1 00:00:04,400 --> 00:00:07,800 Speaker 1: Welcome to tech Stuff, a production from I Heart Radio. 2 00:00:11,840 --> 00:00:14,720 Speaker 1: Hey there, and welcome to tech Stuff. I'm your host, 3 00:00:14,880 --> 00:00:18,440 Speaker 1: Jonathan Strickland. I'm an executive producer with I Heart Radio 4 00:00:18,560 --> 00:00:21,520 Speaker 1: and how the tech are you? We're gonna do a 5 00:00:21,800 --> 00:00:25,200 Speaker 1: tech stuff tidbits episode for those of y'all who are 6 00:00:25,280 --> 00:00:28,600 Speaker 1: not familiar wants in a while, I try and do 7 00:00:28,880 --> 00:00:32,760 Speaker 1: kind of a shorter, more concentrated episode on a tech 8 00:00:32,840 --> 00:00:36,360 Speaker 1: topic that doesn't merit a full episode. Now, it's not 9 00:00:36,400 --> 00:00:39,440 Speaker 1: to say that this particular topic doesn't merit a full episode, 10 00:00:39,640 --> 00:00:44,120 Speaker 1: but it will actually require a lot more in depth discussion, 11 00:00:44,360 --> 00:00:46,720 Speaker 1: and I wanted to kind of give an overview just 12 00:00:46,840 --> 00:00:50,360 Speaker 1: to give y'all something to think about. So let's get 13 00:00:50,400 --> 00:00:53,920 Speaker 1: to it. Late last November, the Open Ai organization launched 14 00:00:54,120 --> 00:00:58,000 Speaker 1: chat GPT, which if you haven't used it, I recommend 15 00:00:58,040 --> 00:01:00,520 Speaker 1: you go play with it. Try it out. You can 16 00:01:00,560 --> 00:01:03,440 Speaker 1: make a little account for free and interact with this 17 00:01:03,520 --> 00:01:06,679 Speaker 1: chat bot. So the GPT for chat GPT stands for 18 00:01:06,840 --> 00:01:11,800 Speaker 1: Generative pre Trained Transformer. But it is a chat bot. Now, 19 00:01:11,800 --> 00:01:13,880 Speaker 1: it's a chat bot that can respond to prompts and 20 00:01:13,959 --> 00:01:18,880 Speaker 1: queries with sometimes astonishing results. Uh. This is the tool 21 00:01:19,120 --> 00:01:21,679 Speaker 1: that has certain teachers worried that their students will be 22 00:01:21,720 --> 00:01:25,960 Speaker 1: hoisting all of their homework onto this chat bot and 23 00:01:26,000 --> 00:01:28,039 Speaker 1: make it, you know, write essays for them, and it 24 00:01:28,080 --> 00:01:31,000 Speaker 1: may not always be easy to spot when that happens. 25 00:01:31,360 --> 00:01:34,360 Speaker 1: There are people who are actually offering tools that purport 26 00:01:34,720 --> 00:01:38,560 Speaker 1: to be able to detect if it was generated by 27 00:01:38,680 --> 00:01:41,160 Speaker 1: a chat bot as opposed to written by a student. 28 00:01:41,280 --> 00:01:47,080 Speaker 1: So there's already this kind of seesaw struggle between the 29 00:01:47,160 --> 00:01:49,680 Speaker 1: chat bots and the detectors, which is kind of interesting. 30 00:01:50,160 --> 00:01:53,160 Speaker 1: Some people argue that Pandora's box has been opened and 31 00:01:53,240 --> 00:01:56,280 Speaker 1: chat GPT came roaring out, and it's going to make 32 00:01:56,760 --> 00:01:59,960 Speaker 1: teaching even more difficult and challenging that already is. By 33 00:02:00,040 --> 00:02:02,280 Speaker 1: the way, if you're a teacher, my hat's off to you. 34 00:02:02,480 --> 00:02:05,480 Speaker 1: Both my parents were teachers, they're both retired now. I 35 00:02:05,520 --> 00:02:09,800 Speaker 1: have nothing but the utmost respect for teachers. Y'all have 36 00:02:10,120 --> 00:02:18,600 Speaker 1: a incredibly important, incredibly challenging, and criminally undervalued job. Anyway, 37 00:02:18,680 --> 00:02:21,760 Speaker 1: for chat GPT to work as well as it does, 38 00:02:21,840 --> 00:02:24,080 Speaker 1: there are a ton of different elements that had to 39 00:02:24,120 --> 00:02:28,200 Speaker 1: be implemented. Open Ai had to develop a sophisticated machine 40 00:02:28,280 --> 00:02:31,919 Speaker 1: learning process to train chat GPT, so that it can 41 00:02:31,960 --> 00:02:37,760 Speaker 1: formulate relevant and convincing but not always accurate or correct 42 00:02:38,040 --> 00:02:41,200 Speaker 1: responses to prompts, and relevant is really important. Right. If 43 00:02:41,240 --> 00:02:45,239 Speaker 1: you ask something a question and you get an answer 44 00:02:45,280 --> 00:02:48,959 Speaker 1: that's about something entirely different, that's a really frustrating experience. 45 00:02:49,560 --> 00:02:51,399 Speaker 1: So one thing that the app needs to be able 46 00:02:51,400 --> 00:02:55,440 Speaker 1: to do is parse human language. And this is a 47 00:02:55,520 --> 00:02:59,360 Speaker 1: non trivial engineering problem, and we're going to talk about 48 00:02:59,400 --> 00:03:02,680 Speaker 1: that briefly in this Tech Stuff Tidbits episode. So while 49 00:03:02,760 --> 00:03:06,120 Speaker 1: chat GPT is the launching off point, this is not 50 00:03:06,560 --> 00:03:11,800 Speaker 1: exclusively applicable to chat GPT. It's actually applicable to a 51 00:03:11,880 --> 00:03:15,600 Speaker 1: wide range of applications. So when you hear the phrase 52 00:03:16,000 --> 00:03:20,080 Speaker 1: natural language, that means it's how we humans tend to 53 00:03:20,080 --> 00:03:24,440 Speaker 1: communicate with one another with languages that naturally evolved out 54 00:03:24,480 --> 00:03:30,240 Speaker 1: of culture and society. There are manufactured languages like Esperanto, 55 00:03:30,400 --> 00:03:34,080 Speaker 1: where people created a language with the purpose of creating 56 00:03:34,080 --> 00:03:37,920 Speaker 1: a language, but generally natural language just refers to a 57 00:03:38,080 --> 00:03:43,440 Speaker 1: language naturally evolving over time, and we've developed these over millennia, right, 58 00:03:43,520 --> 00:03:47,320 Speaker 1: And a language includes things like the rules we use 59 00:03:47,440 --> 00:03:50,640 Speaker 1: to form phrases and sentences, as well as the vocabulary 60 00:03:50,760 --> 00:03:55,040 Speaker 1: we use to populate those phrases and sentences. So we've 61 00:03:55,040 --> 00:03:58,840 Speaker 1: got syntax and grammar, and we've got vocabulary. All of 62 00:03:58,880 --> 00:04:02,280 Speaker 1: these elements come together to form language, and you can't 63 00:04:02,320 --> 00:04:05,320 Speaker 1: just throw them in willy nilly, or you will make 64 00:04:05,360 --> 00:04:08,480 Speaker 1: no sense. That's why we have the structure and rules. 