WEBVTT - TechStuff Tidbits: ChatGPT and Natural Language Processing 0:00:04.400 --> 0:00:07.800 Welcome to tech Stuff, a production from I Heart Radio. 0:00:11.840 --> 0:00:14.720 Hey there, and welcome to tech Stuff. I'm your host, 0:00:14.880 --> 0:00:18.440 Jonathan Strickland. I'm an executive producer with I Heart Radio 0:00:18.560 --> 0:00:21.520 and how the tech are you? We're gonna do a 0:00:21.800 --> 0:00:25.200 tech stuff tidbits episode for those of y'all who are 0:00:25.280 --> 0:00:28.600 not familiar wants in a while, I try and do 0:00:28.880 --> 0:00:32.760 kind of a shorter, more concentrated episode on a tech 0:00:32.840 --> 0:00:36.360 topic that doesn't merit a full episode. Now, it's not 0:00:36.400 --> 0:00:39.440 to say that this particular topic doesn't merit a full episode, 0:00:39.640 --> 0:00:44.120 but it will actually require a lot more in depth discussion, 0:00:44.360 --> 0:00:46.720 and I wanted to kind of give an overview just 0:00:46.840 --> 0:00:50.360 to give y'all something to think about. So let's get 0:00:50.400 --> 0:00:53.920 to it. Late last November, the Open Ai organization launched 0:00:54.120 --> 0:00:58.000 chat GPT, which if you haven't used it, I recommend 0:00:58.040 --> 0:01:00.520 you go play with it. Try it out. You can 0:01:00.560 --> 0:01:03.440 make a little account for free and interact with this 0:01:03.520 --> 0:01:06.679 chat bot. So the GPT for chat GPT stands for 0:01:06.840 --> 0:01:11.800 Generative pre Trained Transformer. But it is a chat bot. Now, 0:01:11.800 --> 0:01:13.880 it's a chat bot that can respond to prompts and 0:01:13.959 --> 0:01:18.880 queries with sometimes astonishing results. Uh. This is the tool 0:01:19.120 --> 0:01:21.679 that has certain teachers worried that their students will be 0:01:21.720 --> 0:01:25.960 hoisting all of their homework onto this chat bot and 0:01:26.000 --> 0:01:28.039 make it, you know, write essays for them, and it 0:01:28.080 --> 0:01:31.000 may not always be easy to spot when that happens. 0:01:31.360 --> 0:01:34.360 There are people who are actually offering tools that purport 0:01:34.720 --> 0:01:38.560 to be able to detect if it was generated by 0:01:38.680 --> 0:01:41.160 a chat bot as opposed to written by a student. 0:01:41.280 --> 0:01:47.080 So there's already this kind of seesaw struggle between the 0:01:47.160 --> 0:01:49.680 chat bots and the detectors, which is kind of interesting. 0:01:50.160 --> 0:01:53.160 Some people argue that Pandora's box has been opened and 0:01:53.240 --> 0:01:56.280 chat GPT came roaring out, and it's going to make 0:01:56.760 --> 0:01:59.960 teaching even more difficult and challenging that already is. By 0:02:00.040 --> 0:02:02.280 the way, if you're a teacher, my hat's off to you. 0:02:02.480 --> 0:02:05.480 Both my parents were teachers, they're both retired now. I 0:02:05.520 --> 0:02:09.800 have nothing but the utmost respect for teachers. Y'all have 0:02:10.120 --> 0:02:18.600 a incredibly important, incredibly challenging, and criminally undervalued job. Anyway, 0:02:18.680 --> 0:02:21.760 for chat GPT to work as well as it does, 0:02:21.840 --> 0:02:24.080 there are a ton of different elements that had to 0:02:24.120 --> 0:02:28.200 be implemented. Open Ai had to develop a sophisticated machine 0:02:28.280 --> 0:02:31.919 learning process to train chat GPT, so that it can 0:02:31.960 --> 0:02:37.760 formulate relevant and convincing but not always accurate or correct 0:02:38.040 --> 0:02:41.200 responses to prompts, and relevant is really important. Right. If 0:02:41.240 --> 0:02:45.239 you ask something a question and you get an answer 0:02:45.280 --> 0:02:48.959 that's about something entirely different, that's a really frustrating experience. 