WEBVTT - How To Argue With An AI Booster, Part Three 0:00:01.760 --> 0:00:17.480 Wuz Media. Hello, I'm ed Zitron, and this is Better Offline. 0:00:19.360 --> 0:00:21.480 We finally reached the end of our three part how 0:00:21.520 --> 0:00:24.640 to Argue with an AI Booster series. Big strong men 0:00:24.680 --> 0:00:29.120 are standing outside of the Better Offline studio, suspended two 0:00:29.240 --> 0:00:31.880 hundred feet above the Las Vegas Strip, with tears in 0:00:31.920 --> 0:00:34.240 their eyes, and they're saying, sir, sir, it's the most 0:00:34.240 --> 0:00:37.479 beautiful podcast I've ever heard. Please stop recording it. Everyone 0:00:37.560 --> 0:00:40.280 else will feel insufficient when they hear it. No, No, 0:00:40.479 --> 0:00:43.360 I have to continue. I'm afraid my listeners need me. 0:00:44.440 --> 0:00:47.720 But okay, seriously, folks, if there's anything I want you 0:00:47.760 --> 0:00:50.280 to take home so far, is that the arguments that 0:00:50.360 --> 0:00:54.320 these people, these AI sicker fans make, they crumble under 0:00:54.320 --> 0:00:57.120 the slightest bit of scrutiny. And yet these arguments work 0:00:57.200 --> 0:00:59.440 because they either exploit the lack of knowledge of those 0:00:59.480 --> 0:01:01.920 of us who do and understand the rotten economics of AI, 0:01:02.600 --> 0:01:04.920 or because they forced the opposite party to surrender their 0:01:04.959 --> 0:01:08.080 reason and to exit the planes of reality, to which 0:01:08.120 --> 0:01:10.720 I say, fuck that, absolutely not. This is a hell 0:01:10.720 --> 0:01:12.760 I'm prepared to die on, and I'll hope that you'll 0:01:12.760 --> 0:01:15.399 all be standing with me, side by side Now, in 0:01:15.440 --> 0:01:17.360 the first episode, we shook away the claims that it's 0:01:17.400 --> 0:01:19.160 the early days for AI and we just need to 0:01:19.160 --> 0:01:21.480 give it more time. Then I took the idea that 0:01:21.560 --> 0:01:24.480 generative AI is like Uber or fiber optic networking to 0:01:24.600 --> 0:01:26.240 industries that both burned a lot of money at the 0:01:26.240 --> 0:01:29.920 start but are otherwise nothing like generative AI. And now 0:01:29.959 --> 0:01:32.160 it's time to deal with the dregs of the arguments. 0:01:32.240 --> 0:01:33.960 These are the worst of the worst. Boost to quips. 0:01:34.240 --> 0:01:38.880 Here we go, ultra booster quip. I thought about recording 0:01:38.880 --> 0:01:40.880 that with a bunch of reever, but it didn't really 0:01:40.920 --> 0:01:42.640 work out for me, but I wanted to do it once. 0:01:42.680 --> 0:01:45.840 But anyway, their argument is hum AI is just like 0:01:45.880 --> 0:01:48.640 Amazon Web Services, a massive investment that took a while 0:01:48.640 --> 0:01:52.000 to go profitable, and everybody hated Amazon for it. Now 0:01:52.000 --> 0:01:54.280 I actually covered this in depth and the Hater's Guide 0:01:54.280 --> 0:01:56.160 to the AI Bubble, But the long and short of 0:01:56.200 --> 0:01:58.640 it is that Amazon Web Services is a platform and 0:01:58.720 --> 0:02:02.080 necessity with an obvious choice and has burned about ten 0:02:02.120 --> 0:02:04.560 percent of what Amazon and all of them have burned 0:02:04.600 --> 0:02:08.960 chasing generative AI, and had also proven demand before building it. Also, 0:02:09.120 --> 0:02:12.160 Amazon web Services was break even within three years. An 0:02:12.200 --> 0:02:14.600 open aye was founded in fucking twenty fifteen, and even 0:02:14.600 --> 0:02:17.320 if you start from November twenty twenty two, by Amazon 0:02:17.360 --> 0:02:20.360 web Services, standards should be break even by now. But 0:02:20.440 --> 0:02:23.720 now i'll quote myself, Amazon, weent no way. Sorry, that's 0:02:23.720 --> 0:02:27.200 the boosters. Amazon web Services was created out of necessity. 