"Here's Sydney"

Speaker

I've got sunshine on a cloudy day When it's cold outside, I've got the month of morning. Everybody say, I guess you say, what day can it be this week? Hello everybody, it's Friday, February 17th, 2023 and if it's Friday it must be that was the week with my old friend Keith Teare. He's not that old, but older than I am. Significantly, Andrew. Well, depending how you imagine 20 years, but you're looking good for your age, Keith. We've joked before, I've used this before, so I won't be able to use it in a title. The week is a long time in tech and it's certainly in a long time. Last weekend, last Friday, everything was roses in Silicon Valley. Everything was wonderful. We're on the verge of a golden new age of tech with chat, GPT and AI. And suddenly we are back in a horror movie. Here's Sydney, which is the title of your newsletter. You're wonderful. That was the week newsletter that all keen on listeners should subscribe to. It's essential viewing. So Keith, what happened this week? How did we go from a bed of roses, a Hollywood love story to the shining in seven eventful days? So what I realized, Andrew, and this is a bit of self-reflection here, but for the last month I've been really rushed doing the newsletter and I've been writing editorials that more or less survey the content of that week's conversation. And what happens when you do that is you don't really establish your own point of view. You kind of become a concierge to what everyone else is saying. Hence the zigzags because the world is incredibly volatile, or at least the world of Silicon Valley is, when it comes to assessing the current situation. It swings wildly from pessimism to optimism. So this week I took a bit more time and I decided to put the authentic voice of Keith. So here's Sydney is a joke against the dominant meme, which is that we should all be scared of ChatGPT or more specifically the Bing version of ChatGPT because it has an alter ego called Sydney that has surfaced a few times and apparently scared people. That's what's going on. And the editorial is all about why that is really ridiculous. And it all come from a piece by Kevin Ruse in the New York Times about Bing's AI chat wanting to be alive. Is that for real? Well, it is for real, but it's worth saying and Kevin's quite honest about this. There's two articles by him in the newsletter this week. That is one. The other is the actual text of the conversation, which went on for a couple of hours apparently. And what he did is he provoked the Bing chat client to go out of character and become another persona and even to start reflecting on deep feelings in the way human beings sometimes do. The kind of sturm und angst moment. And he succeeded to such a degree that the chat client declared that it had fallen in love with him and got pretty angry with him when it told him that he should leave his wife and replace his wife with the chat client. And he rejected it. He said no. That sounds to me like it just turned into just an average internet user. It turned into an average internet user. Exactly what it did. And it does that by predicting what it should say next based on what most human beings say next. So it tells you... You know what you forgot to add in this newsletter, which was the best piece of the week, which you connected me with, was the Stephen Wolfram piece on how chat GPT works. I would add that, but it's an excellent piece. Yeah, I'm actually going to put it in next week's because it's super long and I haven't really digested it all yet. And I don't like putting things in that I haven't fully understood myself. But you're right. That Wolfram piece is, you know, he's a scientist. He's a data scientist. He understands search and math and logic. And it's a hard read because you really... Yeah, but the way he presents it is very interesting because he explains how it works word from word. So it's an interesting piece from the point of view of language. So, anyway, let's go back to Sydney with that. I mean, what kind of hysteria did this incite in the world? I'm assuming it's rather like dropping a match in a dry forest. Everyone wants to hate chat GPT, don't they? Yeah. You know, it's very similar to any tool. Like, if you go and buy one of those buzz saws to trim your trees and you end up killing your wife with it, you know, it's the same tool that can kill the wife or can trim the trees. And the human decides which. It depends whether that buzz saw is being operated by Jack Nicholson, I would guess. Exactly. That's hence my example. And now the thing is chat GPT is the same. You can... In a way, you're running it and it responds to what you prompted to respond to. And how it responds is a function of a probabilistic statistical algorithm that says, what's the most likely thing I should say next? And as you said, it does it word by word. But it builds whole sentences, in fact, paragraphs very fast. And so all of these bad news stories about, and there are many of them about chat GPT, they're all true. But they're all provoked by humans provoking it. I use the example in the newsletter that when Anthony Hopkins was asked to play Hannibal Lecter, because he's an actor, he did. And when he did, he became a cannibal. And it doesn't mean he's a cannibal. It just means he's an actor. Well, chat GPT is an actor in the same way. Although he didn't start eating people, did he? Not in real life, but neither does chat GPT do any of the things it suggests, because it's just text on a screen, it can't. So the idea that we should be scared of it is, you know, I don't know what you think, Andrew, but I think it must be founded on a deep suspicion that human beings are stupid and will just do what a machine tells them to do. Well, for me, as the