0:02:49.560 --> 0:02:51.399 So one thing that the app needs to be able 0:02:51.400 --> 0:02:55.440 to do is parse human language. And this is a 0:02:55.520 --> 0:02:59.360 non trivial engineering problem, and we're going to talk about 0:02:59.400 --> 0:03:02.680 that briefly in this Tech Stuff Tidbits episode. So while 0:03:02.760 --> 0:03:06.120 chat GPT is the launching off point, this is not 0:03:06.560 --> 0:03:11.800 exclusively applicable to chat GPT. It's actually applicable to a 0:03:11.880 --> 0:03:15.600 wide range of applications. So when you hear the phrase 0:03:16.000 --> 0:03:20.080 natural language, that means it's how we humans tend to 0:03:20.080 --> 0:03:24.440 communicate with one another with languages that naturally evolved out 0:03:24.480 --> 0:03:30.240 of culture and society. There are manufactured languages like Esperanto, 0:03:30.400 --> 0:03:34.080 where people created a language with the purpose of creating 0:03:34.080 --> 0:03:37.920 a language, but generally natural language just refers to a 0:03:38.080 --> 0:03:43.440 language naturally evolving over time, and we've developed these over millennia, right, 0:03:43.520 --> 0:03:47.320 And a language includes things like the rules we use 0:03:47.440 --> 0:03:50.640 to form phrases and sentences, as well as the vocabulary 0:03:50.760 --> 0:03:55.040 we use to populate those phrases and sentences. So we've 0:03:55.040 --> 0:03:58.840 got syntax and grammar, and we've got vocabulary. All of 0:03:58.880 --> 0:04:02.280 these elements come together to form language, and you can't 0:04:02.320 --> 0:04:05.320 just throw them in willy nilly, or you will make 0:04:05.360 --> 0:04:08.480 no sense. That's why we have the structure and rules. 0:04:09.000 --> 0:04:12.240 It also includes the quirks we develop over time, you know, 0:04:12.280 --> 0:04:17.479 exceptions to established rules and things like puns and jokes 0:04:17.480 --> 0:04:20.200 and idioms and that kind of thing. Right, we have 0:04:20.520 --> 0:04:24.159 phrases that we use that means something, but they don't 0:04:24.240 --> 0:04:27.960 necessarily mean the surface level. If I say it's raining 0:04:28.000 --> 0:04:31.760 cats and dogs, I probably do not mean that it 0:04:32.040 --> 0:04:36.240 literally is raining small furry animals. If it is, I 0:04:36.279 --> 0:04:40.240 will be inconsolable. I am saying instead that it's raining 0:04:40.279 --> 0:04:45.640 really hard, but I'm using this idiom to express that. Right, So, 0:04:45.760 --> 0:04:49.040 for humans, it's not that difficult to understand what someone 0:04:49.200 --> 0:04:53.560 is saying, assuming you both speak the same language, even 0:04:53.920 --> 0:04:56.840 if that other person is using some unfamiliar words, like 0:04:57.080 --> 0:04:59.159 you could be sitting in on I think back to 0:04:59.279 --> 0:05:02.800 when I was in a class in English class and 0:05:02.880 --> 0:05:07.279 I had a professor who insisted on using the word paradigm, 0:05:07.320 --> 0:05:10.320 and I had never encountered the word paradigm. I it 0:05:10.440 --> 0:05:14.520 was not in my vocabulary, and I picked up in 0:05:14.640 --> 0:05:18.599 context that the way she was specifically using paradigm was 0:05:18.680 --> 0:05:21.640 essentially as a stand in for the word example. Uh. 0:05:21.680 --> 0:05:23.680 This taught me two things. It taught me one of 0:05:23.720 --> 0:05:25.520 the meetings of the word paradigm, and it taught me 0:05:25.560 --> 0:05:30.320 that my professor was incredibly pretentious. Anyway, context really can 0:05:30.480 --> 0:05:33.320 clue us in. Even if we don't know what a 0:05:33.360 --> 0:05:37.080 specific word or phrase means on the surface, from context, 0:05:37.160 --> 0:05:40.920 we can derive at least some meaning. We only still 0:05:41.000 --> 0:05:44.560 miss the full meaning. Or on occasion, you might pull 0:05:44.600 --> 0:05:47.520 a malapropism and you might use the wrong word in 0:05:47.600 --> 0:05:50.679 place of one you intended to use. This has happened 0:05:50.720 --> 0:05:54.360 to me many times. If you want a recent example 0:05:54.720 --> 0:05:57.800 of a character who uses malapropisms, you should watch the 0:05:57.839 --> 0:06:01.360 movie Glass Onion. There is a character in that who 0:06:01.480 --> 0:06:05.680 frequently uses the wrong word to stand in for something 0:06:05.720 --> 0:06:09.080 that he intends to say. Also a fun side note, 0:06:09.560 --> 0:06:13.480 the word malapropism comes from the world of theater. Richard 0:06:13.560 --> 0:06:17.080 Brinsley Sheridan wrote a play called The Rivals. This was 0:06:17.120 --> 0:06:20.160 back in the eighteenth century, so the