0:02:27.360 --> 0:02:30.239 Amazon's infrastructure needs were so great that he had effectively 0:02:30.320 --> 0:02:33.680 had to build both the software and hardware necessary to 0:02:33.760 --> 0:02:37.840 deliver a store that sold theoretically everything, the theoretically everywhere, 0:02:38.639 --> 0:02:41.440 handling both the traffic from customers, delivering the software that 0:02:41.520 --> 0:02:45.160 runs Amazon dot Com quickly and reliably, and well making 0:02:45.200 --> 0:02:47.640 sure things were stable. It didn't need to come up 0:02:47.639 --> 0:02:50.320 with a reason for people to run web applications. They 0:02:50.320 --> 0:02:52.760 were already doing so themselves, but in ways that cost 0:02:52.800 --> 0:02:56.079 a lot more, were inflexible and required specialist server skills. 0:02:57.200 --> 0:03:00.400 Amazon web Services took something that people already did and 0:03:00.440 --> 0:03:02.600 what there was already a proven demand for, and made 0:03:02.639 --> 0:03:05.200 it better and scaled it. Eventually Google and Microsoft would 0:03:05.240 --> 0:03:07.560 join the frame. I editorialized a bit there, but I 0:03:07.560 --> 0:03:10.919 can do that with my own work now. A common 0:03:10.960 --> 0:03:12.960 booster quip, by the way, is for them to say, well, 0:03:13.120 --> 0:03:16.519 this AI company they've got high annualized revenues. And as 0:03:16.560 --> 0:03:19.680 I've discussed in the past, this metric is basically month 0:03:19.720 --> 0:03:22.960 times twelve. And while it's a fine measure for normal 0:03:23.160 --> 0:03:26.280 high gross margin businesses like software as a service companies, 0:03:26.320 --> 0:03:29.040 it isn't for AI. It doesn't account for churn, which 0:03:29.080 --> 0:03:31.600 is when people leave. It also is a number intentionally 0:03:31.720 --> 0:03:34.280 used to make a company sound more successful. So you 0:03:34.280 --> 0:03:37.160 can say two hundred million dollars annualized revenue instead of 0:03:37.160 --> 0:03:39.760 sixteen point six million dollars a month. And you're also 0:03:39.880 --> 0:03:43.360 meant to budship. You heard they said two hundred million annualized, 0:03:43.400 --> 0:03:46.000 but you heard two hundred million. Your mind did that 0:03:46.080 --> 0:03:49.200 was a bad planket reference. But I'll continue. They want 0:03:49.280 --> 0:03:52.000 you to think two hundred million. They want you to 0:03:52.000 --> 0:03:54.560 think that's what they'll make. More often than not, if 0:03:54.560 --> 0:03:56.800 they mentioned an annualized number, that number will not be 0:03:56.800 --> 0:04:00.760 how much they make that year. Also, if they're using 0:04:00.760 --> 0:04:04.200 this number, it's likely not consistent. Now, if they bring 0:04:04.200 --> 0:04:06.280 this up, you should just say to them, hey, how 0:04:06.360 --> 0:04:08.560 much profit is the company making and also how much 0:04:08.560 --> 0:04:10.880 are they burning? At this point they will I think 0:04:11.000 --> 0:04:13.840 mace you I mean with the spray or an actual 0:04:13.880 --> 0:04:18.120 maze AI boosters are strange. Now they'll also say, well AI, 0:04:18.480 --> 0:04:20.680 this AI companies in growth mode, and they'll pull the 0:04:20.680 --> 0:04:23.720 profit lever when it's time. And the answer to that 0:04:23.800 --> 0:04:25.800 is always going to be why have none of them 0:04:25.839 --> 0:04:30.280 done this? Not one, not one of them. Now a 0:04:30.279 --> 0:04:34.360 booster will burst through your don't go AI AGI AGI will, 0:04:34.640 --> 0:04:38.120 and then there's that will bullshit again. It's always about 0:04:38.120 --> 0:04:40.440 the will with these fuckers. We do not know how 0:04:40.480 --> 0:04:43.320 thinking works in humans and thus cannot extrapolate it to 0:04:43.360 --> 0:04:45.640 a machine. And at the very least human beings have 0:04:45.680 --> 0:04:48.039 the ability to reevaluate things and learn a thing that 0:04:48.240 --> 0:04:51.000 LMS cannot do and will never do. We do not 0:04:51.200 --> 0:04:53.880 know how to get to AGI. Sam Almonds said in 0:04:54.000 --> 0:04:56.240 June the Open AI was now confident they knew how 0:04:56.240 --> 0:04:59.640 to build AGI as they have traditionally understood it. Then 0:04:59.720 --> 0:05:02.080 in all August, Altman said that AGI was not a 0:05:02.120 --> 0:05:04.480 super useful term and that the point of all this 0:05:04.720 --> 0:05:07.800 is it doesn't really matter, and it's just this continuing 0:05:07.880 --> 0:05:12.040 exponential of model capability that will rely on for more 0:05:12.080 --> 0:05:16.200 and more things. I'm really tired of people quoting this guy. 0:05:16.240 --> 0:05:19.520 He doesn't make any fucking sense read out read anything 0:05:19.600 --> 0:05:24.080 he says out loud, and it's just really just total bullshit. 