1 00:00:07,133 --> 00:00:10,453 Speaker 1: You're listening to the Saturday Morning with Jack team podcast 2 00:00:10,573 --> 00:00:11,453 Speaker 1: from News Talks. 3 00:00:11,453 --> 00:00:15,773 Speaker 2: A'd be Jack, Did you mum write that review? Very droll? No? 4 00:00:16,173 --> 00:00:20,173 Speaker 2: That was Chat GPT five. It has just been released 5 00:00:20,213 --> 00:00:25,173 Speaker 2: by open Ai, their latest GPT model. Paul Stenhouse is 6 00:00:25,173 --> 00:00:27,813 Speaker 2: our text Burton. He's been looking into it, Hey, Paul. 7 00:00:28,133 --> 00:00:30,413 Speaker 3: Yeah, it's good at creative writing, they say, Jack. I 8 00:00:30,613 --> 00:00:31,653 Speaker 3: tend to agree, wouldn't you. 9 00:00:31,733 --> 00:00:33,653 Speaker 2: I just reckon it's maybe a bit over the top, 10 00:00:33,893 --> 00:00:35,973 Speaker 2: like it just needs to tone things down about in 11 00:00:35,973 --> 00:00:37,413 Speaker 2: that case, about seventy percent. 12 00:00:38,733 --> 00:00:40,613 Speaker 3: They also say that they've managed to get it to 13 00:00:40,773 --> 00:00:44,813 Speaker 3: you know, you know, hallucinate less or make things up less. 14 00:00:44,853 --> 00:00:45,693 Speaker 2: Okay, but I don't know. 15 00:00:46,293 --> 00:00:48,573 Speaker 3: Based on what I just heard you read out, Job, I. 16 00:00:48,493 --> 00:00:51,693 Speaker 2: Believe Yeah, I walked into that very good. You know. Well, 17 00:00:51,733 --> 00:00:55,453 Speaker 2: they say that it is a little less sycophantic than 18 00:00:55,453 --> 00:00:58,373 Speaker 2: some of the other GPT models, and I'm not convinced 19 00:00:58,373 --> 00:01:00,973 Speaker 2: that it is in my case necessarily either. But never mind. 20 00:01:01,093 --> 00:01:03,573 Speaker 2: So what does it mean that Chat GPT five is here? 21 00:01:04,333 --> 00:01:06,653 Speaker 3: Yeah? I mean Sir Sam Autmin, who's the c of 22 00:01:06,693 --> 00:01:11,093 Speaker 3: open Ai, called it like a step change. He referred 23 00:01:11,173 --> 00:01:14,173 Speaker 3: to it as you know when your iPhone got the 24 00:01:14,213 --> 00:01:16,893 Speaker 3: retina display or your Samsung got one of their high 25 00:01:16,893 --> 00:01:19,573 Speaker 3: definition screens, and you looked at it and you went, WHOA, 26 00:01:19,733 --> 00:01:22,253 Speaker 3: this looks incredible. And then you looked back at your 27 00:01:22,253 --> 00:01:24,733 Speaker 3: old iPhone and thought, man, did I how did I 28 00:01:24,773 --> 00:01:27,373 Speaker 3: look at this? How did I see all these little 29 00:01:27,413 --> 00:01:30,573 Speaker 3: dots and pixels on the screen. He thinks that once 30 00:01:30,613 --> 00:01:33,453 Speaker 3: you've used chet GBT five you would never want to 31 00:01:33,493 --> 00:01:36,933 Speaker 3: go back to four or any other iteration. He says, 32 00:01:36,973 --> 00:01:40,813 Speaker 3: it's just better. They say that it feels a little 33 00:01:40,813 --> 00:01:41,413 Speaker 3: more human. 