late seventeen hundreds 0:06:20.320 --> 0:06:24.919 and the rivals includes a character h a a a 0:06:24.960 --> 0:06:29.599 caretaker sort of or a chaperone almost, and her name 0:06:29.680 --> 0:06:34.040 is Mrs Malaprop, and she often employs the wrong words 0:06:34.120 --> 0:06:36.679 to comedic effects. She means one thing, but she says 0:06:36.680 --> 0:06:41.240 another because she's using the incorrect words, which makes the 0:06:41.279 --> 0:06:44.200 meaning of what she's actually saying change and that's where 0:06:44.200 --> 0:06:46.080 the humor is right. And I think this is a 0:06:46.080 --> 0:06:50.039 fun bit of history to know where Malaprop comes from, 0:06:50.080 --> 0:06:53.760 because there's actually another famous character from theater, from English 0:06:53.760 --> 0:06:57.880 theater who predated Mrs Malaprop by a couple of hundred years. 0:06:58.760 --> 0:07:03.479 That character is Ugberry from Billy Shakespeare's Much Ado About Nothing, 0:07:04.040 --> 0:07:07.440 and dog Berry also frequently uses the wrong words in 0:07:07.520 --> 0:07:12.320 an intended comedic effect. The difference is Mrs Malaprop is 0:07:12.360 --> 0:07:16.360 actually funny and dog Berry more often than not isn't 0:07:17.280 --> 0:07:20.720 and mostly being a snobby tease. Here up, please keep 0:07:20.720 --> 0:07:23.400 in mind that back in college I majored in English 0:07:23.440 --> 0:07:26.520 Lit with a focus on Shakespeare. I love Shakespeare dearly, 0:07:27.000 --> 0:07:30.360 but I have issues with some of his comedy. Although 0:07:30.360 --> 0:07:34.680 I guess you could argue the the broad gap in 0:07:34.760 --> 0:07:37.280 time between his time and mind could play a part 0:07:37.280 --> 0:07:41.040 in that anyway. My point is it's not too hard 0:07:41.080 --> 0:07:43.960 for humans to communicate the same thought in lots of 0:07:44.000 --> 0:07:46.960 different ways, and if we encounter a new turn of 0:07:47.000 --> 0:07:49.520 phrase or a new word, we can pick it up 0:07:49.760 --> 0:07:54.960 without too much problem. This stands in stark contrast to 0:07:55.120 --> 0:07:59.240 computers and machines. Now often in the show I talk 0:07:59.400 --> 0:08:04.400 about binary information, about bits and bytes, because when you 0:08:04.440 --> 0:08:09.280 dig way down into how many machines process information, you're 0:08:09.320 --> 0:08:13.280 looking at circuits that run various mathematical operations upon strings 0:08:13.280 --> 0:08:17.040 of data, and that that data is grouped into zeros 0:08:17.080 --> 0:08:22.640 and ones. This is binary or machine language. Computers can 0:08:22.680 --> 0:08:26.440 interpret this quickly. The computers are effectively looking at a 0:08:26.520 --> 0:08:30.480 series of off or on indicators. I often say a 0:08:30.680 --> 0:08:33.720 binary digit or bit is a lot like a light switch. 0:08:33.880 --> 0:08:37.080 It's either off or on. It's zero or one. When 0:08:37.080 --> 0:08:40.200 you put this through logic gates that are have specific 0:08:40.200 --> 0:08:44.400 designs to them, you run them through various operations, these 0:08:44.480 --> 0:08:48.480 zeros and ones can almost magically become complicated processes that 0:08:48.880 --> 0:08:51.679 let you do anything from type out the script for 0:08:51.720 --> 0:08:57.040 a podcast to playing the latest video game. Now make 0:08:57.080 --> 0:09:01.840 sure you keep in mind I said, almost magically in fact, 0:09:01.920 --> 0:09:05.360 it's not magical at all. It's just that we humans 0:09:05.400 --> 0:09:08.440 can't really process huge banks of zeros and ones and 0:09:08.520 --> 0:09:11.839 really make much sense of it. While on a similar note, 0:09:12.200 --> 0:09:15.400 a machine without the proper programming can make no sense 0:09:15.880 --> 0:09:21.400 of our human languages. It is complete gibberish. It's meaningless 0:09:21.440 --> 0:09:25.360 to a computer. Now, it would be exceedingly difficult to 0:09:25.440 --> 0:09:29.640 do any significant kind of programming if programmers had to 0:09:29.679 --> 0:09:35.720 depend upon hard coding zeros and ones while composing their work. Fortunately, 0:09:36.080 --> 0:09:42.160 computer scientists came up