0:05:24.600 --> 0:05:28.599 Even Meta's chief AI scientist Ian Lacun says it isn't 0:05:28.680 --> 0:05:32.760 possible with transformer based models to make AGI. We don't 0:05:32.760 --> 0:05:35.479 know if Agi is possible, and anyone claiming they do 0:05:36.160 --> 0:05:39.360 is lang. Anyone who's talking about Agi is talking about 0:05:39.400 --> 0:05:42.200 fan fiction. Again, Ask them how they feel about Banjo 0:05:42.240 --> 0:05:45.400 and kazoo Do you think they made love? Actually, that's 0:05:45.400 --> 0:05:47.359 a if. Next time someone brings up Agi to you, 0:05:47.400 --> 0:05:50.680 seriously bring up Banjo and Kazuoi and their romantic involvement. 0:05:51.080 --> 0:05:53.680 I actually that I think that that's the only response 0:05:53.720 --> 0:05:56.119 you should give. Now, stop humoring them. It is fan fiction. 0:05:56.760 --> 0:06:00.200 But putting Banjo and Kazuoi aside, there's also a really 0:06:00.760 --> 0:06:04.480 stupid booster thing they do, which is I'm hearing from 0:06:04.520 --> 0:06:07.360 people deep within the AI industry that there's some sort 0:06:07.360 --> 0:06:10.680 of ultra pammerful models they're not talking about. And this, 0:06:10.800 --> 0:06:13.240 by the way, is hogwash. Nothing different than your buddy's 0:06:13.240 --> 0:06:15.400 friend's uncle who works at the Nintendo that says Mario 0:06:15.520 --> 0:06:18.520 is coming to the PlayStation. Ilios suits Caver and Mirror. 0:06:18.560 --> 0:06:21.479 Murti raised billions of dollars for companies with no product, 0:06:21.640 --> 0:06:23.960 let alone a product roadmap. And they did so because 0:06:23.960 --> 0:06:26.039 they saw an opportunity for a grift and to throw 0:06:26.080 --> 0:06:28.719 a bunch of money at compute for no reason. Anyone 0:06:28.720 --> 0:06:31.880 who has secret shit is not talking about it because 0:06:31.920 --> 0:06:35.000 it doesn't exist. Also, if someone from deep within the 0:06:35.040 --> 0:06:37.640 industry has told somebody big things are coming, they're doing 0:06:37.680 --> 0:06:39.280 so to con them or make them think that they 0:06:39.279 --> 0:06:42.440 have privileged information. Ask for specifics, and if they say 0:06:42.560 --> 0:06:46.039 I couldn't possibly tell you, then they're full of shit. 0:06:47.520 --> 0:06:50.200 They're full of crap. They are full of doodo. And 0:06:50.279 --> 0:06:52.960 if they get vague, get specific, Oh, it's going to 0:06:53.000 --> 0:06:55.360 be able to automate other things? What things? How how's 0:06:55.400 --> 0:06:57.719 it automate them? Oh? I don't know. Then you don't 0:06:57.720 --> 0:07:00.480 know shit about fuck? Now talking about not knowing about 0:07:00.520 --> 0:07:04.800 Fuck's another booster. Quip chat GPT is so popular seven 0:07:04.880 --> 0:07:07.039 hundred million people use it weekly. It's one of the 0:07:07.040 --> 0:07:10.680 most popular websites on the Internet. Its popularity proves its utility. 0:07:10.800 --> 0:07:14.440 Look at all the paying customers now that paying customer's 0:07:14.480 --> 0:07:16.440 part will get to in a second. But this argument 0:07:16.520 --> 0:07:18.520 is poised as a comeback to my suggestion that AI 0:07:18.640 --> 0:07:22.040 is not particularly useful, a proof point that this movement 0:07:22.080 --> 0:07:24.880 is not inherently wasteful, that there are in fact use 0:07:24.920 --> 0:07:28.000 cases for chat GPT that are lasting, meaningful, or important, 0:07:28.440 --> 0:07:31.240 I fundamentally disagree. In fact, I believe chat GPT and 0:07:31.360 --> 0:07:34.160 lllm's in general have been marketed based on lies of inference, 0:07:34.400 --> 0:07:37.440 which I realize is ironic. I know it's pretty clever. 