34 00:01:42,733 --> 00:01:43,533 Speaker 1: They say it is. 35 00:01:43,973 --> 00:01:47,253 Speaker 3: Apparently great at creative writing and also coding as well. 36 00:01:47,333 --> 00:01:49,573 Speaker 3: That two of the things that it's been kind of 37 00:01:49,653 --> 00:01:53,333 Speaker 3: trained to do and do well. But I think one 38 00:01:53,333 --> 00:01:55,853 Speaker 3: of the things that most importantly it's done, and I 39 00:01:55,853 --> 00:01:58,333 Speaker 3: don't know if what you read out says this or not, 40 00:01:59,053 --> 00:02:03,813 Speaker 3: but they've trained the model to fail, and fail graciously 41 00:02:04,693 --> 00:02:07,373 Speaker 3: because some of the issues with chet GPT four was 42 00:02:07,853 --> 00:02:10,653 Speaker 3: when you actually got it to start doing tasks and 43 00:02:10,693 --> 00:02:14,293 Speaker 3: it couldn't solve those tasks right, especially in the coding world. 44 00:02:14,413 --> 00:02:16,813 Speaker 3: It just took shortcuts. So if you were trying to 45 00:02:16,813 --> 00:02:18,733 Speaker 3: get it to return something from a database and it 46 00:02:18,733 --> 00:02:20,613 Speaker 3: couldn't figure out how to do it. It would just return 47 00:02:20,693 --> 00:02:23,493 Speaker 3: you the same value every time and hard code it, 48 00:02:23,853 --> 00:02:26,653 Speaker 3: which obviously isn't what you wanted it to do. It 49 00:02:26,733 --> 00:02:29,333 Speaker 3: also sometimes would lie about finishing things. 50 00:02:29,693 --> 00:02:32,173 Speaker 2: Yeah right, okay, and so that's not good. 51 00:02:32,213 --> 00:02:34,293 Speaker 3: So they've been really working to make sure that it 52 00:02:34,413 --> 00:02:37,333 Speaker 3: actually fails and when it can't do something, it tries 53 00:02:37,413 --> 00:02:40,413 Speaker 3: to say that it can't do something, yeah, which is 54 00:02:40,493 --> 00:02:41,893 Speaker 3: which is what you can probably want it to do. 55 00:02:42,493 --> 00:02:44,413 Speaker 2: Right, Okay, that's good. I mean that sounds good. I 56 00:02:44,733 --> 00:02:46,573 Speaker 2: saw an example they were doing where they were they 57 00:02:46,573 --> 00:02:49,773 Speaker 2: were basically putting in text prompts into the AI bot 58 00:02:49,893 --> 00:02:52,693 Speaker 2: and asking it to make games, and it seemed to 59 00:02:52,733 --> 00:02:54,413 Speaker 2: be pretty good at that, which is, you know, which 60 00:02:54,453 --> 00:02:57,453 Speaker 2: is impressive. But whether or not, I suppose it's, you know, 61 00:02:57,573 --> 00:03:00,213 Speaker 2: that much better than the last gptwo model. Will wait 62 00:03:00,253 --> 00:03:03,613 Speaker 2: and see. So so check GPT though and open AI 63 00:03:04,213 --> 00:03:06,733 Speaker 2: probably still the leaders when it comes to check bots 64 00:03:06,773 --> 00:03:09,693 Speaker 2: by some significant margin, right Chap. GPT's absolutely huge at 65 00:03:09,693 --> 00:03:10,053 Speaker 2: the moment. 66 00:03:10,573 --> 00:03:13,333 Speaker 3: Oh yeah, oh yeah, I mean they have. Then they 67 00:03:13,453 --> 00:03:18,173 Speaker 3: released the numbers this week seven hundred million weekly users, 68 00:03:18,733 --> 00:03:22,933 Speaker 3: which is just an astronomical number. Hespirts that's getting interesting 69 00:03:22,973 --> 00:03:25,253 Speaker 3: when you're if you're interested in the business aspect of this. 70 00:03:25,373 --> 00:03:30,173 Speaker 3: They have five million paying business users four million developers 71 00:03:30,333 --> 00:03:33,493 Speaker 3: utilizing the API wow, so that's a little bit of 72 00:03:33,533 --> 00:03:37,253 Speaker 3: an interesting ratio with five million paying customers