65 00:04:09,000 --> 00:04:12,240 Speaker 1: It also includes the quirks we develop over time, you know, 66 00:04:12,280 --> 00:04:17,479 Speaker 1: exceptions to established rules and things like puns and jokes 67 00:04:17,480 --> 00:04:20,200 Speaker 1: and idioms and that kind of thing. Right, we have 68 00:04:20,520 --> 00:04:24,159 Speaker 1: phrases that we use that means something, but they don't 69 00:04:24,240 --> 00:04:27,960 Speaker 1: necessarily mean the surface level. If I say it's raining 70 00:04:28,000 --> 00:04:31,760 Speaker 1: cats and dogs, I probably do not mean that it 71 00:04:32,040 --> 00:04:36,240 Speaker 1: literally is raining small furry animals. If it is, I 72 00:04:36,279 --> 00:04:40,240 Speaker 1: will be inconsolable. I am saying instead that it's raining 73 00:04:40,279 --> 00:04:45,640 Speaker 1: really hard, but I'm using this idiom to express that. Right, So, 74 00:04:45,760 --> 00:04:49,040 Speaker 1: for humans, it's not that difficult to understand what someone 75 00:04:49,200 --> 00:04:53,560 Speaker 1: is saying, assuming you both speak the same language, even 76 00:04:53,920 --> 00:04:56,840 Speaker 1: if that other person is using some unfamiliar words, like 77 00:04:57,080 --> 00:04:59,159 Speaker 1: you could be sitting in on I think back to 78 00:04:59,279 --> 00:05:02,800 Speaker 1: when I was in a class in English class and 79 00:05:02,880 --> 00:05:07,279 Speaker 1: I had a professor who insisted on using the word paradigm, 80 00:05:07,320 --> 00:05:10,320 Speaker 1: and I had never encountered the word paradigm. I it 81 00:05:10,440 --> 00:05:14,520 Speaker 1: was not in my vocabulary, and I picked up in 82 00:05:14,640 --> 00:05:18,599 Speaker 1: context that the way she was specifically using paradigm was 83 00:05:18,680 --> 00:05:21,640 Speaker 1: essentially as a stand in for the word example. Uh. 84 00:05:21,680 --> 00:05:23,680 Speaker 1: This taught me two things. It taught me one of 85 00:05:23,720 --> 00:05:25,520 Speaker 1: the meetings of the word paradigm, and it taught me 86 00:05:25,560 --> 00:05:30,320 Speaker 1: that my professor was incredibly pretentious. Anyway, context really can 87 00:05:30,480 --> 00:05:33,320 Speaker 1: clue us in. Even if we don't know what a 88 00:05:33,360 --> 00:05:37,080 Speaker 1: specific word or phrase means on the surface, from context, 89 00:05:37,160 --> 00:05:40,920 Speaker 1: we can derive at least some meaning. We only still 90 00:05:41,000 --> 00:05:44,560 Speaker 1: miss the full meaning. Or on occasion, you might pull 91 00:05:44,600 --> 00:05:47,520 Speaker 1: a malapropism and you might use the wrong word in 92 00:05:47,600 --> 00:05:50,679 Speaker 1: place of one you intended to use. This has happened 93 00:05:50,720 --> 00:05:54,360 Speaker 1: to me many times. If you want a recent example 94 00:05:54,720 --> 00:05:57,800 Speaker 1: of a character who uses malapropisms, you should watch the 95 00:05:57,839 --> 00:06:01,360 Speaker 1: movie Glass Onion. There is a character in that who 96 00:06:01,480 --> 00:06:05,680 Speaker 1: frequently uses the wrong word to stand in for something 97 00:06:05,720 --> 00:06:09,080 Speaker 1: that he intends to say. Also a fun side note, 98 00:06:09,560 --> 00:06:13,480 Speaker 1: the word malapropism comes from the world of theater. Richard 99 00:06:13,560 --> 00:06:17,080 Speaker 1: Brinsley Sheridan wrote a play called The Rivals. This was 100 00:06:17,120 --> 00:06:20,160 Speaker 1: back in the eighteenth century, so the late seventeen hundreds 101 00:06:20,320 --> 00:06:24,919 Speaker 1: and the rivals includes a character h a a a 102 00:06:24,960 --> 00:06:29,599 Speaker 1: caretaker sort of or a chaperone almost, and her name 103 00:06:29,680 --> 00:06:34,040 Speaker 1: is Mrs Malaprop, and she often employs the wrong words 104 00:06:34,120 --> 00:06:36,679 Speaker 1: to comedic effects. She means one thing, but she says 105 00:06:36,680 --> 00:06:41,240 Speaker 1: another because she's using the incorrect words, which makes the 106 00:06:41,279 --> 00:06:44,200 Speaker 1: meaning of what she's actually saying change and that's where 107 00:06:44,200 --> 00:06:46,080 Speaker 1: the humor is right. And I think this is a 108 00:06:46,080 --> 00:06:50,039 Speaker 1: fun bit of history to know where Malaprop comes from, 109 00:06:50,080 --> 00:06:53,760 Speaker 1: because there's actually another famous character from theater, from English 110 00:06:53,760 --> 00:06:57,880 Speaker 1: theater who predated Mrs Malaprop by a couple of hundred years. 111 00:06:58,760 --> 00:07:03,479 Speaker 1: That character is Ugberry from Billy Shakespeare's Much Ado About Nothing, 112 00:07:04,040 --> 00:07:07,440 Speaker 1: and dog Berry also frequently uses the wrong words in 113 00:07:07,520 --> 00:07:12,320 Speaker 1: an intended comedic effect. The difference is Mrs Malaprop is 114 00:07:12,360 --> 00:07:16,360 Speaker 1: actually funny and dog Berry more often than not isn't 115 00:07:17,280 --> 00:07:20,720 Speaker 1: and mostly being a snobby tease. Here up, please keep 116 00:07:20,720 --> 00:07:23,400 Speaker 1: in mind that back in college I majored in English 117 00:07:23,440 --> 00:07:26,520 Speaker 1: Lit with a focus on Shakespeare. I love Shakespeare dearly, 118 00:07:27,000 --> 00:07:30,360 Speaker 1: but I have issues with some of his comedy. Although 119 00:07:30,360 --> 00:07:34,680 Speaker 1: I guess you could argue the the broad gap in 120 00:07:34,760 --> 00:07:37,280 Speaker 1: time between his time and mind could play a part 121 00:07:37,280 --> 00:07:41,040 Speaker 1: in that anyway. My point is it's not too hard 122 00:07:41,080 --> 00:07:43,960 Speaker 1: for humans to communicate the same thought in lots of 123 00:07:44,000 --> 00:07:46,960 Speaker 1: different ways, and if we encounter a new turn of 124 00:07:47,000 --> 00:07:49,520 Speaker 1: phrase or a new word, we can pick it up 125 00:07:49,760 --> 00:07:54,960 Speaker 1: without too much problem. This stands in stark contrast to 126 00:07:55,120 --> 00:07:59,240 Speaker 1: computers and machines. Now often in the show I talk 127 00:07:59,400 --> 00:08:04,400 Speaker 1: about binary information, about bits and bytes, because when you 128 00:08:04,440 --> 00:08:09,280 Speaker 1: dig way down into how many machines process information, you're 129 00:08:09,320 --> 00:08:13,280 Speaker 1: looking at circuits that run various mathematical operations upon strings 130 00:08:13,280 --> 00:08:17,040 Speaker 1: of data, and that that data is grouped into zeros 131 00:08:17,080 --> 00:08:22,640 Speaker 1: and ones. This is binary or machine language. Computers can 132 00:08:22,680 --> 00:08:26,440 Speaker 1: interpret this quickly. The computers are effectively looking at a 133 00:08:26,520 --> 00:08:30,480 Speaker 1: series of off or on indicators. I often say a 134 00:08:30,680 --> 00:08:33,720 Speaker 1: binary digit or bit is a lot like a light switch. 