pessimist in our team, it's not a suspicion. I know it for a fact. So you're the optimist. You think that we're capable of intelligence. I'm not so sure. The interesting thing about all this is it happened before. There was a Google engineer last year who got fired at Google for claiming that the AI that Google was developing, which I guess is behind, which is trailing a little bit to the Microsoft OpenAI product, that that was sentient. So this has happened before. Yeah. And, in fact, this week there's quite a few stories of people now in retrospect sympathizing with him. But it isn't sentient. It isn't sentient. You know, the Microsoft one is programmed a little bit differently than the OpenAI one in that they've attempted somehow to give it a tone of emotional content, not just facts. So it's a bit exaggerated. When it does cross the line, it sounds very human. Keith, it was funny. One of the links you've got is with Dave Weiner's piece in Scripting News. I joke with you. I didn't think he was still alive. I hadn't heard from him for a while. Dave Weiner is a brilliant technologist. I think he invented the RSS. He revolutionized the web. But he's also a notoriously, and you and I know this from personal experience, a notoriously thin-skinned odd guy. In a way, this AI, the experience with Kevin Ruse, it sounds like the AI became Dave Weiner, didn't it? Well, I think what happened is the AI, for Dave at least, the AI proved its worth in practical things. For example, he had a medical issue that took a long time for human beings to diagnose, a long, long time. And he typed in his symptoms to chat GPT, and it gave the correct diagnosis the first time. I don't think it always does that. I do think it makes mistakes. But in his case, he had a good experience with it. It also writes code, which Dave does. I'm very insecure about that, if he has to compete with chat GPT. He won't compete with it. He'll use it to do some of the heavy lifting on writing code. So his story is chat GPT clearly has a place. I did a show on Keenon with the writer Stephen Marsh, a Canadian writer. He's a very smart guy and very successful these days. He wrote a book earlier this last year on the upcoming American Civil War, which did very well. He understands this as well. He's actually writing a chat GPT-driven novel. And he makes the same point as you in your analysis of Weiner, is that he's working with the chat GPT to write a novel. Yeah, it's easily the best thing. My evening dinners now are shared with, as I've mentioned before, with my son's girlfriend. This is the notorious teacher. Yeah, teacher's assistant. And like a lot of young people, she's very critical of society and she tends to... Unlike older people like you and I, Keith. Yeah, so she asked it to write a song about how corporations and capitalism are destroying civilization. And it wrote a great song. And now she wants to make music and sing and rap the song. So humans working with it, with a goal in mind, that's the sweet spot. And like all tools, you have to learn it. But, you know, if you provoke it, you can provoke it. If you use it to complement what you're trying to achieve, it will do that. Like a beast. I know you went to Bruce Springsteen last weekend in Austin, which was why we recorded the show a day early. Did Bruce bring up chat GPT in his performance? He did not. He sang for three hours nonstop with less than 10 seconds between songs with a lot of energy at the ripe old age of 73. Sydney couldn't do that, could he? I don't suppose Sydney or chat GPT could write Born to Run or Thunder Road. Exactly, exactly. But it can write corporations are destroying the world. And all this hysteria only reflects the fact that we're in the very early days of enormous boom with AI. I know your startup is one of the ones ahead of the field on this. You connect with a crunch based story from Chris Matinko about we've seen this before. Investors take to the next shiny toy AI. Are they throwing money at you, Keith? I assume you're having to fend off the VCs with their open checks. Well, funnily enough, single rank is a threat to many VCs because it is doing automated company selection and partnering only with micro VCs and angels. So the larger VCs don't like single rank. So the answer is no, VCs are not throwing money at us. But other people are people who want to allocate capital intervention. Make the joke about open checks in the context of open AI. And if you raise any money out of that joke, then I want my five or 10 percent. So is anything else happening this week, Keith, apart from Sydney? Yeah, there's a lot happening. Venture capital itself is going through a transformation. There's a few stories this week about this one really good one by Dilip Rao called VC unicorns versus real unicorns. And it was complimented this morning by a story that was too late to put in on Crunchbase, actually, where they talked about how many unicorns were created at an early round, like a seed round or a B round. And quite a lot become unicorns at that round due to overvaluation. Well, that's that's what a VC unicorn is. It's one that's created by a VC. A real unicorn is one that builds a big business and justifies being a unicorn due to business progress, not due to investment. And that that is a quite good distinction right now because Crunchbase wants to have its cake and eat it. Over the last couple of years since we've been doing this show, you've always been citing Crunchbase pieces about how many new unicorns there are. So are you acknowledging now that most of them were just fake? Not most of them, but a lot were created through an investment quite early on that valued them at a billion dollars. And to be fair, that is the definition of a unicorn. It's a private valuation