Aug 23, 2024 ยท 2024 #28. Read the transcript grouped by speaker, inspect word-level timecodes, and optionally turn subtitles on for direct video playback
Edit labels for this show, save them in this browser, or download a JSON override for the production folder.
Transcript Playback
Kids Love AI
Human Transcript
Timed transcript
Blocks are grouped by speaker for readability. Expand a block to inspect word-level timing.
Speaker 1
hello everybody it's friday august 23rd 2024 when i'm going to talk trump or harris got more important things to think about like ian bogost he was on the show a few years ago on the keen on show One of technologies I think smarter, more experienced, more incisive critics, observers. When he was on my show, he had a new book out, Play Anything, The Pleasure of Limits, The Uses of Boredom and the Secret of Games. It's a good book. And Bogost is a regular contributor to The Atlantic. He's one of their tech contributors. His piece this week was entitled, AI Cheating is Getting Worse. Colleges Still Don't Have a Plan. It's an interesting, provocative piece. He argues that AI in colleges now has resulted in the situation devolving into what he calls absurdism. But not everyone agrees with Bogost. Steven Sinofsky, a very significant tech executive. He was Mr. Windows at Microsoft for many years. He has an excellent blog called Hardware Software. And he responded explicitly saying using AI for school is not cheating. So we have Sinofsky. And we have Bogost. And then, of course, we have the great Keith Teer, who in this week's That Was The Week newsletter comes out very clearly in the Sinofsky camp. Kids love AI, his newsletter says, and teachers should too. He acknowledges in the editorial that it was 100% inspired by Sinofsky's essay. So, Keith, perhaps you might explain why... uh, Sanofsky's inspired you for this week's newsletter?
I actually, you know, I had this dirty little secret I was harboring, which is that I write about it in the editorial that when I was in, uh, Paris for the Olympics, my youngest son had to do some summer school work, uh, English work, all about rhetorical devices, understanding rhetorical devices. You'd think it would be genetic based on him being my child, but it apparently isn't. And, um, It was multi-answer questions. And so he was doing these modules, 10 questions at a time, four possible answers per question. he did them and then he photographed them and sent them to me and I fed them to chat GPT. And I said, you know, can you read this question, look at the answers and, and give me your reasoning why you either agree or disagree with Luke's answer. And it gave a, you know, maybe a 50 to a hundred word reasoned response saying he is right or is wrong or whatever. Across 200 questions,
It disagreed with my son, I think, about five times. He changed his mind, I think, a couple of those times to ChatGPT's answer. And on balance, ChatGPT was wrong some of the time. My son was wrong some of the time. His score probably ended up being roughly what it would have been without ChatGPT. But what was gained in the process was I sat with him, read the reasoning... and discussed whether he should change his answer, which meant that his understanding of each thing went way deeper than it would have done otherwise. It was almost like having a teacher sitting next to you. And the dirty little secret was I was asking myself, are we cheating? And it didn't feel like it should be cheating because pretty much none of the answers changed. A small number did, but weren't always right. And he disagreed with ChatGPT often as well and was right some of the time to disagree. So it didn't feel like cheating. It felt like just a new process available to a 17-year-old doing 11th grade English.
So I've got two responses to that, Keith, and maybe we can deal with them separately. The first is that not everyone has as... technically sophisticated and dedicated parents as yourself. And what you did, I don't think any parent or any kid would be critical of. You understand this technology. You set it up for your son. You're very much involved with your kids. So what I fear about AI, like all these technologies, is that For our, shall we say, class of upper middle class, technically literate people, this technology is great. For everyone else, it's less good. For the kids who don't have parents to show them around, it's problematic. I'm not sure you really addressed the heart of the Bogost piece, which inspired the Sinofsky response. In college, he says, and I'm quoting Bogost from the Atlantic piece, if the first year of AI college ended in a feeling of dismay, the situation is now, and this is in the second year, has devolved into absurdism. Teachers struggle to continue teaching, even as they wonder whether they're grading students or computers. In the meantime, an endless AI cheating and detection arms race play out in the background. I mean, what you're talking about, what Luke dealt with, he wasn't using chat GPT to write. He was using it to reason. Is that fair?
Let's then deal with the Sinofsky response and explain. I know his is quite anecdotal. He writes about his own personal experience of the beginnings of computing in his piece. He even has a picture of a Smith Corona typewriter, sort of early proto-PC that he used. What's his argument about the way in which kids now are using AI to write their own stuff in college?
Yeah. And after, let's go back to what Bogart, as it said.
Words and timings
Yeah.Andafter,let'sgobacktowhatBogart,asitsaid.
Speaker 1
Bogart.
Words and timings
Bogart.
Speaker 3
Bogart said, because... Not Bogart.
Words and timings
Bogartsaid,because...NotBogart.
Speaker 1
That was an actor cue.
Words and timings
Thatwasanactorcue.