with solutions, namely compilers and computer languages, 0:09:42.760 --> 0:09:48.120 and we're gonna start with computer languages. A computer language 0:09:48.280 --> 0:09:51.640 creates levels of abstraction that make it a bit less 0:09:51.760 --> 0:09:55.800 daunting to write software for computers. So the idea is 0:09:55.840 --> 0:09:59.720 that each computer language has its own set of rules, 0:09:59.760 --> 0:10:03.480 like its own syntax, its own vocabulary, and as long 0:10:03.520 --> 0:10:05.760 as you work within those rules, and you do so 0:10:05.920 --> 0:10:08.760 precisely and with as few errors as you possibly can, 0:10:09.360 --> 0:10:11.920 you can create a program to make the computer do 0:10:12.000 --> 0:10:14.559 whatever it is you want it to do. Because the 0:10:14.600 --> 0:10:19.160 computer language itself has built into it the ability to 0:10:19.280 --> 0:10:24.079 be converted into machine language. That will get us two compilers, 0:10:24.400 --> 0:10:26.040 which we will chat about in the minute. In fact, 0:10:26.040 --> 0:10:29.440 we'll talk more about computer languages and compilers after we 0:10:29.520 --> 0:10:41.400 come back from this quick break. Okay, before the break, 0:10:41.880 --> 0:10:44.520 I had introduced this idea of computer languages. Now, not 0:10:44.640 --> 0:10:48.199 all computer languages are equal, and there are dozens of them. 0:10:48.480 --> 0:10:51.480 In fact, there are some notable differences between computer languages. 0:10:51.520 --> 0:10:53.760 Some of them are very similar, some of them share 0:10:53.880 --> 0:10:57.560 a common root computer language, and some of them could 0:10:57.600 --> 0:11:01.240 not be more different. Now, in general, we can separate 0:11:01.280 --> 0:11:03.679 them into very two, very broad categories. There are a 0:11:03.720 --> 0:11:07.320 lot of different ways to categorize programming languages, but we're 0:11:07.360 --> 0:11:10.040 looking at one of the most basic, which is low 0:11:10.120 --> 0:11:14.720 level languages and high level languages. So a low level 0:11:14.800 --> 0:11:20.640 language is relatively close to machine language. It only provides 0:11:21.080 --> 0:11:25.680 a thin layer of abstraction, and as such it can 0:11:25.760 --> 0:11:29.320 still be very challenging for programmers to work with these 0:11:29.400 --> 0:11:32.600 languages because they're not that far off from the basic 0:11:32.679 --> 0:11:37.720 machine languages. Uh, it's a little easier. It's it's designed 0:11:37.720 --> 0:11:41.319 so that humans can interact with it a little more 0:11:41.520 --> 0:11:45.199 naturally than they would with pure machine language, but it's 0:11:45.240 --> 0:11:48.320 not that easy. A high level programming language, on the 0:11:48.320 --> 0:11:52.040 other hand, has a great deal of abstraction, and it's 0:11:52.440 --> 0:11:54.960 much closer to a human language in that way. So 0:11:55.080 --> 0:11:58.319 these languages are far easier for humans to work with 0:11:58.400 --> 0:12:00.840 on a day to day basis, And by humans I 0:12:00.880 --> 0:12:03.920 mean programmers. Like, if you had never seen a computer 0:12:04.080 --> 0:12:07.240 programming language and one day you just sat down to 0:12:07.400 --> 0:12:10.559 work on one, it would not seem easy or intuitive 0:12:10.600 --> 0:12:14.079 to you, probably, But both low level and high level 0:12:14.160 --> 0:12:17.600 languages have their own sets of rules. If you break 0:12:17.640 --> 0:12:19.680 those rules, your program is not going to behave the 0:12:19.679 --> 0:12:22.000 way you intended to. It might not run at all. 0:12:22.080 --> 0:12:25.840 You might just get error messages. But yeah, they have rules, 0:12:25.840 --> 0:12:30.839 and if you follow them, then programs can potentially work. Now, 0:12:30.880 --> 0:12:34.839 if computers process information in zeros and ones, and programming 0:12:34.920 --> 0:12:38.520 languages provide levels of abstraction that approach human language at 0:12:38.520 --> 0:12:42.120 the higher levels, how do we reconcile that? How does 0:12:42.240 --> 0:12:47.360 a computer interpret a program written in uh language like Python? 