hey,
34:23
Speaker 1: this sounds like it could be a good thing in
34:25
Speaker 1: some ways, and it can be in some ways. But
34:28
Speaker 1: it's the wild West right now, so it's not like
34:31
Speaker 1: there's any there's anyone saying, well, you can't use it
34:35
Speaker 1: for that, you can only use it for this, you
34:37
Speaker 1: know what I'm saying.
34:37
Speaker 2: Plus, also, everything that we've come up with as just
34:41
Speaker 2: Internet users in the general public has been what we
34:44
Speaker 2: could come up with in given three months, with no
34:48
Speaker 2: warning that we should start thinking about this. It's just like, hey,
34:51
Speaker 2: this is here, what are you going to do with it?
34:53
Speaker 2: And people are just finding new things to do with
34:55
Speaker 2: it every day, And yeah, some of them are benign,
34:58
Speaker 2: like having a draft a blog post for your business.
35:03
Speaker 2: I thought they were already doing that based on some
35:05
Speaker 2: of the emails that I get from like businesses, right, Yeah,
35:08
Speaker 2: but they definitely are now if they weren't before. And
35:11
Speaker 2: that's totally cool because there's there's a it's just taking
35:16
Speaker 2: some of the weight off of the humans that are
35:19
Speaker 2: already doing this. Work. Right, what's going to be problematic
35:24
Speaker 2: is when it comes for the full job, or enough
35:28
Speaker 2: of the job that the company can transfer whatever's left
35:32
Speaker 2: of that person's job to other people and make them
35:35
Speaker 2: just work a little harder while they're supported by the AI.
35:39
Speaker 1: Yeah, here's some stats that they were pretty shocking to me.
35:43
Speaker 1: I didn't know it was moving this fast. But there's
35:47
Speaker 1: a networking app called Fishbowl and in twenty twenty three,
35:50
Speaker 1: just earlier this year, they found that forty percent of
35:55
Speaker 1: what they call working professionals are already using some kind
35:58
Speaker 1: of either chat, GPT or some kind of AI tool
36:01
Speaker 1: while they work, whether it's generating idealists or brainstorming lists,
36:07
Speaker 1: or actually writing stuff or maybe looking at code. And
36:12
Speaker 1: this is the troubling part. Forty of those forty percent,
36:16
Speaker 1: almost seventy percent are doing that in secret and hadn't
36:19
Speaker 1: told their bosses that they were doing that.
36:21
Speaker 2: Right, those are just working professionals. We haven't even started
36:23
Speaker 2: talking about students yet.
36:25
Speaker 1: Yeah. I mean you combine that with work from home,
36:27
Speaker 1: you got a real racket going on.
36:29
Speaker 2: For sure, you know. Yeah, no, totally again though, I mean,
36:33
Speaker 2: like if you can use it to do good work,
36:35
Speaker 2: and you can now do more work. I think you
36:38
Speaker 2: should be paid for more work. Like if your productivity's
36:40
Speaker 2: gone through the roof, great, you figured it out. I've
36:44
Speaker 2: got no problem with that. It's the opposite that I
36:46
Speaker 2: have the problem with.
36:48
Speaker 1: Well, let's skip students for a second and then and
36:51
Speaker 1: talk about that since you brought it up, Because here's
36:54
Speaker 1: the thing this is the United States doesn't have a
36:59
Speaker 1: G eight track record of ignoring the bottom line in
37:04
Speaker 1: favor of just keeping hardworking humans at their jobs. So
37:10
Speaker 1: I think it was a Goldman. Sachs said that they
37:13
Speaker 1: found that there could actually be an increase in the
37:17
Speaker 1: annual GDP by about seven percent over ten years because
37:21
Speaker 1: productivity increases. And I guess the idea is that productivity
37:26
Speaker 1: is increasing because let's say you've got twenty to thirty
37:29
Speaker 1: percent of stuff being done by AI. That opens up
37:32
Speaker 1: twenty to thirty percent of your time for your employees
37:36
Speaker 1: to maybe innovate or you know, get do other capitalistic things.