135 00:08:33,880 --> 00:08:37,080 Speaker 1: It's either off or on. It's zero or one. When 136 00:08:37,080 --> 00:08:40,200 Speaker 1: you put this through logic gates that are have specific 137 00:08:40,200 --> 00:08:44,400 Speaker 1: designs to them, you run them through various operations, these 138 00:08:44,480 --> 00:08:48,480 Speaker 1: zeros and ones can almost magically become complicated processes that 139 00:08:48,880 --> 00:08:51,679 Speaker 1: let you do anything from type out the script for 140 00:08:51,720 --> 00:08:57,040 Speaker 1: a podcast to playing the latest video game. Now make 141 00:08:57,080 --> 00:09:01,840 Speaker 1: sure you keep in mind I said, almost magically in fact, 142 00:09:01,920 --> 00:09:05,360 Speaker 1: it's not magical at all. It's just that we humans 143 00:09:05,400 --> 00:09:08,440 Speaker 1: can't really process huge banks of zeros and ones and 144 00:09:08,520 --> 00:09:11,839 Speaker 1: really make much sense of it. While on a similar note, 145 00:09:12,200 --> 00:09:15,400 Speaker 1: a machine without the proper programming can make no sense 146 00:09:15,880 --> 00:09:21,400 Speaker 1: of our human languages. It is complete gibberish. It's meaningless 147 00:09:21,440 --> 00:09:25,360 Speaker 1: to a computer. Now, it would be exceedingly difficult to 148 00:09:25,440 --> 00:09:29,640 Speaker 1: do any significant kind of programming if programmers had to 149 00:09:29,679 --> 00:09:35,720 Speaker 1: depend upon hard coding zeros and ones while composing their work. Fortunately, 150 00:09:36,080 --> 00:09:42,160 Speaker 1: computer scientists came up with solutions, namely compilers and computer languages, 151 00:09:42,760 --> 00:09:48,120 Speaker 1: and we're gonna start with computer languages. A computer language 152 00:09:48,280 --> 00:09:51,640 Speaker 1: creates levels of abstraction that make it a bit less 153 00:09:51,760 --> 00:09:55,800 Speaker 1: daunting to write software for computers. So the idea is 154 00:09:55,840 --> 00:09:59,720 Speaker 1: that each computer language has its own set of rules, 155 00:09:59,760 --> 00:10:03,480 Speaker 1: like its own syntax, its own vocabulary, and as long 156 00:10:03,520 --> 00:10:05,760 Speaker 1: as you work within those rules, and you do so 157 00:10:05,920 --> 00:10:08,760 Speaker 1: precisely and with as few errors as you possibly can, 158 00:10:09,360 --> 00:10:11,920 Speaker 1: you can create a program to make the computer do 159 00:10:12,000 --> 00:10:14,559 Speaker 1: whatever it is you want it to do. Because the 160 00:10:14,600 --> 00:10:19,160 Speaker 1: computer language itself has built into it the ability to 161 00:10:19,280 --> 00:10:24,079 Speaker 1: be converted into machine language. That will get us two compilers, 162 00:10:24,400 --> 00:10:26,040 Speaker 1: which we will chat about in the minute. In fact, 163 00:10:26,040 --> 00:10:29,440 Speaker 1: we'll talk more about computer languages and compilers after we 164 00:10:29,520 --> 00:10:41,400 Speaker 1: come back from this quick break. Okay, before the break, 165 00:10:41,880 --> 00:10:44,520 Speaker 1: I had introduced this idea of computer languages. Now, not 166 00:10:44,640 --> 00:10:48,199 Speaker 1: all computer languages are equal, and there are dozens of them. 167 00:10:48,480 --> 00:10:51,480 Speaker 1: In fact, there are some notable differences between computer languages. 168 00:10:51,520 --> 00:10:53,760 Speaker 1: Some of them are very similar, some of them share 169 00:10:53,880 --> 00:10:57,560 Speaker 1: a common root computer language, and some of them could 170 00:10:57,600 --> 00:11:01,240 Speaker 1: not be more different. Now, in general, we can separate 171 00:11:01,280 --> 00:11:03,679 Speaker 1: them into very two, very broad categories. There are a 172 00:11:03,720 --> 00:11:07,320 Speaker 1: lot of different ways to categorize programming languages, but we're 173 00:11:07,360 --> 00:11:10,040 Speaker 1: looking at one of the most basic, which is low 174 00:11:10,120 --> 00:11:14,720 Speaker 1: level languages and high level languages. So a low level 175 00:11:14,800 --> 00:11:20,640 Speaker 1: language is relatively close to machine language. It only provides 176 00:11:21,080 --> 00:11:25,680 Speaker 1: a thin layer of abstraction, and as such it can 177 00:11:25,760 --> 00:11:29,320 Speaker 1: still be very challenging for programmers to work with these 178 00:11:29,400 --> 00:11:32,600 Speaker 1: languages because they're not that far off from the basic 179 00:11:32,679 --> 00:11:37,720 Speaker 1: machine languages. Uh, it's a little easier. It's it's designed 180 00:11:37,720 --> 00:11:41,319 Speaker 1: so that humans can interact with it a little more 181 00:11:41,520 --> 00:11:45,199 Speaker 1: naturally than they would with pure machine language, but it's 182 00:11:45,240 --> 00:11:48,320 Speaker 1: not that easy. A high level programming language, on the 183 00:11:48,320 --> 00:11:52,040 Speaker 1: other hand, has a great deal of abstraction, and it's 184 00:11:52,440 --> 00:11:54,960 Speaker 1: much closer to a human language in that way. So 185 00:11:55,080 --> 00:11:58,319 Speaker 1: these languages are far easier for humans to work with 186 00:11:58,400 --> 00:12:00,840 Speaker 1: on a day to day basis, And by humans I 187 00:12:00,880 --> 00:12:03,920 Speaker 1: mean programmers. Like, if you had never seen a computer 188 00:12:04,080 --> 00:12:07,240 Speaker 1: programming language and one day you just sat down to 189 00:12:07,400 --> 00:12:10,559 Speaker 1: work on one, it would not seem easy or intuitive 190 00:12:10,600 --> 00:12:14,079 Speaker 1: to you, probably, But both low level and high level 191 00:12:14,160 --> 00:12:17,600 Speaker 1: languages have their own sets of rules. If you break 192 00:12:17,640 --> 00:12:19,680 Speaker 1: those rules, your program is not going to behave the 193 00:12:19,679 --> 00:12:22,000 Speaker 1: way you intended to. It might not run at all. 194 00:12:22,080 --> 00:12:25,840 Speaker 1: You might just get error messages. But yeah, they have rules, 195 00:12:25,840 --> 00:12:30,839 Speaker 1: and if you follow them, then programs can potentially work. Now, 196 00:12:30,880 --> 00:12:34,839 Speaker 1: if computers process information in zeros and ones, and programming 197 00:12:34,920 --> 00:12:38,520 Speaker 1: languages provide levels of abstraction that approach human language at 198 00:12:38,520 --> 00:12:42,120 Speaker 1: the higher levels, how do we reconcile that? How does 199 00:12:42,240 --> 00:12:47,360 Speaker 1: a computer interpret a program written in uh language like Python? 