0:12:47.559 --> 0:12:51.720 For example? Python is much closer to a natural language 0:12:51.720 --> 0:12:53.760 than it is a machine language. So how do we 0:12:53.800 --> 0:12:55.800 get to the point where a computer can take that 0:12:55.880 --> 0:13:00.160 information and actually execute a program? Well, the program I'm 0:13:00.200 --> 0:13:03.000 once composed has to go through what we call a compiler, 0:13:03.160 --> 0:13:06.520 and the compiler's job is to take this program and 0:13:06.600 --> 0:13:10.600 according to the rules of that programming language, convert the 0:13:10.640 --> 0:13:14.480 program from the programming language into machine code so that 0:13:14.559 --> 0:13:17.960 a computer can actually do something with it. So the 0:13:18.000 --> 0:13:21.880 compiler is kind of like a translator. And the compiler 0:13:21.880 --> 0:13:23.959 actually has to do a few jobs to make this happen, 0:13:24.040 --> 0:13:27.160 and has to scan the program's source code for recognizable 0:13:27.200 --> 0:13:30.640 commands and terms. That has to analyze the syntax a 0:13:30.760 --> 0:13:34.640 gave the structure of the code to understand the order 0:13:34.679 --> 0:13:37.840 of operations. That has to break all that down into 0:13:37.960 --> 0:13:41.000 machine code that follows what the language says. And if 0:13:41.040 --> 0:13:43.680 the programmer made a mistake, well that ends up getting 0:13:43.880 --> 0:13:46.280 translated to and then you don't find out un till 0:13:46.280 --> 0:13:48.040 you try and run the program, and then it gets 0:13:48.080 --> 0:13:51.280 back to debugging figuring out where did you make that 0:13:51.360 --> 0:13:55.280 mistake in the actual program. Now here's the thing. The 0:13:55.360 --> 0:13:59.280 programming languages, while they can create high levels of abstraction, 0:13:59.679 --> 0:14:02.840 are still not necessarily accessible to the average person. Like 0:14:02.880 --> 0:14:04.880 I had mentioned before the break, you know, I'm talking 0:14:04.880 --> 0:14:07.800 about the average person who has little to know experience 0:14:07.800 --> 0:14:11.280 with computer languages or programming. So to people like that, 0:14:11.360 --> 0:14:14.920 and I'll include myself here, a sheet of code written, 0:14:14.920 --> 0:14:18.280 and even a very high level computer language might end 0:14:18.360 --> 0:14:20.880 up being indecipherable. You could look at and say, I 0:14:20.920 --> 0:14:24.320 don't know what this program is supposed to do because 0:14:24.320 --> 0:14:27.080 I don't know enough about this programming language to understand 0:14:27.120 --> 0:14:30.520 what any of this means. As such, for folks such 0:14:30.520 --> 0:14:34.920 as myself, programming a computer is a daunting task because 0:14:35.040 --> 0:14:38.080 we lack the basic knowledge of the programming languages we 0:14:38.120 --> 0:14:41.760 would need to use to make an effective program. Now 0:14:41.800 --> 0:14:44.280 you you can teach yourself these things lots of people have. 0:14:44.400 --> 0:14:47.920 In fact, a lot of the most famous hackers and 0:14:47.960 --> 0:14:53.479 even like leaders in in tech business are self taught programmers. 0:14:53.800 --> 0:14:56.440 So there's nothing stopping you from doing this, even if 0:14:56.480 --> 0:14:59.560 you never took a class in computer science or programming. 0:15:00.040 --> 0:15:02.640 In my case, it's literally that I haven't sat down 0:15:02.640 --> 0:15:05.240 to do any programming since the days when I did 0:15:05.280 --> 0:15:09.960 it an Apple Basic, and I only vaguely remember those days. 0:15:09.960 --> 0:15:12.000 But you know, what if you could interact with a 0:15:12.040 --> 0:15:16.200 machine through the use of natural language, not a programming language. 0:15:16.520 --> 0:15:19.640 What if the computer we're able to take your queries 0:15:19.680 --> 0:15:23.320 and commands that were either written or spoken in everyday 0:15:23.400 --> 0:15:26.080 human language and then suss out what it was you 0:15:26.160 --> 0:15:28.880 wanted and then give it to you. That's the goal 0:15:29.000 --> 0:15:33.520 of natural language processing, and we see it in different implementations. 