37:42
Speaker 1: But what it to me, and this is just my opinion,
37:45
Speaker 1: and again we're really early in all this, but it's
37:48
Speaker 1: a bottom line world, and especially a bottom line country
37:51
Speaker 1: that we live in, and I imagine what it would
37:55
Speaker 1: likely mean is by by jobs more than it means, well, hey,
37:59
Speaker 1: you've got more time, and why don't you innovate it
38:01
Speaker 1: your job, because for most jobs, it'll probably be like, oh,
38:06
Speaker 1: wait a minute, if we can teach it to do
38:07
Speaker 1: forty center your job, I bet we could train it
38:10
Speaker 1: to do a one hundred percent.
38:12
Speaker 2: Yeah, or we can get rid of, you know, a
38:15
Speaker 2: bunch of you and just keep some of you to
38:17
Speaker 2: do the other sixty percent.
38:19
Speaker 1: You know. But now see these people are out of
38:20
Speaker 1: jobs that it's going to bite them in the rear though,
38:22
Speaker 1: because it's not ultimately going to be well, who knows.
38:26
Speaker 1: It doesn't seem like it could be good for the
38:28
Speaker 1: overall economy if all of a sudden, all these people
38:30
Speaker 1: are out of jobs. Because people being out of jobs
38:32
Speaker 1: mean they're not that means the economy is going to tank.
38:35
Speaker 1: They're not spending. And it's not like a situation where
38:40
Speaker 1: you know, the tractor replaced the plow and then the
38:45
Speaker 1: robot tractor replaced the tractor. But hey, now we've got
38:48
Speaker 1: these better jobs where you're designing and building these robot
38:51
Speaker 1: tractors and they're higher paying and they're great. It's not
38:55
Speaker 1: like that because you know, the farmer was replaced who
38:59
Speaker 1: drove that tractor and isn't skilled in the practice of
39:02
Speaker 1: designing robot tractors. And in this case, in most cases,
39:08
Speaker 1: they're not being there's not some other job waiting for
39:12
Speaker 1: someone who got fired in the world of designing AI.
39:16
Speaker 1: Does that make sense?
39:17
Speaker 2: No, it makes total sense. But yeah, and in this case,
39:19
Speaker 2: one of the big differences is instead of the farmer
39:22
Speaker 2: having to go figure out how to work a computer,
39:25
Speaker 2: that people working computers now have to go figure out
39:27
Speaker 2: how to be farmers in order to sustain themselves. Right,
39:31
Speaker 2: But you're right, we don't have a track record of
39:34
Speaker 2: taking care of people very well, at least who are
39:38
Speaker 2: out of a job. And I mean, without getting on
39:40
Speaker 2: a soapbox here, what's either going to come out of this,
39:44
Speaker 2: because there's going to be one or the other. The
39:45
Speaker 2: status quo as it is now or as it wasn't
39:48
Speaker 2: as of up to twenty twenty two. We don't know
39:51
Speaker 2: that that's going to be around anymore. Instead, we'll either
39:55
Speaker 2: do something like create universal basic income for people to
39:59
Speaker 2: be like, hey, your industry literally does not exist anymore
40:03
Speaker 2: and it just happened overnight. Basically, we're just gonna make
40:07
Speaker 2: sure that everybody's at least minimally taken care of while
40:09
Speaker 2: we're figuring out what comes next, or it's gonna be
40:13
Speaker 2: like good luck, chump, you're fired, You're out on your own. Instead,
40:17
Speaker 2: we're gonna take all this extra wealth, this extra two
40:20
Speaker 2: trillion dollars that's gonna be generated, and push it upward
40:23
Speaker 2: toward the wealthy instead and everybody else. Is just the
40:28
Speaker 2: divide between wealthy and not wealthy is just going to
40:32
Speaker 2: exponentially grow. One of those two things is gonna happen,
40:35
Speaker 2: because I don't see how there's just gonna be a
40:37
Speaker 2: regular middle ground like there is now where it's kind
40:40
Speaker 2: of shaky, and how we're taking care of people, because
40:44
Speaker 2: there's just gonna be so many layoffs and fairly skilled
40:48
Speaker 2: workers being laid off too. We've just never encountered that before.