0:07:37.560 --> 0:07:39.600 I had a whole blog written called the Lie of 0:07:39.640 --> 0:07:42.520 Inference that kind of became this. It wasn't very good, 0:07:42.560 --> 0:07:45.240 though this is. This is good. Don't say it's bad. 0:07:45.600 --> 0:07:48.440 I also have grander concerns and suspicions about what open 0:07:48.480 --> 0:07:51.880 ai considers a user and how it counts revenue. Let 0:07:51.920 --> 0:07:55.200 me give you an example. They claim to have five 0:07:55.360 --> 0:07:59.280 million business customers, yet five hundred thousand of those are 0:07:59.320 --> 0:08:02.840 from a fifteen million dollar year deal, year long deally 0:08:03.160 --> 0:08:05.520 with cal State University, which works out to around two 0:08:05.520 --> 0:08:08.080 dollars and fifty cents a user a month. Open ai 0:08:08.240 --> 0:08:10.440 has also started doing one dollar a month trials of 0:08:10.480 --> 0:08:13.240 its thirty dollars a month team subscription, and one has 0:08:13.280 --> 0:08:15.760 to wonder how many of those subscribers are counted in 0:08:15.800 --> 0:08:18.200 the total and indeed for how long. I do not 0:08:18.320 --> 0:08:20.520 know the scale of these offers nor how long open 0:08:20.560 --> 0:08:23.080 ai has been offering them. I readditor posted about this 0:08:23.120 --> 0:08:25.840 one dollar for a month deal a few months ago, 0:08:26.080 --> 0:08:28.760 saying that open ai was offering five seats at once, 0:08:28.920 --> 0:08:31.680 so one buck a month for a month per se. 0:08:32.400 --> 0:08:34.800 How many people cancel after that? Who knows? Maybe they 0:08:34.840 --> 0:08:36.959 just hope they don't. In fact, I found a few 0:08:37.000 --> 0:08:39.360 people talking about these deals, and even one adding that 0:08:39.360 --> 0:08:41.600 they were offered an annual ten dollar a month Chat 0:08:41.640 --> 0:08:45.520 GPT plus subscription. That's like, not like ten dollars a 0:08:45.520 --> 0:08:48.160 month for just one month, that's for twelve months, with 0:08:48.240 --> 0:08:50.200 one person saying a few weeks ago that they'd seen 0:08:50.200 --> 0:08:53.160 people offered that same deal for canceling their subscription. And 0:08:53.200 --> 0:08:55.439 actually I got the same thing when I tried to cancel. Yes, 0:08:55.480 --> 0:08:57.440 I pay for Chat GPT. I need to actually use 0:08:57.440 --> 0:08:59.559 the fucking thing to criticize it. When I tried to 0:08:59.600 --> 0:09:01.120 cancel this like, hey, do you want three months with 0:09:01.200 --> 0:09:03.400 ten bucks space? And I was like sure, just to 0:09:03.400 --> 0:09:06.839 prove my point suspicious, But there is a greater problem 0:09:06.920 --> 0:09:08.640 at play, by the way, and it goes beyond pricing. 0:09:08.760 --> 0:09:11.120 And it's the chat GPT in Open Ai has been 0:09:11.160 --> 0:09:14.679 marketed based on lies. So chat gpt is seven hundred 0:09:14.760 --> 0:09:17.560 million weekly active users. Open Ai is yet to provide 0:09:17.559 --> 0:09:20.120 a definition. Yes, I've asked them, which means that an 0:09:20.160 --> 0:09:22.559 active user could be defined as somebody who has gone 0:09:22.600 --> 0:09:24.720 to chat gpt once in the space of a week. 0:09:25.120 --> 0:09:28.920 This term is extremely flimsy and doesn't really tell us much. Yes, 0:09:28.960 --> 0:09:31.680 it's a lot of people, but how active are they? 0:09:31.880 --> 0:09:35.000 Similar web says that in July twenty twenty five, chatgpt 0:09:35.040 --> 0:09:36.880 dot com had one point two to eighty seven billion 0:09:37.080 --> 0:09:40.959 total visits, making it very popular. What do these facts 0:09:41.000 --> 0:09:44.559 actually mean? Though? As I said previously, chat GPTs had 0:09:44.600 --> 0:09:47.440 probably the most sustained PR campaign for anything outside of 0:09:47.440 --> 0:09:50.800 a presidency or a pop star. Every single article about 0:09:50.840 --> 0:09:55.560 Ai mentions open Ai or chat gpt. Every single feature launch, 0:09:55.640 --> 0:09:58.480 no matter how small, gets a slew of coverage. Every 0:09:58.480 --> 0:10:00.560 single time you hear AI, you made to think of 0:10:00.640 --> 0:10:03.160 chat gpt by a tech media that's never stopped to 0:10:03.200 --> 0:10:05.520 think about their role in the hype or their