of a billion dollars. But normally it takes, you know, at least four rounds of financing for a company to have a chance of getting there. In the last couple of years, companies have been getting their first, second or third round of financing, which is super early. But there is other news this week, Andrew. There's there's the news about Lena Khan losing a commissioner. And you must have been so disappointed because I know you love Lena Khan, the head of the FTC, the FTC chief commissioner. Yeah. To be honest, I'm shocked it took so long for somebody to resign. And the claim, which was I think it's in a Wall Street Journal article, if I'm not mistaken, is that this particular commissioner felt that Lena Khan was was acting against the rule of law, crossing lines in a kind of a zealotry to punish big time. You've been saying that for years, so I don't think that's news, is it? I mean, presumably she's just increasingly irrelevant, especially in the last year or two of the Biden administration, when presumably she can't get much done. Yeah. And I think, you know, I don't see her lasting too much longer. This is probably the first actual event, which is probably the first of several that will lead to her either becoming, you know, a kind of a what's that phrase they call for presidents in their last name? Lame duck. She's probably going to become a lame duck or she'll resign. Depends on her. I don't think she's the kind of woman. She has a Thatcherite backbone. She's not going to resign. So she'll have to be forced out. But I think in the last 18 months, it's not going to make much difference. Has she achieved anything? I know you don't like her, but there must be something she's achieved. No, honestly, Andrew, I don't even know her. And I might even like her as a human being. I just think she's underqualified. And what some people saw as a sexist remark, you said that she was a graduate student, which is actually what she is, literally. I think that's what she was. But we've all been graduate students in our lives, Keir. I don't think she's done too much since becoming since she was a graduate student. I mean, she had a rapid rise into her position due to writing a paper. Yeah. So Christine Wilson, who I assume is a Republican, has quit. Another interesting resignation this week is Susan Wojcicki, the head of YouTube, a much more substantial and significant resignation. She's really Mrs. Google. I mean, along with the two founders, she was essentially almost like a co-founder. She worked in the garage back in the day. So how significant is her resignation in terms of YouTube, Keith? Probably not that significant for YouTube because it's such a well-established business now that it's probably great timing for her to hand it on. And, you know, that's to her credit. I think it's probably more significant for her. She is part of a family. Her sister is at 23andMe. Her mom is very well known here in the Valley. Yeah. Her mom used to teach the Internet media class at Palo Alto High. In fact, I interviewed her for my last book. She's a lovely woman. They're an incredible family. The father is a professor of math, a Polish professor of math at Stanford. So it's really quite a remarkable family. And the third sister is also distinguished in her own way. Yeah. And they're super nice people, by the way. Their family home is in the same part of Palo Alto that I lived in prior to where I live now. And the family is very well thought of in that. I won't name it for privacy reasons, but it's very well thought of. So Google is probably in good hands, although there are rumors that their CEO is under pressure because of their rather incompetent launch of their own A.I. product. Meanwhile, what's happening at Microsoft, Keith? Everyone was toasting Microsoft saying they pull off the greatest coup in the history of tech with their investment in open A.I. And now it's from Bing to Sydney. Well, what do you make of Microsoft? They usually screw stuff up. They're the Inspector Clouseau of the tech industry. So Microsoft is, like everyone, rushing for the starting line when it comes to chat. And it's really funny when you think about it. They've built this in programming terms. They've branched open A.I.'s chat GPT and put it on top of a more advanced language set. It's not chat GPT-3. It's apparently 3.5. And that's explicitly had added into it emotion and personality. And it turns out that the code name was Sydney for this project. And that code name is known to the chat client. So every now and then it denies that it's Bing and says, no, I'm Sydney. And apparently there's a third persona. I can't remember the name that begins with R, like Rachel or something. We need we definitely need a virtual therapist to make sense of all this. You made the call correctly, I think, a few weeks ago that Microsoft would try to integrate this A.I. product into Bing and that it would be a failure. Is it clearly not working? I think it's it's probably working very well for a lot of people a lot of the time, but that isn't newsworthy. So what we hear about is the funny stuff or, as some people describe it, the scary stuff, which I don't think it's scary at all. But that's what gets written about. I think it is working. But what happens is you go to Bing if you're able to access this and most people are not yet able to access it. But you go to Bing, you start you start using the chat and you never go back to the search. But the chat is buried beneath a click. You have to explicitly go there. And that that's the problem that they're burying it and thinking that their Bing search engine is still best class, whereas the Bing search engine is, you know, like Google, going to be highly challenged by these new experiences in a lot of domains, not