40:52
Speaker 1: Yeah. I mean, that's the thing that these the largest
40:56
Speaker 1: corporations might want to think about. Is all that's gonna
40:59
Speaker 1: take is one CEO of a huge corporation to say,
41:06
Speaker 1: wait a minute, it I think I can get rid
41:09
Speaker 1: of seventy five percent of the vps in my in
41:12
Speaker 1: my company, right, and like who unless who accept the
41:18
Speaker 1: person at the very very top of that food chain
41:21
Speaker 1: is protected. And the answer is nobody.
41:24
Speaker 2: No, No, the offense essentially.
41:26
Speaker 1: At the end of the day, because they make a
41:28
Speaker 1: lot of money. If you it's one thing to lay
41:29
Speaker 1: off a bunch of you know, technical writers that are
41:32
Speaker 1: all sitting in their cubicles, But if you start laying
41:35
Speaker 1: off those those vps who get those big bonuses, that's
41:38
Speaker 1: more bonus money. And you know, are we looking at
41:41
Speaker 1: a situation where a corporation is run by one human?
41:44
Speaker 2: I mean, it's entirely possible, Like you can make a
41:47
Speaker 2: really good case that what it is going to wipe
41:49
Speaker 2: out is the middle management, right, vps, This is exactly
41:53
Speaker 2: like you said, and that we still will need some
41:55
Speaker 2: humans to do some stuff.
41:57
Speaker 1: Like the board take care of the board, right, sure.
41:59
Speaker 2: Of course, yeah, but yes, I mean who knows, we
42:02
Speaker 2: have no idea at this point. Ultimately, it could very
42:07
Speaker 2: easily provide for a much better healthier society, at least
42:13
Speaker 2: financially speaking, it could do that, especially given a long
42:17
Speaker 2: enough period of time.
42:19
Speaker 1: I'm a cynic when it comes to that kind of
42:21
Speaker 1: trust though.
42:21
Speaker 2: I am as well for sure. But if you look
42:23
Speaker 2: back in history at the history of technology overall, especially
42:28
Speaker 2: if you just turn a blind eye to human suffering
42:30
Speaker 2: for a second and you just look at the progress
42:32
Speaker 2: of society. Right in a lot of ways it has
42:36
Speaker 2: has gotten better and better things to technology. There's also
42:39
Speaker 2: a lot of downsides to it. Nothing's black and white,
42:43
Speaker 2: it's just not that's just not how things are. So
42:46
Speaker 2: there's of course going to be problems. There's going to
42:48
Speaker 2: be suffering, there's going to be people left behind, there's
42:50
Speaker 2: going to be people that fall through the cracks. It's
42:52
Speaker 2: just inevitable. We just don't know how many people for
42:56
Speaker 2: how long and what will happen to those people on
42:59
Speaker 2: the other of this transition.
43:02
Speaker 1: Yeah, I was talking with somebody the other day about
43:06
Speaker 1: the writer's strike in Hollywood. The WGA is striking right now.
43:12
Speaker 1: For those of you who don't know, it's kind of
43:14
Speaker 1: all over the place. But one of the things that
43:16
Speaker 1: they have argued for in this round of negotiations is, hey,
43:21
Speaker 1: you can't replace us with AI, and the studios all
43:25
Speaker 1: came back and said, well, how about this, We'll assess
43:28
Speaker 1: that on a year to year basis. And that's frightening
43:33
Speaker 1: if you're if you're either a writer in Hollywood or
43:36
Speaker 1: you're somebody who loves TV and films and quality TV
43:41
Speaker 1: in films, because I don't know if I think ideation
43:46
Speaker 1: and initial scripts maybe even right now, could I could
43:52
Speaker 1: see that happening where they said they're like, all right now,
43:54
Speaker 1: we'll bring in a human to refine this thing at
43:58
Speaker 1: a much lower wage. That it's probably what they're most
44:01
Speaker 1: afraid of, rather than being wholesale replaced, because like you said,
44:04
Speaker 1: these programs are they're all about just data and numbers.
44:09
Speaker 1: They're not They don't have human feelings, and that's what
44:12
Speaker 1: art is. And so I think I would be more
44:15
Speaker 1: concerned if I was writing pamphlets for Verizon or something,
44:20
Speaker 1: or if I was.
44:22
Speaker 2: Some pamphlet writer for Verizon.