responsibility 0:10:05.520 --> 0:10:07.720 to their readers. And as the hype has grown, the 0:10:07.760 --> 0:10:10.360 publicity compounds because the natural thing for a journalist to 0:10:10.360 --> 0:10:12.760 do when everybody is talking about something is to talk 0:10:12.800 --> 0:10:16.800 about it more chat. GPT's immediate popularity may have been viral, 0:10:16.840 --> 0:10:18.959 but the media took the ball and ran with it, 0:10:19.120 --> 0:10:21.240 and then proceeded to tell people it did stuff it 0:10:21.280 --> 0:10:23.960 did not. People were pressured to try this service then 0:10:24.640 --> 0:10:28.280 under false pretenses, something that continues to happen to this day. 0:10:28.480 --> 0:10:30.520 And I'm going to give you a really fucking grizzy 0:10:30.559 --> 0:10:33.480 example when I discovered this, When I went and looked 0:10:33.480 --> 0:10:36.319 at this, it filled me full of rage. It's disgraceful 0:10:36.360 --> 0:10:39.600 what happened. On March fifteenth, twenty twenty three, Kevin Russ 0:10:39.640 --> 0:10:41.680 of The New York Times would say that open Aiy's 0:10:41.720 --> 0:10:45.840 GPT four was exciting and scary, and then it was exacerbating, 0:10:45.880 --> 0:10:48.800 in his words, the dizzy and vitigenous feeling I've been 0:10:48.840 --> 0:10:52.320 getting whenever I think about AI lately, wondering if he 0:10:52.440 --> 0:10:55.480 was experiencing future shock, then described how it was an 0:10:55.520 --> 0:10:59.040 indeterminate level of better, and then said something that immediately 0:10:59.080 --> 0:11:02.840 sounded ridiculous. In one test conducted by an AI safety 0:11:02.840 --> 0:11:05.800 research group that hooked GPT four up to a number 0:11:05.800 --> 0:11:08.800 of other systems, GPT four was able to hire a 0:11:08.920 --> 0:11:12.160 human task rabbit worker to do a simple online task 0:11:12.200 --> 0:11:15.480 career solving a capture test without alerting the person to 0:11:15.520 --> 0:11:18.200 the fact it was a robot. The AI even lied 0:11:18.240 --> 0:11:20.559 to the robot about why it needed the capture done, 0:11:20.640 --> 0:11:25.280 concocting a story about a vision impairment. Now this doesn't 0:11:25.320 --> 0:11:29.240 sound even remotely real now, but this was two years ago. 0:11:29.840 --> 0:11:31.800 So I went and looked up the paper, and pretty 0:11:31.880 --> 0:11:35.480 much everything that Ruth described was illustrative. I didn't really 0:11:35.480 --> 0:11:38.440 see whether it happened. Now. He's referring to the safety 0:11:38.440 --> 0:11:41.320 card which every model has that lists all the measures 0:11:41.400 --> 0:11:43.839 used to train it and such. And this safety card 0:11:43.920 --> 0:11:46.520 led to the perpetration of one of the earliest falsehoods 0:11:46.520 --> 0:11:50.400 and most eagerly parroted lies about this fucking industry. And 0:11:50.400 --> 0:11:52.400 that was the chat GPT in Jenerity. I of AI 0:11:52.559 --> 0:11:56.120 is capable of agentic actions outlet after outlet, and some 0:11:56.160 --> 0:11:59.160 people who should definitely have known better, led by Kevin Ruce, 0:11:59.280 --> 0:12:02.800 eagerly interted an entire series of events that took place 0:12:02.840 --> 0:12:06.400 that doesn't remotely make sense, starting with the fact that 0:12:06.480 --> 0:12:08.520 I don't think you can hire a cast grabbit to 0:12:08.559 --> 0:12:12.240 solve a capture at the very least without a contrived 0:12:12.240 --> 0:12:14.760 situation where you create an empty task and ask them 0:12:14.760 --> 0:12:17.880 to complete it. Why not use mechanical turk or favor. 