200 00:12:47,559 --> 00:12:51,720 Speaker 1: For example? Python is much closer to a natural language 201 00:12:51,720 --> 00:12:53,760 Speaker 1: than it is a machine language. So how do we 202 00:12:53,800 --> 00:12:55,800 Speaker 1: get to the point where a computer can take that 203 00:12:55,880 --> 00:13:00,160 Speaker 1: information and actually execute a program? Well, the program I'm 204 00:13:00,200 --> 00:13:03,000 Speaker 1: once composed has to go through what we call a compiler, 205 00:13:03,160 --> 00:13:06,520 Speaker 1: and the compiler's job is to take this program and 206 00:13:06,600 --> 00:13:10,600 Speaker 1: according to the rules of that programming language, convert the 207 00:13:10,640 --> 00:13:14,480 Speaker 1: program from the programming language into machine code so that 208 00:13:14,559 --> 00:13:17,960 Speaker 1: a computer can actually do something with it. So the 209 00:13:18,000 --> 00:13:21,880 Speaker 1: compiler is kind of like a translator. And the compiler 210 00:13:21,880 --> 00:13:23,959 Speaker 1: actually has to do a few jobs to make this happen, 211 00:13:24,040 --> 00:13:27,160 Speaker 1: and has to scan the program's source code for recognizable 212 00:13:27,200 --> 00:13:30,640 Speaker 1: commands and terms. That has to analyze the syntax a 213 00:13:30,760 --> 00:13:34,640 Speaker 1: gave the structure of the code to understand the order 214 00:13:34,679 --> 00:13:37,840 Speaker 1: of operations. That has to break all that down into 215 00:13:37,960 --> 00:13:41,000 Speaker 1: machine code that follows what the language says. And if 216 00:13:41,040 --> 00:13:43,680 Speaker 1: the programmer made a mistake, well that ends up getting 217 00:13:43,880 --> 00:13:46,280 Speaker 1: translated to and then you don't find out un till 218 00:13:46,280 --> 00:13:48,040 Speaker 1: you try and run the program, and then it gets 219 00:13:48,080 --> 00:13:51,280 Speaker 1: back to debugging figuring out where did you make that 220 00:13:51,360 --> 00:13:55,280 Speaker 1: mistake in the actual program. Now here's the thing. The 221 00:13:55,360 --> 00:13:59,280 Speaker 1: programming languages, while they can create high levels of abstraction, 222 00:13:59,679 --> 00:14:02,840 Speaker 1: are still not necessarily accessible to the average person. Like 223 00:14:02,880 --> 00:14:04,880 Speaker 1: I had mentioned before the break, you know, I'm talking 224 00:14:04,880 --> 00:14:07,800 Speaker 1: about the average person who has little to know experience 225 00:14:07,800 --> 00:14:11,280 Speaker 1: with computer languages or programming. So to people like that, 226 00:14:11,360 --> 00:14:14,920 Speaker 1: and I'll include myself here, a sheet of code written, 227 00:14:14,920 --> 00:14:18,280 Speaker 1: and even a very high level computer language might end 228 00:14:18,360 --> 00:14:20,880 Speaker 1: up being indecipherable. You could look at and say, I 229 00:14:20,920 --> 00:14:24,320 Speaker 1: don't know what this program is supposed to do because 230 00:14:24,320 --> 00:14:27,080 Speaker 1: I don't know enough about this programming language to understand 231 00:14:27,120 --> 00:14:30,520 Speaker 1: what any of this means. As such, for folks such 232 00:14:30,520 --> 00:14:34,920 Speaker 1: as myself, programming a computer is a daunting task because 233 00:14:35,040 --> 00:14:38,080 Speaker 1: we lack the basic knowledge of the programming languages we 234 00:14:38,120 --> 00:14:41,760 Speaker 1: would need to use to make an effective program. Now 235 00:14:41,800 --> 00:14:44,280 Speaker 1: you you can teach yourself these things lots of people have. 236 00:14:44,400 --> 00:14:47,920 Speaker 1: In fact, a lot of the most famous hackers and 237 00:14:47,960 --> 00:14:53,479 Speaker 1: even like leaders in in tech business are self taught programmers. 238 00:14:53,800 --> 00:14:56,440 Speaker 1: So there's nothing stopping you from doing this, even if 239 00:14:56,480 --> 00:14:59,560 Speaker 1: you never took a class in computer science or programming. 240 00:15:00,040 --> 00:15:02,640 Speaker 1: In my case, it's literally that I haven't sat down 241 00:15:02,640 --> 00:15:05,240 Speaker 1: to do any programming since the days when I did 242 00:15:05,280 --> 00:15:09,960 Speaker 1: it an Apple Basic, and I only vaguely remember those days. 243 00:15:09,960 --> 00:15:12,000 Speaker 1: But you know, what if you could interact with a 244 00:15:12,040 --> 00:15:16,200 Speaker 1: machine through the use of natural language, not a programming language. 245 00:15:16,520 --> 00:15:19,640 Speaker 1: What if the computer we're able to take your queries 246 00:15:19,680 --> 00:15:23,320 Speaker 1: and commands that were either written or spoken in everyday 247 00:15:23,400 --> 00:15:26,080 Speaker 1: human language and then suss out what it was you 248 00:15:26,160 --> 00:15:28,880 Speaker 1: wanted and then give it to you. That's the goal 249 00:15:29,000 --> 00:15:33,520 Speaker 1: of natural language processing, and we see it in different implementations. 250 00:15:33,680 --> 00:15:38,000 Speaker 1: Right with chat GBT it's a text based interaction, but 251 00:15:38,120 --> 00:15:43,040 Speaker 1: with smart speakers it's through speaking it into a microphone 252 00:15:43,200 --> 00:15:46,720 Speaker 1: and giving a response. But the basic idea is still 253 00:15:46,760 --> 00:15:50,160 Speaker 1: the same, and at a shallow level, it appears that 254 00:15:50,200 --> 00:15:52,960 Speaker 1: a computer is able to understand you and can respond 255 00:15:53,000 --> 00:15:55,800 Speaker 1: in kind, but in reality, what is going on in 256 00:15:55,840 --> 00:15:59,600 Speaker 1: the background is a very complex analysis to determine what 257 00:15:59,760 --> 00:16:02,960 Speaker 1: it as you are saying or asking or typing or whatever. 258 00:16:03,640 --> 00:16:06,920 Speaker 1: And like a compiler, a natural language processor has to 259 00:16:06,960 --> 00:16:10,479 Speaker 1: identify all the components of a query and to analyze 260 00:16:10,480 --> 00:16:13,320 Speaker 1: the syntax and then respond in a way that's most 261 00:16:13,360 --> 00:16:15,640 Speaker 1: likely to be relevant. For example, if I were to 262 00:16:15,680 --> 00:16:20,440 Speaker 1: ask a smart speaker what's the weather in Walt Disney 263 00:16:20,440 --> 00:16:24,000 Speaker 1: World today, it would have to understand that I've got 264 00:16:24,240 --> 00:16:26,960 Speaker 1: a location I've given It's not my location. I've asked 265 00:16:27,120 --> 00:16:31,560 Speaker 1: for Walt Disney World, I've asked a specific set of 266 00:16:31,640 --> 00:16:34,240 Speaker 1: data what is the weather, and I've given a timeframe 267 00:16:34,280 --> 00:16:37,640 Speaker 1: of today. It would have to understand all that, analyze 268 00:16:37,680 --> 00:16:42,400 Speaker 1: all that, and then get the correct response and present 269 00:16:42,520 --> 00:16:45,320 Speaker 1: it to me, and that's that's incredible, Like, that's an 270 00:16:45,320 --> 00:16:47,200 Speaker 1: incredible amount of work going on in the back end. 271 00:16:47,400 --> 00:16:51,320 Speaker 1: It happens almost instantly when whenever we interact with these systems, 272 00:16:52,000 --> 00:16:53,680 Speaker 1: but it's a lot of stuff that has to happen 273 00:16:53,800 --> 00:16:57,240 Speaker 1: or over that to work. Otherwise we would end up 274 00:16:57,240 --> 00:16:59,920 Speaker 1: with devices that just do irritating things like I imagine 275 00:17:00,000 --> 00:17:02,440 Speaker 1: and telling my smart hub to dim the lights, and 276 00:17:02,480 --> 00:17:04,359 Speaker 1: instead it tells me what the weather is, that the 277 00:17:04,359 --> 00:17:07,720 Speaker 1: Magic Kingdom, and I'm