44:23
Speaker 1: Just went in gulp, No, I'm so sorry. But like
44:28
Speaker 1: BuzzFeed back in the day, instead of having a dozen
44:30
Speaker 1: writers writing clickbait articles, why not have just one human
44:34
Speaker 1: that is a prompt engineer that's managing a virtual AI
44:39
Speaker 1: clickbait room that's just pumping out these articles that you
44:43
Speaker 1: know they were paying someone down to forty grand a
44:46
Speaker 1: year to write previously.
44:47
Speaker 2: Yeah, I mean it's a great question, like that was
44:50
Speaker 2: a horrific, horrible job to have. Not too many years ago.
44:54
Speaker 2: So it's great to have a computer do it, but
44:56
Speaker 2: that means that we need these other people to go
44:58
Speaker 2: on to be to have writing jobs that are more
45:02
Speaker 2: satisfying to them than that. But that's not necessarily the case,
45:05
Speaker 2: because as these things get smarter and better, they're just
45:10
Speaker 2: going to be relied upon more. We're not going to
45:12
Speaker 2: go back. There's no going back now. It just happened
45:15
Speaker 2: like it just happened basically as of March twenty twenty three.
45:19
Speaker 2: And one of the big problems that people have already
45:23
Speaker 2: projected running into is if computers replace humans, say writers,
45:33
Speaker 2: basically entirely. Eventually all the stuff that humans have written
45:36
Speaker 2: on the Internet is going to become dated. It's going
45:38
Speaker 2: to stop, yeah, and it will have been replaced and
45:41
Speaker 2: picked up on by generative, pre trained transformers. Right, And
45:48
Speaker 2: eventually all the writing on the Internet after a certain
45:52
Speaker 2: date will have been written by computers, but will be
45:54
Speaker 2: being scraped by computers. When humans go ask the computer
45:57
Speaker 2: a question, the computer then goes in reference is something
46:01
Speaker 2: written by a computer, So humans will be completely taken
46:04
Speaker 2: out of the equation in that in that respect, we'll
46:06
Speaker 2: be getting all of our information, at least non historical
46:09
Speaker 2: information from non humans, and that could be a really
46:13
Speaker 2: big problem, not just in the fact that we're losing
46:16
Speaker 2: jobs or in the fact that computers are now telling
46:19
Speaker 2: us all of our information, but also that there's some
46:23
Speaker 2: there's some part of what humans put into things that
46:27
Speaker 2: will be lost that I think we're going to demand.
46:31
Speaker 2: I saw somebody put it like, I think, I can't
46:34
Speaker 2: remember who it was, but they said, we're going like
46:37
Speaker 2: people will go seek out human written stuff. There will
46:40
Speaker 2: always be audiences for human written stuff. Yeah. Maybe, like
46:43
Speaker 2: you said, will rely on computers to write the Verizon pamphlets,
46:47
Speaker 2: but we're not going to rely on computers to write
46:49
Speaker 2: great works of literature or to create great works of art,
46:52
Speaker 2: Like we're just not going to They'll still do that.
46:55
Speaker 2: They're going to be writing books and movies and all that,
46:56
Speaker 2: but there will always be a taste in a market
46:59
Speaker 2: for human created stuff. This guy said, I think he's right.
47:04
Speaker 1: Yeah. And Justine Bateman, I don't know if you saw that.
47:09
Speaker 1: I don't know if it was a blog poster.
47:11
Speaker 2: Are you having a hallucination right now? Did you mean
47:14
Speaker 2: Justine Bateman?
47:15
Speaker 1: Yeah, yeah, Justin Bateman, the Jason Bateman's sister the actor,
47:20
Speaker 1: and she's done all kinds of things since then. I
47:23
Speaker 1: know she has a computer science degree, so she's very
47:26
Speaker 1: smart and knows a lot about this stuff. But she
47:29
Speaker 1: basically said, and this is beyond just the chatbot stuff.