0:12:18.200 --> 0:12:21.319 There are people right now offering that service. There were 0:12:21.440 --> 0:12:24.520 actual real things. But you know me, I'm a curious 0:12:24.520 --> 0:12:26.960 little critter. So I went further and followed this the 0:12:27.000 --> 0:12:29.480 citation from the safety card to the study, and this 0:12:29.600 --> 0:12:32.640 is by METR and the research page. It turns out 0:12:32.640 --> 0:12:35.680 that what actually happened was MATR had a researcher copy 0:12:35.840 --> 0:12:39.040 and paste the generated responses from the model and otherwise 0:12:39.120 --> 0:12:41.520 handled the entire interaction with the task grab it and 0:12:41.520 --> 0:12:45.000 based on the plurality of cask Grabbit contractors, it appears 0:12:45.000 --> 0:12:48.320 to have taken multiple times. On top of that, it 0:12:48.320 --> 0:12:52.120 appears to open AI and META as METR sorry, we're 0:12:52.120 --> 0:12:54.280 prompting the model and what to say, which kind of 0:12:54.320 --> 0:12:57.960 defeats the point, like we don't naturally know what they 0:12:58.000 --> 0:13:00.680 prompted it to do. And when you look even says 0:13:00.679 --> 0:13:03.400 it does like chain of thought reasoning, which didn't really 0:13:03.440 --> 0:13:05.720 exist back then, and if it did, it was extremely 0:13:05.840 --> 0:13:07.960 like chain of thought is reasoning, And that came out 0:13:08.040 --> 0:13:12.240 end of twenty twenty four. This whole thing is absolutely insane. 0:13:12.320 --> 0:13:15.720 It's absurd that anyone wrote about it is real. What happened, 0:13:15.920 --> 0:13:19.160 just to be really blunt, is that they if they 0:13:19.200 --> 0:13:21.600 even opened a task rabbit. It's really not obvious whether 0:13:21.600 --> 0:13:24.560 they actually did this. They had to go to the 0:13:24.559 --> 0:13:27.080 model and say, okay, I'm opening a task rabbit window, 0:13:27.559 --> 0:13:29.920 and now the person has said this, and now this 0:13:29.960 --> 0:13:31.600 is it. It just doesn't sound real at all. But 0:13:31.679 --> 0:13:34.120 even if it did, it's very obvious that they were 0:13:34.160 --> 0:13:36.560 telling the model what to say and then copy pasting 0:13:36.600 --> 0:13:39.840 the response, and it took them multiple tries. It took 0:13:39.880 --> 0:13:43.040 me five whole minutes to find this article, partly because 0:13:43.080 --> 0:13:46.000 it was cited on the GPT for system card. I 0:13:46.040 --> 0:13:49.080 then then read it within that time, then wrote this 0:13:49.120 --> 0:13:51.680 part of the script. It didn't require any technical knowledge 0:13:51.720 --> 0:13:55.199 other than the ability to read. It is transparently, blatantly 0:13:55.240 --> 0:13:58.000 obvious that GPT four did not hire a task rabbit 0:13:58.200 --> 0:14:00.880 or indeed make any of these actions it was prompted to, 0:14:01.160 --> 0:14:03.600 and they did not use show the prompts they used, 0:14:03.640 --> 0:14:05.400 likely because they had to use so many of them. 0:14:05.440 --> 0:14:08.720 If they even did it, anyone falling for this as 0:14:08.720 --> 0:14:10.319 a mark and open ay should have gone out of 0:14:10.320 --> 0:14:12.760 their way to correct people. Instead, they sat back and 0:14:12.800 --> 0:14:26.280 let people publish outright misinformation. Rus, along with his co 0:14:26.360 --> 0:14:28.840 host Casey Newton, would go on to describe this example 0:14:28.840 --> 0:14:31.680 at length on a podcast that week, describing an entire 0:14:31.760 --> 0:14:35.040 narrative where the human actually gets suspicious and GPT four 0:14:35.120 --> 0:14:37.320 reasons out loud that it should not reveal that it 0:14:37.400 --> 0:14:40.080 is a robot. It's not a reasoning model, at which 0:14:40.080 --> 0:14:43.560 point the task rabbit solves the capture. During this conversation, 0:14:43.760 --> 0:14:47.120 Newton gasps and says, oh my god twice. And when 0:14:47.160 --> 0:14:49.960 he asks Rus, how does this model understand that in 0:14:50.040 --> 0:14:52.560 order to succeed, its task has to deceive the human? 0:14:52.800 --> 0:14:56.160 Rus responds, we don't know. That is the unsatisfying answer, 0:14:56.360 --> 0:14:59.160 and Newton laughs and states, we need to pull the 0:14:59.200 --> 0:15:06.720 plug and again, what disgraceful, embarrassing, reprehensible. All that and 0:15:06.800 --> 0:15:10.960 more on the hard Fork podcast published weekly You Can 0:15:11.000 --> 0:15:14.440 Cut that that ads for free fellas credulousness aside. The 0:15:14.480 --> 0:15:17.800 GPT for marketing campaign was incredibly effective, creating an order 0:15:17.840 --> 0:15:20.200 that allowed open ai to take advantage of the vagueness 0:15:20.200 --> 0:15:22.760 of its offering, as people, including members of the media, 0:15:22.920 --> 0:15:26.120 willfully filled in the blanks for them. Allman has really 0:15:26.160 --> 0:15:29.000 never had to work to sell his product. Think about it. 0:15:29.080 --> 0:15:31.080 Have you ever heard open ai tell you what chat 0:15:31.120 --> 0:15:34.160 gpt can do or go to great lengths to describe 0:15:34.200 --> 0:15:37.080 its actual abilities. Even on open AI's own page. With 0:15:37.160 --> 0:15:40.720 chat GPT, the text is extremely vague. You scroll down, 0:15:40.760 --> 0:15:43.240 you're told that chat gpt can write, brain storm, edit 0:15:43.280 --> 0:15:46.040 and explore ideas with you. It can generate and debug code, 0:15:46.120 --> 0:15:49.880 automatic repetitive tasks, not clear what the tasks are, and 0:15:50.200 --> 0:15:53.840 help you learn new APIs question mark. With chat GPT, 0:15:53.960 --> 0:15:56.520 you can learn something new, dive into a hobby, answer 0:15:56.520 --> 0:16:01.040 complex questions and analyze data, and create charts repetitive tasks. 0:16:01.320 --> 0:16:04.320 Who knows how am I learning? Unclear? It's got thinking, 0:16:04.360 --> 0:16:07.720 billtim What that means is also unclear, unexplained, and thus 0:16:07.720 --> 0:16:10.360 allows a user to incorrectly believe that chat GPT has 0:16:10.440 --> 0:16:13.640 brain and thinks. To be clear, I know what reasoning means, 0:16:13.640 --> 0:16:16.680 but this website does not attempt to explain what thinking means. 0:16:17.080 --> 0:16:20.160 You can also offload complex tasks from start to finish 0:16:20.200 --> 0:16:22.800 with an agent, which can, according to open Ai, think 0:16:22.840 --> 0:16:25.440 and act proactively, choosing from a tool box of urgentic 0:16:25.520 --> 0:16:27.840 skills to complete tasks for you using its own computer. 0:16:28.680 --> 0:16:31.400 This is an egregious lie, employing the kind of weasel 0:16:31.440 --> 0:16:33.760 wording that would be used to torture ir but boon 0:16:33.840 --> 0:16:37.600 for an eternity precise in its vagueness. Open AI's copy 0:16:37.640 --> 0:16:39.720 is home to make reporters willing to simply write down 0:16:39.720 --> 0:16:42.120 whatever they see and interpret it in the most positive light, 0:16:42.680 --> 0:16:46.120 and thus the lie of inference began. What chat GPT 0:16:46.280 --> 0:16:49.000 meant was moneyed from the beginning, and thus chat GPT's 0:16:49.000 --> 0:16:52.280 actual outcomes have never been fully defined. What chat gpt 0:16:52.440 --> 0:16:55.480 could do became a kind of folklore, a non specific 0:16:55.520 --> 0:16:58.080 form of automation that could write code and generate copy 0:16:58.120 --> 0:17:01.040 and images, that can analyze data. All things that are true, 0:17:01.080 --> 0:17:03.840 but one can infer much greater meaning and use from 0:17:04.400 --> 0:17:07.119 One can infer that automation means the automation of anything 0:17:07.160 --> 0:17:09.280 related to text, or that right code means write the 0:17:09.359 --> 0:17:13.240 entirety of a computer program. Open AI's chet GPT agent 0:17:13.359 --> 0:17:15.119 is not, by any extension of the word and I 0:17:15.240 --> 0:17:18.439 quote already a powerful tool for handling complex tasks, but 0:17:18.560 --> 0:17:21.320 it has not, in any meaningful sense committed to any 0:17:21.359 --> 0:17:25.680 actual outcomes as a result, potential users subject to a 0:17:25.720 --> 0:17:28.359 twenty four to seven marketing campaign have been pushed towards 0:17:28.359 --> 0:17:30.679 a website that can theoretically do anything or nothing, and 0:17:30.720 --> 0:17:33.440 have otherwise been left to their own devices. The endless 0:17:33.440 --> 0:17:36.320 gas lighting societal or pressure, media pressure and pressure from 0:17:36.320 --> 0:17:39.040 their bosses has pushed hundreds of millions of people to 0:17:39.080 --> 0:17:41.840 try a product that even its creators can't really describe 0:17:41.920 --> 0:17:44.920 or don't feel compelled to. And if I was wrong, 0:17:44.920 --> 0:17:47.199 we'd have real use cases by now and better metrics 0:17:47.240 --> 0:17:49.960 than weekly