thinking, well, that's nice, but the 278 00:17:07,800 --> 00:17:10,600 Speaker 1: lights are still too bright. Please dim them so I 279 00:17:10,640 --> 00:17:13,119 Speaker 1: don't have to get up off my couch, walk across 280 00:17:13,160 --> 00:17:18,240 Speaker 1: the room and turn a dial, because you know, lazy. Anyway, 281 00:17:18,480 --> 00:17:21,919 Speaker 1: As for natural language processing, that's a simple phrase that 282 00:17:22,000 --> 00:17:27,080 Speaker 1: hides how insanely complicated the actual processes. In reality, natural 283 00:17:27,119 --> 00:17:31,760 Speaker 1: language processing is a multidisciplinary area of development. It incorporates 284 00:17:31,760 --> 00:17:36,280 Speaker 1: elements of artificial intelligence, machine learning, human linguistics. There's a 285 00:17:36,320 --> 00:17:38,800 Speaker 1: bit of psychology that goes in there too. And the 286 00:17:38,880 --> 00:17:42,440 Speaker 1: evolution of natural language processing is a little difficult to trace, 287 00:17:42,520 --> 00:17:45,520 Speaker 1: particularly if you're looking at it from the perspective of 288 00:17:45,720 --> 00:17:50,119 Speaker 1: a user. So way back in the day, kiddos a 289 00:17:50,160 --> 00:17:54,359 Speaker 1: lot of computer games didn't have graphics. They were text 290 00:17:54,480 --> 00:17:58,040 Speaker 1: based games, kind of like a choose your own adventure novel. 291 00:17:58,400 --> 00:18:02,480 Speaker 1: To say, do they still have those? If you're familiar 292 00:18:02,520 --> 00:18:05,760 Speaker 1: with them. You read these books and at the end 293 00:18:05,760 --> 00:18:08,800 Speaker 1: of certain pages you are presented with a choice, and 294 00:18:09,080 --> 00:18:11,280 Speaker 1: it gives you two different page numbers to go to 295 00:18:11,359 --> 00:18:14,160 Speaker 1: depending upon whichever choice you make, and then you continue 296 00:18:14,200 --> 00:18:16,960 Speaker 1: the story from there. Well, a text based adventure was 297 00:18:17,080 --> 00:18:20,520 Speaker 1: very similar to that. There were fewer overt prompts in 298 00:18:20,560 --> 00:18:24,520 Speaker 1: your typical text based adventure. You could theoretically you could 299 00:18:24,560 --> 00:18:27,440 Speaker 1: choose to type in whatever you wanted, so you could 300 00:18:27,520 --> 00:18:30,080 Speaker 1: type commands into a prompt line and then the computer 301 00:18:30,119 --> 00:18:33,560 Speaker 1: game would produce a response. So you might type something 302 00:18:33,600 --> 00:18:36,679 Speaker 1: like look and that would prompt the game to produce 303 00:18:36,680 --> 00:18:39,800 Speaker 1: a description of the environment that you were in at 304 00:18:39,880 --> 00:18:43,760 Speaker 1: the moment. Or you might type inventory to find out 305 00:18:43,840 --> 00:18:46,199 Speaker 1: what stuff you happen to be carrying on your character. 306 00:18:46,880 --> 00:18:50,880 Speaker 1: Or you might type put bit in T in order 307 00:18:50,880 --> 00:18:53,399 Speaker 1: to get the dag nabbed and probability drive working. I 308 00:18:53,440 --> 00:18:57,480 Speaker 1: am still traumatized by the text adventure The Hitchhecker's Guide 309 00:18:57,480 --> 00:19:02,520 Speaker 1: to the Galaxy. Decades later, that game was fiendishly hard 310 00:19:03,000 --> 00:19:08,680 Speaker 1: and non intuitive anyway, then the program would respond appropriately. 311 00:19:08,720 --> 00:19:11,479 Speaker 1: It would give you the response based upon the command 312 00:19:11,520 --> 00:19:14,160 Speaker 1: you typed in, and on a surface level, it looked 313 00:19:14,160 --> 00:19:18,880 Speaker 1: like the computer game understood what you were saying, except 314 00:19:19,119 --> 00:19:20,840 Speaker 1: as soon as you typed in a phrase that the 315 00:19:20,880 --> 00:19:24,719 Speaker 1: programmers hadn't accounted for or just didn't support for whatever reason, 316 00:19:25,320 --> 00:19:28,399 Speaker 1: you would get a pretty standardized message saying something along 317 00:19:28,400 --> 00:19:31,800 Speaker 1: the lines of I'm sorry, I don't understand, and so 318 00:19:31,880 --> 00:19:35,880 Speaker 1: it turned out the game didn't understand you at all. Instead, 319 00:19:35,880 --> 00:19:39,760 Speaker 1: the game had a list of inputs that mapped to 320 00:19:39,960 --> 00:19:43,720 Speaker 1: specific outcomes, and if you provided the input while, you'd 321 00:19:43,720 --> 00:19:47,719 Speaker 1: get the outcome. But anything outside of that list was 322 00:19:47,800 --> 00:19:49,879 Speaker 1: not something the game could handle, and it had to 323 00:19:49,880 --> 00:19:53,760 Speaker 1: give you a response saying I'm sorry, I can't do that, 324 00:19:54,200 --> 00:19:56,720 Speaker 1: or something along those lines. So it gave the illusion 325 00:19:56,840 --> 00:20:00,320 Speaker 1: of understanding, but the player would quickly come to learned 326 00:20:00,320 --> 00:20:03,359 Speaker 1: that there was no such thing actually going on in 327 00:20:03,440 --> 00:20:07,640 Speaker 1: the background. On a similar note, we have chat bots, 328 00:20:07,680 --> 00:20:10,960 Speaker 1: and these have been around for ages, and developers have 329 00:20:11,040 --> 00:20:14,160 Speaker 1: worked for a long time to make chatbots sophisticated enough 330 00:20:14,560 --> 00:20:17,240 Speaker 1: so that you might start to think that maybe the 331 00:20:17,320 --> 00:20:20,720 Speaker 1: chat bot actually understands what you're saying, or maybe there's 332 00:20:20,760 --> 00:20:23,919 Speaker 1: a real human on the other side posing as a 333 00:20:24,000 --> 00:20:27,600 Speaker 1: chat bot. We'll talk about that again in just a second, 334 00:20:27,600 --> 00:20:40,760 Speaker 1: but first let's take another quick break. Okay, so you've 335 00:20:40,920 --> 00:20:46,000 Speaker 1: likely heard of the Turing test, which is this mythical 336 00:20:46,119 --> 00:20:49,080 Speaker 1: test for artificial intelligence and to determine whether or not 337 00:20:49,119 --> 00:20:52,400 Speaker 1: a computer may or may not have sentience or consciousness 338 00:20:52,520 --> 00:20:57,120 Speaker 1: or whatever. Uh That that is really kind of snowballed 339 00:20:57,160 --> 00:21:00,159 Speaker 1: from what it originally was, but the basic idea is 340 00:21:00,200 --> 00:21:03,160 Speaker 1: that it's a take on a game called the imitation game. 341 00:21:03,960 --> 00:21:08,720 Speaker 1: And in this game, a person an interrogator, sits down 342 00:21:08,760 --> 00:21:13,320 Speaker 1: at a computer terminal and they compose questions and they 343 00:21:13,359 --> 00:21:18,240 Speaker 1: get answers displayed on a computer display in front of them, 344 00:21:18,280 --> 00:21:21,560 Speaker 1: and their job is to determine whether or not the 345 00:21:21,720 --> 00:21:26,080 Speaker 1: entity that's creating the responses is another human, or in fact, 346 00:21:26,200 --> 00:21:28,879 Speaker 1: it's a machine that's attempting to pose as a human. 