47:32
Speaker 1: But she was like, right now, there are major Hollywood
47:36
Speaker 1: stars being scanned, and there may be a brand new
47:41
Speaker 1: Tom Cruise movie in sixty years. Yeah, long after he's
47:45
Speaker 1: dead starring Tom Cruise. He may be making movies for
47:49
Speaker 1: the next two hundred years. And like, is this what
47:52
Speaker 1: you want? Actors? Do you want to be scanned and
47:55
Speaker 1: have them use your image like this in perpetuity for
47:59
Speaker 1: you know, there will be money involved. It's not like
48:02
Speaker 1: they can just say, Okay, we can just do whatever
48:03
Speaker 1: we want. But what if they're like, here's a billion dollars,
48:06
Speaker 1: Tom Cruise, just for the use of your image in perpetuity,
48:11
Speaker 1: because we will be able to duplicate that so realistically
48:15
Speaker 1: that people won't know human voices, same thing that's already happening.
48:20
Speaker 1: What yeah, it is. So that stuff is kind of scary.
48:26
Speaker 1: And you know when you read I didn't really know
48:29
Speaker 1: this was kind of already happening in companies. But Olivia
48:33
Speaker 1: found this stuff IBM CEO Arvin Krishna said just last
48:37
Speaker 1: month in May that he believed thirty percent of back
48:39
Speaker 1: office jobs could be replaced over five years, and it
48:44
Speaker 1: was pausing hiring for close to eight thousand positions because
48:48
Speaker 1: they might be able to use AI instead. And then
48:52
Speaker 1: Dropbox talked about the AI era when they announced a
48:56
Speaker 1: round of layoffs. So it is happening right now in
49:00
Speaker 1: real time.
49:00
Speaker 2: Pretty amazing. Yeah, that's I mean, there's proof positive right there,
49:05
Speaker 2: Like that guy couldn't even wait a couple months a year,
49:09
Speaker 2: Like this really started up in March, and he's saying
49:12
Speaker 2: this in May. Already in May, they're like, wait, wait,
49:14
Speaker 2: stop hiring. We're gonna eventually replace these guys with AI
49:18
Speaker 2: so soon that we're going to stop hiring those positions
49:21
Speaker 2: for now until the AI is competent enough to take over.
49:24
Speaker 1: I mean, how many people does ib employ? What, what's
49:27
Speaker 1: thirty percent of that?
49:28
Speaker 2: I don't know. I would say at least at least
49:31
Speaker 2: one hundred people, right, So, yeah, like you said, it's
49:34
Speaker 2: happening already. And then one other thing to look out
49:37
Speaker 2: for too, that's I believe is already at least theoretically
49:40
Speaker 2: possible since AI can write code. Now they'll be able
49:45
Speaker 2: to create new large language models themselves, So the computers
49:50
Speaker 2: will be able to create new AI.
49:54
Speaker 1: Well, that's the singularity, right.
49:56
Speaker 2: No, the singularity is when one of them a understands
50:00
Speaker 2: what it is and become Yes, that's the singularity.
50:05
Speaker 1: But this leads to that though, doesn't it It does.
50:07
Speaker 2: It's hypothetically yes, but we just understand what's going on
50:11
Speaker 2: so little that you just can't say either way. Really,
50:13
Speaker 2: you definitely can't say that it no, it won't happen.
50:16
Speaker 2: It's just fantasy. And you also can't say yes, it's
50:18
Speaker 2: definitely gonna happen.
50:19
Speaker 1: Yeah. And here's the thing, man, I'm not a paranoid technophobe.
50:26
Speaker 2: You don't any measure foiled cap on.
50:28
Speaker 1: No, by any measure. I'm a pretty positive thinker, and
50:33
Speaker 1: this this is pretty scary to me.
50:37
Speaker 2: I'm just gonna leave that there, agreed, Chuck. Okay, if
50:44
Speaker 2: you want to know more about large language models, everybody
50:47
Speaker 2: just to start looking around Earth and when you see
50:51
Speaker 2: people running from explosions, go toward it and ask what's
50:54
Speaker 2: going on?
50:55
Speaker 1: You almost said, type it into a search engine.
50:57
Speaker 2: Yeah, steer clear of those. Yeah, there's so much more,
50:59
Speaker 2: we could have talked about. But this is if you
51:02
Speaker 2: ask me, this is round one. I think we definitely
51:04
Speaker 2: need to do at least one or so.