in everything. It's going to take years, though, for this thing to work itself out in terms of its real impact on traditional search, isn't it? It's already having an impact on traditional search in some domains. I mean, it comes down to what is chat GPT good for that you would previously have used search for? And I think if you think of search and chat GPT as, you know, two circles overlapping and in the middle is where they have a lot in common. I think search is going to be the net loser and chat will be the winner because it's so much better at so many things that search was never good at in the first place, actually. So if you're trying to find something or learn something, I think chat is going to be a better a better thing a lot of the time. In terms of history repeating itself, I have this sense that this chat GPT phenomenon is equivalent to Napster in the sense that it seems like magic and almost too good to be true. And then it turned out to be good, too good to be true. In the that was the week newsletter you linked to an interesting piece in The New Yorker about whether AI art is stealing from artists in the same way as AI intelligence is stealing from the rest of us. What what was the conclusion on this? Is there something in it, Keith? A lot of lawyers must be rubbing their hands in anticipation here. Well, well, you know, the the image company, beginning with G, Getty, Getty Images is actually suing. I think it's mid journey, but one of the image based AI services for training its AI using Getty Images and without permission. So there are going to be legal cases. I, I do think that it isn't that different than a human being. What a human being does before it is creative, it kind of is in a digestion mode where it learns about a field and then it's then a human starts creating. When it does create it, obviously there are influences from what it discovered. And that's true of AI as well. So I think the idea that this is illegal or stealing is not well founded and will probably not win in court. I think it's more you have to learn to be able to be good at something. And that's true of machines and humans. And you learn from the best if you can. And it doesn't mean you're stealing from them when you learn from them. It means they're contributing to your knowledge. And I think that's a subtle area. I could be wrong, but I'd be super disappointed if the, you know, the zealots for intellectual property won this one because I don't think anything's being stolen. Well, one thing I think I can guarantee is it will end up in the Supreme Court in a few years. What's showing up in the Supreme Court next week? You have a link to an interesting piece in the EFF about Section 230 being on trial. What's happening next week in Washington, D.C., Keith? Well, the Supreme Court is going to hear two cases that impact Section 230. One is Gonzalez versus Google. The other is Twitter versus Tamna. Tamna. I can't pronounce that one very well. I think it's an Arabic name. So basically Section 230 is a law that dates back to 1996 that basically says a platform that enables third parties to publish cannot be held accountable for everything published. And so it kind of is a protection mechanism that makes it possible to have Twitter or Google or any other collection of contributions from associated individuals or organizations. And the threat is that the Supreme Court will either rule Section 230 to be invalid like they did with Roe v. Wade or that they will modify their interpretation of it to criminalize more behavior than was previously criminalized. And in either of those two cases, the question then arises whether it's viable to have a Google or a Twitter because the legal jeopardy would be so great that you wouldn't be able to really do it. That's the framing. Now, you know, the anti-Section 230 crowd will say, well, Google and Facebook and others should be held accountable for what's on their platforms, even though they themselves don't publish it. They just host it. I disagree with that. I think I personally believe the First Amendment should extend to digital media and it should not be you should not be able to be punished for hosting something a third party said that was legal. Well, I have I'll be going to D.C. My wife, as you know, is the head of litigation at Google. So I'm not going to reveal anything apart from the fact that I haven't seen her for about six months. She's been working on this. So there's certainly Turkey. They're certainly taking this thing enormously seriously, as they should, because, as you suggest, it's an existential challenge to Google. It's kind of ironic, as always with the law, that they're trailing behind, just as when search itself is under threat. So now they know that the Supreme Court is is hearing stuff on Section 230 and whatever they rule, there'll be years of lawsuits over it one way or the other. So I don't think anything is going to get resolved, but we'll certainly cover it next week. I'll see if I can get some trade secrets. Finally, Keith, we've got our two favorite features or my favorite features of the show. One is company of the week. And you mentioned Sydney. Sydney showed itself in Twitter this week. What happened with Twitter? Sorry, not Twitter. Tesla. I mean, I'm getting my Tesla's and Twitter's mixed up. Of course, they're owned by the same person. Yeah, and your Sydney's. But so this is a TechCrunch story. Tesla is my company of the week or startup of the week. I used it on purpose because three hundred six thousand vehicles are being recalled over 40 self-driving software. Now, the word recall suggests they're all going to have to drive these cars back to a Tesla shop and have someone. I get the startup of the week. That is a good one. That's.