0:15:33.680 --> 0:15:38.000 Right with chat GBT it's a text based interaction, but 0:15:38.120 --> 0:15:43.040 with smart speakers it's through speaking it into a microphone 0:15:43.200 --> 0:15:46.720 and giving a response. But the basic idea is still 0:15:46.760 --> 0:15:50.160 the same, and at a shallow level, it appears that 0:15:50.200 --> 0:15:52.960 a computer is able to understand you and can respond 0:15:53.000 --> 0:15:55.800 in kind, but in reality, what is going on in 0:15:55.840 --> 0:15:59.600 the background is a very complex analysis to determine what 0:15:59.760 --> 0:16:02.960 it as you are saying or asking or typing or whatever. 0:16:03.640 --> 0:16:06.920 And like a compiler, a natural language processor has to 0:16:06.960 --> 0:16:10.479 identify all the components of a query and to analyze 0:16:10.480 --> 0:16:13.320 the syntax and then respond in a way that's most 0:16:13.360 --> 0:16:15.640 likely to be relevant. For example, if I were to 0:16:15.680 --> 0:16:20.440 ask a smart speaker what's the weather in Walt Disney 0:16:20.440 --> 0:16:24.000 World today, it would have to understand that I've got 0:16:24.240 --> 0:16:26.960 a location I've given It's not my location. I've asked 0:16:27.120 --> 0:16:31.560 for Walt Disney World, I've asked a specific set of 0:16:31.640 --> 0:16:34.240 data what is the weather, and I've given a timeframe 0:16:34.280 --> 0:16:37.640 of today. It would have to understand all that, analyze 0:16:37.680 --> 0:16:42.400 all that, and then get the correct response and present 0:16:42.520 --> 0:16:45.320 it to me, and that's that's incredible, Like, that's an 0:16:45.320 --> 0:16:47.200 incredible amount of work going on in the back end. 0:16:47.400 --> 0:16:51.320 It happens almost instantly when whenever we interact with these systems, 0:16:52.000 --> 0:16:53.680 but it's a lot of stuff that has to happen 0:16:53.800 --> 0:16:57.240 or over that to work. Otherwise we would end up 0:16:57.240 --> 0:16:59.920 with devices that just do irritating things like I imagine 0:17:00.000 --> 0:17:02.440 and telling my smart hub to dim the lights, and 0:17:02.480 --> 0:17:04.359 instead it tells me what the weather is, that the 0:17:04.359 --> 0:17:07.720 Magic Kingdom, and I'm thinking, well, that's nice, but the 0:17:07.800 --> 0:17:10.600 lights are still too bright. Please dim them so I 0:17:10.640 --> 0:17:13.119 don't have to get up off my couch, walk across 0:17:13.160 --> 0:17:18.240 the room and turn a dial, because you know, lazy. Anyway, 0:17:18.480 --> 0:17:21.919 As for natural language processing, that's a simple phrase that 0:17:22.000 --> 0:17:27.080 hides how insanely complicated the actual processes. In reality, natural 0:17:27.119 --> 0:17:31.760 language processing is a multidisciplinary area of development. It incorporates 0:17:31.760 --> 0:17:36.280 elements of artificial intelligence, machine learning, human linguistics. There's a 0:17:36.320 --> 0:17:38.800 bit of psychology that goes in there too. And the 0:17:38.880 --> 0:17:42.440 evolution of natural language processing is a little difficult to trace, 0:17:42.520 --> 0:17:45.520 particularly if you're looking at it from the perspective of 0:17:45.720 --> 0:17:50.119 a user. So way back in the day, kiddos a 0:17:50.160 --> 0:17:54.359 lot of computer games didn't have graphics. They were text 0:17:54.480 --> 0:17:58.040 based games, kind of like a choose your own adventure novel. 0:17:58.400 --> 0:18:02.480 To say, do they still have those? If you're familiar 0:18:02.520 --> 0:18:05.760 with them. You read these books and at the end 0:18:05.760 --> 0:18:08.800 of certain pages you are presented with a choice, and 0:18:09.080 --> 0:18:11.280 it gives you two different page numbers to go to 0:18:11.359 --> 0:18:14.160 depending upon whichever choice you make, and then you continue 0:18:14.200 --> 0:18:16.960 the story from there. Well, a text based adventure was 0:18:17.080 --> 0:18:20.520 very similar to that. There were fewer overt prompts in 0:18:20.560 --> 0:18:24.520 your typical text based adventure. You could theoretically you could 0:18:24.560 --> 0:18:27.440 choose to type in whatever you wanted, so you could 0:18:27.520 --> 0:18:30.080 type commands into a prompt line and then the computer 0:18:30.119 --> 0:18:33.560 game would produce a response. So you might type something 0:18:33.600 --> 0:18:36.679 like look and that would prompt the game to produce 0:18:36.680 --> 0:18:39.800 a description of the environment that you were in at 0:18:39.880 --> 0:18:43.760 the moment. Or you might type inventory to find out 0:18:43.840 --> 0:18:46.199 what stuff you happen to be carrying on your character. 