51:06
Speaker 1: More on this, Okay, Yeah, and then one day, like
51:08
Speaker 1: I said, Ai, Josh and Chuckle, just wrap it all
51:11
Speaker 1: up and spank it on the bottom and say no
51:13
Speaker 1: problems here.
51:14
Speaker 2: Hopefully they'll give us a billion dollars rather than like
51:17
Speaker 2: a month free of blue Apron instead.
51:20
Speaker 1: Yeah, I mean we could talk here.
51:23
Speaker 2: Well, since Chuck said we can talk real confidential like
51:27
Speaker 2: that means it's time for listener.
51:28
Speaker 1: May I'm gonna call this conception, not inception, but conception.
51:35
Speaker 2: Oh I saw this one. I don't know how I
51:36
Speaker 2: feel about this.
51:39
Speaker 1: Hey, guys. Last year, my wife and I were attempting
51:41
Speaker 1: to get pregnant. A couple of months in we made
51:43
Speaker 1: plans to stay with some friends in another town for
51:45
Speaker 1: a weekend, and when we can arrive, it happened to
51:47
Speaker 1: coincide with my wife's ovulation cycle. As shy people, we
51:52
Speaker 1: both felt a little bit awkward about, you know, hugging
51:55
Speaker 1: and kissing in a friend's guest room, but we really
51:59
Speaker 1: didn't want to miss that chance and that time of
52:01
Speaker 1: the month, so we went about getting in the mood
52:04
Speaker 1: as quietly as possible, and my wife suggested we play
52:07
Speaker 1: a podcast from my phone so that, you know, if
52:09
Speaker 1: any noise is made outside the room, it would sound
52:12
Speaker 1: like we were just doing a little pre bedtime listening.
52:15
Speaker 1: I knew I needed something with nice, simple production values,
52:18
Speaker 1: so we wouldn't get distracted, of course, by the whiz
52:21
Speaker 1: bang sounds and whatnot. And since you were my intro
52:23
Speaker 1: to the world of podcast, I've always had a steady
52:26
Speaker 1: supply of yours downloaded. I picked the least interesting sounding
52:29
Speaker 1: one in the feed at the time, How Coal Works.
52:33
Speaker 2: Okay, I thought that one turned out to be surprisingly interesting. Yeah,
52:38
Speaker 2: I could see how we would have thought that though.
52:40
Speaker 1: Yeah, for sure. We put that on and we did
52:43
Speaker 1: our business. Six weeks later we got a positive pregnancy test,
52:47
Speaker 1: and now over a year later, we've welcomed our son
52:50
Speaker 1: into the world. Name is Cole, and of course we
52:53
Speaker 1: named him Cole. That is what this person said.
52:57
Speaker 2: Wait a minute, Wait a minute, they really did name
52:59
Speaker 2: them Cole.
53:00
Speaker 1: No, he said it as a joke. Oh but great
53:04
Speaker 1: minds right, good joke for both of them. It's almost
53:07
Speaker 1: like you're both chatbox. And this person said You're fine
53:13
Speaker 1: to read this, but give me a fake name. And
53:15
Speaker 1: so I just want to say thanks to Gene for
53:18
Speaker 1: writing in about this.
53:20
Speaker 2: Gene is in Gene Transfer.
53:23
Speaker 1: Sure, thanks a lot, Gene.
53:25
Speaker 2: We appreciate that. I think again, I'm still figuring that
53:29
Speaker 2: one out. And if you want to be like Gene,
53:32
Speaker 2: I'm making air quotes here, you can send us an
53:36
Speaker 2: email to wrap it up. Spank it on the bottom.
53:39
Speaker 2: Only humans can do that.
53:41
Speaker 1: I wonder if when you said spanking on the bottom,
53:43
Speaker 1: if that created any issues.
53:45
Speaker 2: Yeah, I hadn't thought about that. Maybe playfully how about that? Sure,
53:51
Speaker 2: and send it off to stuff podcast at iHeartRadio dot com.
53:58
Speaker 1: Stuff you Should Know is a production of iHeartRadio. For
54:01
Speaker 1: more podcasts my heart Radio, visit the iHeartRadio app, Apple Podcasts,
54:06
Speaker 1: or wherever you listen to your favorite shows.