347 00:21:29,280 --> 00:21:33,480 Speaker 1: And if you get to a certain percentage of interrogators 348 00:21:33,480 --> 00:21:38,760 Speaker 1: who cannot be certain or they mistakenly misidentify a machine 349 00:21:38,760 --> 00:21:41,520 Speaker 1: as a human, you would say that machine passes the 350 00:21:41,560 --> 00:21:46,639 Speaker 1: Turing test, and it can convincingly pose as a human. Well, 351 00:21:47,640 --> 00:21:50,680 Speaker 1: we've seen lots of different examples of chat bots that 352 00:21:50,800 --> 00:21:54,800 Speaker 1: have supposedly passed the Turing test. But again, this isn't 353 00:21:54,920 --> 00:21:58,119 Speaker 1: like a solid tests. It's not it's not like the 354 00:21:58,240 --> 00:22:00,399 Speaker 1: S A T S or something. There's not like a 355 00:22:00,480 --> 00:22:04,040 Speaker 1: solid grading structure. It's more it's more interpretive than that. 356 00:22:05,040 --> 00:22:09,080 Speaker 1: But some of the early chatbots we saw do this 357 00:22:09,160 --> 00:22:13,359 Speaker 1: kind of thing by selecting a subset of human behaviors. 358 00:22:13,400 --> 00:22:16,080 Speaker 1: So an example, there were early chatbots or meant to 359 00:22:16,119 --> 00:22:20,520 Speaker 1: simulate someone who had paranoid schizophrenia, or it would pose 360 00:22:20,560 --> 00:22:23,600 Speaker 1: as a therapist, which mostly involved taking whatever it was 361 00:22:23,680 --> 00:22:26,000 Speaker 1: you last said and then turning it into a question, 362 00:22:26,119 --> 00:22:28,920 Speaker 1: which is, why do you think that your coworkers don't 363 00:22:28,960 --> 00:22:31,639 Speaker 1: like you? I gotta say, dr spate, so give me 364 00:22:31,720 --> 00:22:36,920 Speaker 1: a lot of complicated feelings. Anyway, by selecting this subset 365 00:22:36,920 --> 00:22:40,159 Speaker 1: of human behaviors, the programmers are limiting the sort of 366 00:22:40,200 --> 00:22:42,879 Speaker 1: things that the chat bought would be expected to chat about, 367 00:22:43,440 --> 00:22:46,400 Speaker 1: and they would also plant an expectation in the part 368 00:22:46,480 --> 00:22:50,000 Speaker 1: of the human interrogator. It lowers expectations. In other words, 369 00:22:50,240 --> 00:22:52,919 Speaker 1: if you are told that, hey, you're gonna be chatting 370 00:22:52,920 --> 00:22:57,480 Speaker 1: with someone, and it might be a young boy from 371 00:22:57,520 --> 00:23:00,520 Speaker 1: another country who only has a passing understand ending of 372 00:23:00,680 --> 00:23:05,480 Speaker 1: say English, and uh, they are like fifteen. Well, that's 373 00:23:05,480 --> 00:23:08,480 Speaker 1: gonna set your expectations right. You're no longer gonna think, oh, 374 00:23:08,480 --> 00:23:10,200 Speaker 1: this is someone who's going to have a very deep 375 00:23:10,280 --> 00:23:13,560 Speaker 1: knowledge of, say the Vietnam War. That's not gonna happen, 376 00:23:13,960 --> 00:23:19,679 Speaker 1: So there's some leeway there. Uh. Anyway, that's that's one 377 00:23:19,720 --> 00:23:23,240 Speaker 1: of the tricks of creating chat bots that are convincing. 378 00:23:23,240 --> 00:23:26,400 Speaker 1: But these days we're actually seeing much more sophisticated chat 379 00:23:26,440 --> 00:23:30,879 Speaker 1: bots like chat GPT, and they seem to actually understand 380 00:23:30,920 --> 00:23:33,160 Speaker 1: what it is we want, and we can give chat 381 00:23:33,200 --> 00:23:37,280 Speaker 1: GPT a complicated prompt and the program is capable of 382 00:23:37,320 --> 00:23:41,040 Speaker 1: providing a response. So, for example, I actually did this. 383 00:23:41,160 --> 00:23:44,800 Speaker 1: I wrote in the prompt, compose a high KU about 384 00:23:44,920 --> 00:23:49,000 Speaker 1: g p U s as graphics processing units. And this 385 00:23:49,119 --> 00:23:55,200 Speaker 1: is what chat GPT created for me. GPUs speed up 386 00:23:55,200 --> 00:24:04,920 Speaker 1: my code, processing data, lightning, fast, silent, powerful. For now, 387 00:24:04,960 --> 00:24:07,680 Speaker 1: I could point out that this poem does not strictly 388 00:24:07,720 --> 00:24:11,719 Speaker 1: adhere to the structure of hiku because a typical hiku, 389 00:24:12,160 --> 00:24:14,960 Speaker 1: the first and third lines have five syllables, the middle 390 00:24:15,000 --> 00:24:19,600 Speaker 1: line has seven syllables. The hiku quote unquote that chat 391 00:24:19,600 --> 00:24:24,120 Speaker 1: gpt produced had seven syllables, eight syllables, then six syllables. 392 00:24:24,160 --> 00:24:27,040 Speaker 1: So this is not a hiku in the structural sense. 393 00:24:27,080 --> 00:24:30,320 Speaker 1: But you you do see how chat gpt is trying 394 00:24:30,320 --> 00:24:33,560 Speaker 1: to comply with my request. It it's giving something that 395 00:24:33,640 --> 00:24:37,520 Speaker 1: has the feel of a hiku, even though it's not 396 00:24:38,200 --> 00:24:41,359 Speaker 1: strictly speaking a hiku. Now, I've talked in the past 397 00:24:41,400 --> 00:24:45,120 Speaker 1: about how chat gpt pulls data from a huge library 398 00:24:45,200 --> 00:24:49,119 Speaker 1: of information. It is not actively connected to the Internet, 399 00:24:49,160 --> 00:24:54,199 Speaker 1: but instead has this massive repository of information that it 400 00:24:54,240 --> 00:24:59,119 Speaker 1: can pull from, kind of like having a really big encyclopedia, 401 00:24:59,280 --> 00:25:03,400 Speaker 1: like horror coded encyclopedia. Think of something that you would 402 00:25:03,400 --> 00:25:07,040 Speaker 1: have in a home library. Actual books. The information in 403 00:25:07,080 --> 00:25:09,720 Speaker 1: those books is not going to change, not frequently, like 404 00:25:09,800 --> 00:25:13,280 Speaker 1: once a year you might get an updated volume that 405 00:25:13,359 --> 00:25:17,480 Speaker 1: gives information about different things that have developed over the year, 406 00:25:17,920 --> 00:25:20,560 Speaker 1: but otherwise, no, it doesn't change. So chat gpt is 407 00:25:20,600 --> 00:25:24,640 Speaker 1: not pulling the most recent information and then serving that up. 408 00:25:24,680 --> 00:25:29,480 Speaker 1: It's it's going to this big library. Now, Unfortunately, that 409 00:25:29,640 --> 00:25:32,760 Speaker 1: library doesn't guarantee that the responses you get are going 410 00:25:32,800 --> 00:25:35,760 Speaker 1: to be accurate. They will appear to at least be 411 00:25:35,880 --> 00:25:40,800 Speaker 1: relevant because chat gpt s programmers were really good at 412 00:25:40,800 --> 00:25:45,199 Speaker 1: having it analyzed queries and to really hone in on 413 00:25:45,359 --> 00:25:49,240 Speaker 1: what was being asked so that chat GPTs response, it's 414 00:25:49,359 --> 00:25:55,360 Speaker 1: generative response would relate to the