seven hundred 73 00:03:37,293 --> 00:03:41,213 Speaker 3: million weekly users, so no surprise the company is not 74 00:03:41,333 --> 00:03:45,493 Speaker 3: yet profitable. But some other numbers they mention it plans 75 00:03:45,533 --> 00:03:50,693 Speaker 3: to raise from venture capitalists forty billion dollars this year, 76 00:03:51,453 --> 00:03:54,093 Speaker 3: so they obviously need the cash and they but they 77 00:03:54,133 --> 00:03:57,213 Speaker 3: are on pace to pull in revenues of twenty billion 78 00:03:57,253 --> 00:03:59,373 Speaker 3: dollars by the end of the year. So you have 79 00:03:59,453 --> 00:04:01,493 Speaker 3: to kind of do some maths. If they need forty 80 00:04:01,533 --> 00:04:05,773 Speaker 3: million forty billion, they're going to have twenty billion in revenue. 81 00:04:06,413 --> 00:04:11,213 Speaker 3: They must be spending a ton of money. And as 82 00:04:11,253 --> 00:04:13,773 Speaker 3: you go through and use these new models, especially this 83 00:04:13,853 --> 00:04:16,653 Speaker 3: new CHET GPT five, if you've never paid for Chat GPT, 84 00:04:17,053 --> 00:04:20,693 Speaker 3: you've probably never come across a thing called reasoning, which 85 00:04:20,733 --> 00:04:23,053 Speaker 3: basically is where the AI starts to talk to itself 86 00:04:23,093 --> 00:04:25,333 Speaker 3: a little bit, and so it's spends some time thinking. 87 00:04:25,533 --> 00:04:28,213 Speaker 3: But of course that actually makes it kind of more 88 00:04:28,253 --> 00:04:30,933 Speaker 3: expensive because it needs to go back and use more 89 00:04:30,973 --> 00:04:33,373 Speaker 3: tokens and use more computing time and all of these things. 90 00:04:33,373 --> 00:04:36,253 Speaker 3: So I can't imagine their costs are going down anytime soon. 91 00:04:37,813 --> 00:04:41,213 Speaker 3: But I think there's a lot of people who use 92 00:04:41,253 --> 00:04:45,133 Speaker 3: it daily weekly who can't imagine now a life without it. 93 00:04:45,253 --> 00:04:48,453 Speaker 2: Yeah, yeah, too right. I mean I use it in 94 00:04:48,493 --> 00:04:50,453 Speaker 2: all sorts of you know, different parts of my life now. 95 00:04:50,693 --> 00:04:53,973 Speaker 2: It is amazing. It often just it just needs a 96 00:04:54,013 --> 00:04:56,413 Speaker 2: weird tweak from time to time, right, Like it's kind 97 00:04:56,413 --> 00:04:59,533 Speaker 2: of always eighty percent there in my experience and just 98 00:04:59,573 --> 00:05:02,693 Speaker 2: needs a wee kind of human you know component, just 99 00:05:02,973 --> 00:05:04,773 Speaker 2: but see I like that, Yeah, so do I. 100 00:05:04,773 --> 00:05:06,613 Speaker 3: I like that it's not too perfect. No, like that 101 00:05:06,693 --> 00:05:09,293 Speaker 3: it kind of helps refine and then I massage. 102 00:05:09,413 --> 00:05:11,413 Speaker 2: Yeah. Yeah, it's a good way to think about it. Hey, 103 00:05:11,413 --> 00:05:12,133 Speaker 2: thank you so much. 104 00:05:12,173 --> 00:05:12,413 Speaker 3: Paul. 105 00:05:12,493 --> 00:05:14,293 Speaker 2: Paul Steinhouse is our textbook. 106 00:05:14,853 --> 00:05:17,933 Speaker 1: For more from Saturday Morning with Jack Tame, Listen live 107 00:05:18,013 --> 00:05:20,853 Speaker 1: to news Talks ed B from nine am Saturday, or 108 00:05:20,933 --> 00:05:22,813 Speaker 1: follow the podcast on iHeartRadio.