0:18:46.880 --> 0:18:50.880 Or you might type put bit in T in order 0:18:50.880 --> 0:18:53.399 to get the dag nabbed and probability drive working. I 0:18:53.440 --> 0:18:57.480 am still traumatized by the text adventure The Hitchhecker's Guide 0:18:57.480 --> 0:19:02.520 to the Galaxy. Decades later, that game was fiendishly hard 0:19:03.000 --> 0:19:08.680 and non intuitive anyway, then the program would respond appropriately. 0:19:08.720 --> 0:19:11.479 It would give you the response based upon the command 0:19:11.520 --> 0:19:14.160 you typed in, and on a surface level, it looked 0:19:14.160 --> 0:19:18.880 like the computer game understood what you were saying, except 0:19:19.119 --> 0:19:20.840 as soon as you typed in a phrase that the 0:19:20.880 --> 0:19:24.719 programmers hadn't accounted for or just didn't support for whatever reason, 0:19:25.320 --> 0:19:28.399 you would get a pretty standardized message saying something along 0:19:28.400 --> 0:19:31.800 the lines of I'm sorry, I don't understand, and so 0:19:31.880 --> 0:19:35.880 it turned out the game didn't understand you at all. Instead, 0:19:35.880 --> 0:19:39.760 the game had a list of inputs that mapped to 0:19:39.960 --> 0:19:43.720 specific outcomes, and if you provided the input while, you'd 0:19:43.720 --> 0:19:47.719 get the outcome. But anything outside of that list was 0:19:47.800 --> 0:19:49.879 not something the game could handle, and it had to 0:19:49.880 --> 0:19:53.760 give you a response saying I'm sorry, I can't do that, 0:19:54.200 --> 0:19:56.720 or something along those lines. So it gave the illusion 0:19:56.840 --> 0:20:00.320 of understanding, but the player would quickly come to learned 0:20:00.320 --> 0:20:03.359 that there was no such thing actually going on in 0:20:03.440 --> 0:20:07.640 the background. On a similar note, we have chat bots, 0:20:07.680 --> 0:20:10.960 and these have been around for ages, and developers have 0:20:11.040 --> 0:20:14.160 worked for a long time to make chatbots sophisticated enough 0:20:14.560 --> 0:20:17.240 so that you might start to think that maybe the 0:20:17.320 --> 0:20:20.720 chat bot actually understands what you're saying, or maybe there's 0:20:20.760 --> 0:20:23.919 a real human on the other side posing as a 0:20:24.000 --> 0:20:27.600 chat bot. We'll talk about that again in just a second, 0:20:27.600 --> 0:20:40.760 but first let's take another quick break. Okay, so you've 0:20:40.920 --> 0:20:46.000 likely heard of the Turing test, which is this mythical 0:20:46.119 --> 0:20:49.080 test for artificial intelligence and to determine whether or not 0:20:49.119 --> 0:20:52.400 a computer may or may not have sentience or consciousness 0:20:52.520 --> 0:20:57.120 or whatever. Uh That that is really kind of snowballed 0:20:57.160 --> 0:21:00.159 from what it originally was, but the basic idea is 0:21:00.200 --> 0:21:03.160 that it's a take on a game called the imitation game. 0:21:03.960 --> 0:21:08.720 And in this game, a person an interrogator, sits down 0:21:08.760 --> 0:21:13.320 at a computer terminal and they compose questions and they 0:21:13.359 --> 0:21:18.240 get answers displayed on a computer display in front of them, 0:21:18.280 --> 0:21:21.560 and their job is to determine whether or not the 0:21:21.720 --> 0:21:26.080 entity that's creating the responses is another human, or in fact, 0:21:26.200 --> 0:21:28.879 it's a machine that's attempting to pose as a human. 