query. It is one 415 00:25:55,400 --> 00:25:58,520 Speaker 1: common complaint with chat gpt that it presents information in 416 00:25:58,560 --> 00:26:02,159 Speaker 1: such a way as to authoritative though in fact it 417 00:26:02,200 --> 00:26:05,960 Speaker 1: may not quote unquote know what it's talking about. So 418 00:26:06,160 --> 00:26:09,760 Speaker 1: it's working on this very complicated system to parce language 419 00:26:09,760 --> 00:26:13,879 Speaker 1: infirm meaning based upon the words and syntax provided by users, 420 00:26:13,960 --> 00:26:16,720 Speaker 1: and then generate a response following the basic rules of 421 00:26:16,760 --> 00:26:23,520 Speaker 1: grammar and vocabulary well probabilistically picking the most likely response 422 00:26:23,600 --> 00:26:27,080 Speaker 1: to be correct and to be relevant. Again, this is 423 00:26:27,119 --> 00:26:30,560 Speaker 1: something we humans do pretty naturally, but for computers it 424 00:26:30,640 --> 00:26:34,840 Speaker 1: is anything but natural. It required a ton of work 425 00:26:34,840 --> 00:26:37,640 Speaker 1: and evolution to get there. Now, beyond the surface level, 426 00:26:37,680 --> 00:26:40,080 Speaker 1: I feel it's important to say that chat gpt does 427 00:26:40,240 --> 00:26:44,119 Speaker 1: not truly understand what we're saying to it, or what 428 00:26:44,240 --> 00:26:48,120 Speaker 1: it's saying to us. Non the level of deriving meaning 429 00:26:48,560 --> 00:26:51,880 Speaker 1: from it, it's not able to associate different ideas, it's 430 00:26:51,880 --> 00:26:55,240 Speaker 1: not able to come up with something new. It's not thinking. 431 00:26:55,760 --> 00:26:59,439 Speaker 1: It's analyzing and it's responding, and it's doing so in 432 00:26:59,480 --> 00:27:03,440 Speaker 1: a very cool way, but it's not sentient or anything 433 00:27:03,480 --> 00:27:06,400 Speaker 1: like that. Now I will probably do a full episode 434 00:27:06,440 --> 00:27:09,879 Speaker 1: about what's going on behind the curtain with natural language processing. 435 00:27:10,760 --> 00:27:15,240 Speaker 1: It's a pretty challenging topic to cover. It is incredibly complex, 436 00:27:15,760 --> 00:27:22,359 Speaker 1: it is incredibly sophisticated. It requires an interdisciplinary approach that 437 00:27:22,600 --> 00:27:28,639 Speaker 1: is hard to describe easily, and it also typically involves 438 00:27:28,640 --> 00:27:33,200 Speaker 1: several different machine learning strategies that are somewhat challenging to describe, 439 00:27:33,200 --> 00:27:36,600 Speaker 1: particularly without the benefit of visual aids. But I think 440 00:27:36,600 --> 00:27:39,520 Speaker 1: it's worth diving into, and I think we can do it. 441 00:27:39,600 --> 00:27:42,480 Speaker 1: I think we can at least get an appreciation for 442 00:27:42,600 --> 00:27:45,920 Speaker 1: how these systems are working. If nothing else, it can 443 00:27:45,960 --> 00:27:49,040 Speaker 1: remind us that the magic we're experiencing when we tell 444 00:27:49,119 --> 00:27:53,760 Speaker 1: chat GPT to compose, say a punk rock song about 445 00:27:53,840 --> 00:27:58,000 Speaker 1: the iPhone, well we understand that what we see is 446 00:27:58,040 --> 00:28:02,359 Speaker 1: the result of complex sesses and not you know, some 447 00:28:02,440 --> 00:28:05,960 Speaker 1: sort of mystical event. Now. I say that not to 448 00:28:06,040 --> 00:28:09,480 Speaker 1: take anything away from the phenomenal achievements of the hundreds 449 00:28:09,480 --> 00:28:12,240 Speaker 1: of folks who have worked on natural language processing projects, 450 00:28:12,640 --> 00:28:14,840 Speaker 1: but rather to prevent the rest of us from bringing 451 00:28:14,920 --> 00:28:19,560 Speaker 1: meaning where maybe there is no meaning. We don't want 452 00:28:19,560 --> 00:28:22,520 Speaker 1: to project onto this thing. We don't want to make 453 00:28:22,600 --> 00:28:26,239 Speaker 1: assumptions because that could lead us down pathways where we 454 00:28:26,320 --> 00:28:32,399 Speaker 1: start to trust things that are inherently not totally trustworthy. 455 00:28:32,520 --> 00:28:35,359 Speaker 1: One of the issues we've heard multiple times with chat 456 00:28:35,400 --> 00:28:39,040 Speaker 1: GPT is that it's kind of a black box and 457 00:28:39,160 --> 00:28:41,240 Speaker 1: that you ask a question, it gives you an answer, 458 00:28:41,280 --> 00:28:44,680 Speaker 1: but you don't see the process that chat GPT went 459 00:28:44,720 --> 00:28:47,800 Speaker 1: through in order to understand what you were asking and 460 00:28:47,800 --> 00:28:51,840 Speaker 1: then generate the answer that it gives you. And because 461 00:28:52,040 --> 00:28:54,959 Speaker 1: of that, you can't double check its work right. You 462 00:28:55,000 --> 00:28:58,560 Speaker 1: can't check to see what sources did you pull your 463 00:28:58,600 --> 00:29:01,840 Speaker 1: information from to generate your answer, because the sources may 464 00:29:01,920 --> 00:29:05,800 Speaker 1: or may not be reliable. And chat GPT may be 465 00:29:05,800 --> 00:29:08,920 Speaker 1: in a phenomenal tool, but if it's pulling from unreliable resources, 466 00:29:08,920 --> 00:29:10,840 Speaker 1: while the answer you get is still going to be wrong. 467 00:29:11,840 --> 00:29:15,000 Speaker 1: But because chat GPT doesn't really do that, doesn't really 468 00:29:15,400 --> 00:29:18,800 Speaker 1: show its work. Uh. That's where you start to run 469 00:29:18,840 --> 00:29:23,160 Speaker 1: into these problems. And the more dependence you put upon 470 00:29:23,280 --> 00:29:26,080 Speaker 1: these kinds of systems, the more important it is to 471 00:29:26,240 --> 00:29:30,120 Speaker 1: understand how these systems are actually generating the responses. This 472 00:29:30,160 --> 00:29:34,240 Speaker 1: goes beyond chat bots. Obviously, this applies to AI across 473 00:29:34,280 --> 00:29:38,800 Speaker 1: the board. It's an ongoing issue within AI in general. 474 00:29:39,120 --> 00:29:43,320 Speaker 1: Is this desire to make certain that the results that 475 00:29:43,400 --> 00:29:46,160 Speaker 1: AI generates where whatever it may be. Maybe it's facial 476 00:29:46,200 --> 00:29:52,400 Speaker 1: recognition technology, um, maybe it's a robot deciding how to 477 00:29:52,480 --> 00:29:55,560 Speaker 1: open a door. Being able to see that process and 478 00:29:55,640 --> 00:29:59,240 Speaker 1: understand what steps the system went through in order to 479 00:29:59,280 --> 00:30:02,120 Speaker 1: get to its decision are critical in order to be 480 00:30:02,200 --> 00:30:06,320 Speaker 1: able to uh, to judge how well that's that overall 481 00:30:06,400 --> 00:30:11,760 Speaker 1: system works or doesn't work, or if it's reliable or unreliable. Uh. 482 00:30:11,880 --> 00:30:15,280 Speaker 1: That transparency is absolutely necessary for that sort of thing. 