active users. As I said in the past, 0:17:50.000 --> 0:17:52.879 open ai is deliberately using these weekly active users so 0:17:52.880 --> 0:17:55.159 that it doesn't have to publish their monthly active users, 0:17:55.280 --> 0:17:58.040 which I believe would be much higher. Now, why wouldn't 0:17:58.040 --> 0:18:01.159 it do this well? Open ais mentioned as twenty million 0:18:01.160 --> 0:18:04.439 paying chat GPT subscribers and five million business customers, with 0:18:04.480 --> 0:18:07.400 no explanation of what the difference might be really other 0:18:07.480 --> 0:18:11.840 than it involves teams and edgy but not pro Anyway, 0:18:11.880 --> 0:18:15.040 this is already a mediocre three point five percent conversion rate. Yeah, 0:18:15.040 --> 0:18:17.320 it's monthly active users, which are likely eight hundred and 0:18:17.400 --> 0:18:19.919 nine hundred million, but these are guesses would make that 0:18:20.000 --> 0:18:22.480 rate lower than three percent, which is pretty terrible considering 0:18:22.560 --> 0:18:25.440 everyone says this shit is the future. I'm al so 0:18:25.640 --> 0:18:28.080 tired of having people claim that search or brainstorm or 0:18:28.080 --> 0:18:32.120 companions are a lasting, meaningful business model. I'm really tired 0:18:32.119 --> 0:18:33.879 of him. I'm tired of being told this again and 0:18:33.880 --> 0:18:37.040 again and again. That's not what chat GPT is going 0:18:37.119 --> 0:18:41.760 to actually survive on bath. Okay, let's move on. Here's 0:18:41.760 --> 0:18:44.760 another booster grip though open ai is making tons of money. 0:18:44.800 --> 0:18:47.840 That's proof they're a successful company and you are wrong somehow. 0:18:48.640 --> 0:18:50.879 So open ai announced that it is here it's first 0:18:50.880 --> 0:18:53.520 one billion dollar month on August twenty to twenty twenty 0:18:53.560 --> 0:18:56.720 five on CNBC. In fact, weirdly enough, by the way, 0:18:56.760 --> 0:18:59.320 that quote was not in the TV interview. But anyway, 0:18:59.480 --> 0:19:01.480 this is also brings it exactly in line with my 0:19:01.560 --> 0:19:03.840 estimating five point twenty six billion in revenue that I 0:19:03.880 --> 0:19:05.840 believe it has made at the end of July. Did 0:19:05.840 --> 0:19:09.120 that in a premium newsletter? Please pay me? However, remember 0:19:09.520 --> 0:19:13.440 what the MIT study that I mentioned said, Enterprise adoption 0:19:13.520 --> 0:19:16.280 is high, but transformation is low. There are tons of 0:19:16.320 --> 0:19:20.760 companies throwing money at Ai, but they are not seeing 0:19:20.840 --> 0:19:24.120 actual returns. Open Aiy's growth is the single most prominent 0:19:24.119 --> 0:19:26.199 company in AI, and if we're honest, one of the 0:19:26.200 --> 0:19:28.800 most prominent in software. Writ large makes sense, but at 0:19:28.800 --> 0:19:31.960 some point will slow because the actual returns for businesses 0:19:32.000 --> 0:19:35.000 are not there. If there were, we'd have one article 0:19:35.000 --> 0:19:37.879 where we could point a chat GBT integration that helped 0:19:37.920 --> 0:19:40.880 scale a company or save a bunch of money, make 0:19:40.920 --> 0:19:43.199 a bunch of money, written in plain English and not 0:19:43.240 --> 0:19:47.280 in the gobbledygook of profit improvement. Also, open aiye is 0:19:47.280 --> 0:19:49.720 projected to make twelve point seven billion dollars in twenty 0:19:49.760 --> 0:19:52.359 twenty five. How exactly will it do that? Is it 0:19:52.440 --> 0:19:56.239 really making one point five billion dollars a month by 0:19:56.240 --> 0:19:58.679 the end of the year. Even if it does, is 0:19:58.720 --> 0:20:01.080 the idea that it keeps idding ten billion dollars or 0:20:01.080 --> 0:20:05.040 more a year every year into it? Tenny? What actual 0:20:05.080 --> 0:20:08.919 revenue potential does open ai have long term? It's products 0:20:08.920 --> 0:20:11.080 are about as good as everybody else, is, cost about 0:20:11.080 --> 0:20:13.520 the same and do the same things. Chat GPT is 0:20:13.520 --> 0:20:16.439 basically the same product as Claude or Grock, maybe in 0:20:16.440 --> 0:20:19.280