0:21:29.280 --> 0:21:33.480 And if you get to a certain percentage of interrogators 0:21:33.480 --> 0:21:38.760 who cannot be certain or they mistakenly misidentify a machine 0:21:38.760 --> 0:21:41.520 as a human, you would say that machine passes the 0:21:41.560 --> 0:21:46.639 Turing test, and it can convincingly pose as a human. Well, 0:21:47.640 --> 0:21:50.680 we've seen lots of different examples of chat bots that 0:21:50.800 --> 0:21:54.800 have supposedly passed the Turing test. But again, this isn't 0:21:54.920 --> 0:21:58.119 like a solid tests. It's not it's not like the 0:21:58.240 --> 0:22:00.399 S A T S or something. There's not like a 0:22:00.480 --> 0:22:04.040 solid grading structure. It's more it's more interpretive than that. 0:22:05.040 --> 0:22:09.080 But some of the early chatbots we saw do this 0:22:09.160 --> 0:22:13.359 kind of thing by selecting a subset of human behaviors. 0:22:13.400 --> 0:22:16.080 So an example, there were early chatbots or meant to 0:22:16.119 --> 0:22:20.520 simulate someone who had paranoid schizophrenia, or it would pose 0:22:20.560 --> 0:22:23.600 as a therapist, which mostly involved taking whatever it was 0:22:23.680 --> 0:22:26.000 you last said and then turning it into a question, 0:22:26.119 --> 0:22:28.920 which is, why do you think that your coworkers don't 0:22:28.960 --> 0:22:31.639 like you? I gotta say, dr spate, so give me 0:22:31.720 --> 0:22:36.920 a lot of complicated feelings. Anyway, by selecting this subset 0:22:36.920 --> 0:22:40.159 of human behaviors, the programmers are limiting the sort of 0:22:40.200 --> 0:22:42.879 things that the chat bought would be expected to chat about, 0:22:43.440 --> 0:22:46.400 and they would also plant an expectation in the part 0:22:46.480 --> 0:22:50.000 of the human interrogator. It lowers expectations. In other words, 0:22:50.240 --> 0:22:52.919 if you are told that, hey, you're gonna be chatting 0:22:52.920 --> 0:22:57.480 with someone, and it might be a young boy from 0:22:57.520 --> 0:23:00.520 another country who only has a passing understand ending of 0:23:00.680 --> 0:23:05.480 say English, and uh, they are like fifteen. Well, that's 0:23:05.480 --> 0:23:08.480 gonna set your expectations right. You're no longer gonna think, oh, 0:23:08.480 --> 0:23:10.200 this is someone who's going to have a very deep 0:23:10.280 --> 0:23:13.560 knowledge of, say the Vietnam War. That's not gonna happen, 0:23:13.960 --> 0:23:19.679 So there's some leeway there. Uh. Anyway, that's that's one 0:23:19.720 --> 0:23:23.240 of the tricks of creating chat bots that are convincing. 0:23:23.240 --> 0:23:26.400 But these days we're actually seeing much more sophisticated chat 0:23:26.440 --> 0:23:30.879 bots like chat GPT, and they seem to actually understand 0:23:30.920 --> 0:23:33.160 what it is we want, and we can give chat 0:23:33.200 --> 0:23:37.280 GPT a complicated prompt and the program is capable of 0:23:37.320 --> 0:23:41.040 providing a response. So, for example, I actually did this. 0:23:41.160 --> 0:23:44.800 I wrote in the prompt, compose a high KU about 0:23:44.920 --> 0:23:49.000 g p U s as graphics processing units. And this 0:23:49.119 --> 0:23:55.200 is what chat GPT created for me. GPUs speed up 0:23:55.200 --> 0:24:04.920 my code, processing data, lightning, fast, silent, powerful. For now, 0:24:04.960 --> 0:24:07.680 I could point out that this poem does not strictly 0:24:07.720 --> 0:24:11.719 adhere to the structure of hiku because a typical hiku, 0:24:12.160 --> 0:24:14.960 the first and third lines have five syllables, the middle 0:24:15.000 --> 0:24:19.600 line has seven syllables. The hiku quote unquote that chat 0:24:19.600 --> 0:24:24.120 gpt produced had seven syllables, eight syllables, then six syllables. 0:24:24.160 --> 0:24:27.040 So this is not a hiku in the structural sense. 0:24:27.080 --> 0:24:30.320 But you you do see how chat gpt is trying 0:24:30.320 --> 0:24:33.560 to comply with my request. It it's giving something that 0:24:33.640 --> 0:24:37.520 has the feel of a hiku, even though it's not 0:24:38.200 --> 0:24:41.359 strictly speaking a hiku. Now, I've talked in the past 0:24:41.400 --> 0:24:45.120 about how chat gpt pulls data from a huge library 0:24:45.200 --> 0:24:49.119 of information. It is not actively connected to the Internet, 0:24:49.160 --> 0:24:54.199 but instead has this massive repository of information that it 0:24:54.240 --> 0:24:59.119 can pull from, kind of like having a really big encyclopedia, 0:24:59.280 --> 0:25:03.400