483 00:30:15,400 --> 00:30:18,640 Speaker 1: And frankly, a lot of the systems we encounter today 484 00:30:18,680 --> 00:30:21,040 Speaker 1: have a lack of transparency and that makes it kind 485 00:30:21,080 --> 00:30:25,040 Speaker 1: of scary. However, that being said, I think chat GPT 486 00:30:25,320 --> 00:30:28,840 Speaker 1: is a really really cool project. I do share the 487 00:30:28,880 --> 00:30:32,600 Speaker 1: concerns of people relying upon it to do work that 488 00:30:32,640 --> 00:30:35,400 Speaker 1: they should be doing. Um, I think that's I think 489 00:30:35,400 --> 00:30:38,640 Speaker 1: they're cheating themselves. If you aren't doing the work, then 490 00:30:38,680 --> 00:30:40,760 Speaker 1: you're not learning how to think. Which is the most 491 00:30:40,760 --> 00:30:44,800 Speaker 1: important lesson you can learn in your education is learning 492 00:30:44,800 --> 00:30:48,320 Speaker 1: how to actually think and to think critically, and if 493 00:30:48,360 --> 00:30:50,840 Speaker 1: you deny yourself that, then you just set yourself up 494 00:30:50,880 --> 00:30:54,240 Speaker 1: to be led around by the nose by anyone who 495 00:30:54,280 --> 00:30:58,840 Speaker 1: has a convincing enough story, and that rarely turns out well. 496 00:30:59,640 --> 00:31:01,760 Speaker 1: At least it doesn't turn out well except for the 497 00:31:01,760 --> 00:31:04,720 Speaker 1: person who's doing the leading, and even they tend to 498 00:31:05,200 --> 00:31:07,680 Speaker 1: come to a bad ending once it's all said and done. 499 00:31:08,520 --> 00:31:12,760 Speaker 1: All right, that's it. Hope you enjoyed this tech stuff 500 00:31:12,800 --> 00:31:16,520 Speaker 1: tidbits about natural language processing, specifically within the context of 501 00:31:16,600 --> 00:31:18,920 Speaker 1: chat GPT. Like I said, we'll have to do a 502 00:31:19,000 --> 00:31:22,680 Speaker 1: much deeper dive because this isn't even really scratching the 503 00:31:22,720 --> 00:31:26,840 Speaker 1: surface like that. This is such a deep dense topic 504 00:31:27,320 --> 00:31:29,400 Speaker 1: that we could do a couple of episodes about it 505 00:31:29,440 --> 00:31:32,920 Speaker 1: and really kind of explore it. Uh. I would even 506 00:31:32,960 --> 00:31:35,800 Speaker 1: reach out to experts to have on the show to 507 00:31:35,880 --> 00:31:38,600 Speaker 1: talk with them about it. The only issue there is 508 00:31:38,640 --> 00:31:43,760 Speaker 1: that I would worry very quickly that experts would use 509 00:31:44,760 --> 00:31:49,000 Speaker 1: terminology and jargon that I'm not familiar with, let alone 510 00:31:49,080 --> 00:31:51,360 Speaker 1: some of my listeners, some of y'all are way ahead 511 00:31:51,360 --> 00:31:53,960 Speaker 1: of me on this stuff, and that's awesome. Some of 512 00:31:54,080 --> 00:31:55,920 Speaker 1: you all are probably in the same position that I'm in, 513 00:31:56,400 --> 00:31:59,320 Speaker 1: where you know, you could hear someone spelled off a 514 00:31:59,320 --> 00:32:01,760 Speaker 1: lot of jargon and not know what the heck they 515 00:32:01,760 --> 00:32:06,280 Speaker 1: were saying. Context be darned, it would just not be 516 00:32:06,920 --> 00:32:12,160 Speaker 1: be accessible to you, uh or to me. And so yeah, 517 00:32:12,280 --> 00:32:13,680 Speaker 1: I just got to make sure that if I pick 518 00:32:13,760 --> 00:32:17,680 Speaker 1: someone who is an expert, they're also a great communicator 519 00:32:17,960 --> 00:32:23,040 Speaker 1: for that kind of stuff. Uh. Some engineers are phenomenal communicators, 520 00:32:23,160 --> 00:32:28,840 Speaker 1: and some are phenomenal engineers who can make machines sing, 521 00:32:29,160 --> 00:32:32,720 Speaker 1: but may not be able to uh talk with other 522 00:32:32,800 --> 00:32:35,520 Speaker 1: human beings in a way that the other human beings 523 00:32:35,520 --> 00:32:38,880 Speaker 1: can follow. It's it's more that, you know, I lack 524 00:32:39,720 --> 00:32:44,000 Speaker 1: that experience and understanding to do it confidently and accurately. 525 00:32:44,320 --> 00:32:47,720 Speaker 1: But yeah, we'll do more about natural language processing. I 526 00:32:47,720 --> 00:32:52,440 Speaker 1: didn't even really touch on the spoken language processing stuff, 527 00:32:52,480 --> 00:32:54,920 Speaker 1: you know, like things like speech to text and that 528 00:32:55,000 --> 00:32:59,600 Speaker 1: kind of thing. That stuff is also incredibly complex. It 529 00:32:59,640 --> 00:33:05,240 Speaker 1: adds other layers of complexity upon this system. UM. I 530 00:33:05,320 --> 00:33:07,360 Speaker 1: have talked about those in the past. I've done episodes 531 00:33:07,400 --> 00:33:09,600 Speaker 1: where I've talked a bit about, you know, like things 532 00:33:09,640 --> 00:33:13,000 Speaker 1: like like voice recognition and those sorts of things, and 533 00:33:13,040 --> 00:33:16,959 Speaker 1: speech to text but yeah, that's also important to remember 534 00:33:17,000 --> 00:33:20,800 Speaker 1: that that adds yet another layer. Well that's it. I 535 00:33:20,880 --> 00:33:24,520 Speaker 1: hope you're having a great three so far. I mean 536 00:33:24,520 --> 00:33:27,400 Speaker 1: we're four days into it. I hope things haven't gone 537 00:33:27,520 --> 00:33:30,840 Speaker 1: pear shaped already. And if you have any suggestions for 538 00:33:30,880 --> 00:33:32,920 Speaker 1: topics A should cover in future episodes of tech Stuff, 539 00:33:32,920 --> 00:33:34,360 Speaker 1: there a couple of ways you can reach out to me. 540 00:33:34,400 --> 00:33:36,800 Speaker 1: One of those is to download the iHeart Radio app. 541 00:33:37,000 --> 00:33:40,080 Speaker 1: It is free to download into use. You can go 542 00:33:40,120 --> 00:33:43,040 Speaker 1: to that little searchbar type in tech Stuff. It'll take 543 00:33:43,080 --> 00:33:45,440 Speaker 1: you to our page. You'll see there's a little microphone 544 00:33:45,560 --> 00:33:48,120 Speaker 1: icon there. If you click on that, you can leave 545 00:33:48,120 --> 00:33:51,760 Speaker 1: a voice message up to thirty seconds in lengths say hi, 546 00:33:52,400 --> 00:33:54,520 Speaker 1: tell me kind of what topics you would like to 547 00:33:54,520 --> 00:33:57,880 Speaker 1: hear about more. If you prefer, you can pop on 548 00:33:57,920 --> 00:34:01,080 Speaker 1: over to Twitter and send me a message there. The 549 00:34:01,080 --> 00:34:04,080 Speaker 1: the Twitter handle for the show is tech Stuff H 550 00:34:04,200 --> 00:34:11,800 Speaker 1: s W and I'll talk to you again really soon, y. 551 00:34:13,360 --> 00:34:16,399 Speaker 1: Tech Stuff is an I Heart Radio production. For more 552 00:34:16,480 --> 00:34:19,839 Speaker 1: podcasts from I Heart Radio, visit the i Heart Radio app, 553 00:34:20,000 --> 00:34:23,160 Speaker 1: Apple Podcasts, or wherever you listen to your favorite shows.