Speaker 3
Yeah, there you go. So Sunofsky is making actually a very simple point that it's pretty hard to disagree with, which is that all technology which enhances the individual first enters the world accused of cheating. You know, I can remember calculators were a great example, and he speaks at length about that. And so if the individual is able to do better due to some technical innovation, that's considered initially cheating. But after a very short period of time, and he argues a shrinking period of time, it gets incorporated into the mainstream as an asset or a tool. And I think, broadly speaking, that's right.
It is, but my response, and you and I have talked, argued about this endlessly. I mean, there's also the example of Napster and the whole scale stealing of musical content at the turn of the century, which eventually was fixed. So that was a kind of cheating that didn't work out.
Well, yeah, it's a harder parallel because it's not focused on education, but let's go with it. What Napster did is show that digital music separated from physical media was attractive to millions and millions of people. But in order to do that, it had to break copyright laws. Like YouTube, except YouTube got away with it, Napster didn't. Well, YouTube was mainly user-generated content, so it was easier. But yes, it did eventually come up against the same issues and now has a solution. But I do think the parallel does hold to the extent that you couldn't imagine Spotify and Apple Music or YouTube Music without Napster having happened first. In fact, even before Napster, in England, when I was doing Siberia Cafe, there was a guy who created something called Cerberus, which was an MP3 streamer. way back then, and he could rip music off of CDs as well. So, you know, it's a long history of technical innovation leading to change business models. In the case of teachers, going back to Bogus's thing, that's probably not right either. What's his name? Bogost. Bogost. Bogus. I think I prefer Bogus. Anyway,
What it means is his point about teachers is correct if you take a snapshot today. Just as musicians and copyright holders were very, very confused what to do when Napster came out. And teachers basically, their core role has to change. It makes no sense for the teacher to be the primary deliverer of knowledge when you have computers that are going to be way better at delivering knowledge. It probably isn't even the teacher's job to score tests because technology is going to be able to do that. What the teacher needs to be doing is what I did with my kid, which is being the mentor Well, you were being the teacher.
Yeah. And by the way, I didn't know the subject matter very well. So I was really more guiding the thinking process. I would call it emerging critical thinking as opposed to the knowledge or the scoring. Today's teachers are not doing critical thinking. They're doing knowledge transfer and scoring. which are the least valuable parts of a teacher's job. So redefining a teacher as, if you will, like a mentor or a curator of children, using AI as a tool seems to be a far more promising way to think about it, to me. So I agree that they're confused right now, and I agree they're bereft of ideas, but I think that is probably a momentary stage in what Sanofsky wants uh argues will be an ultimate end point where ai is just accepted as a as a really
good tool so what about the bogus piece about colleges where there's this devolution into absurdism where teachers have no idea whether or not the kids are actually writing what they're supposed to write because his piece in the atlantic is about the death essentially of the college essay because initially college, especially the liberal arts education, was premised on kids writing essays and writing them themselves. Now they're using open AI and all these other platforms to do so. What are we supposed to do?
Well, you know, if you remember back to when we learned that, Writing an essay, it was really learning structure. What is a good opening thesis statement or attention-grabbing fictional paragraph? What is the middle all about? What is the end all about? And so essay writing was really scoring learning structure, and the creativity in the writing was always a very minor part of it. So it feels to me as if...
I don't think computers can write essays creatively as well as humans yet, but they can write creative essays as well as the average human already quite well. And so the writing process and the structure, probably just like a calculator knows how to divide and multiply, and you don't need your times table, that's going to become commoditized.
Right, so maybe what you're saying, and this is an argument a lot of people... in AI make is that on the one hand, AI will commodify certain aspects of essay writing, but it actually adds to the value of creativity and the quality of the writing in itself, because the software is never really going to be able to replicate what humans can do when it comes to style. Oddly enough, it actually adds to the human quality of writing.
Well, I think what you'll get is a division between people who hate writing but have to to get the score versus people who love writing and hate being scored.
So this comes back to what I was saying earlier about the fact that kids who have the good fortune to have parents like yourself are going to be fine here. Is this just going to be another in a long line of technologies where the elite benefit massively and everything else and everyone else gets commodified?
Let's not go down that rabbit hole. Nobody is the short answer. So as long as the internet is disseminating knowledge through social media and other forms, kids from all backgrounds are going to learn how to do things. Look at the quality of TikToks from poor kids. It isn't worse than the quality of TikToks from rich kids. In fact, it's often better. So I think... I think there's a leveling effect that technology brings in terms of knowledge transfer and capability to execute. So I think the main benefit the elites have is the survival of the existing education system and its elitist hierarchies. I don't know that that education system can survive or even should, which means the elitist kids, their advantage is going to shrink.
Yeah, and I think one of the problems with, I mean, this is part of a broader structural crisis of education, both K through 12 and university system in America and elsewhere. So, I mean, there are other issues as well. But also there's this idea of the absence of trust is that the authorities, the teachers are always assuming that the students are cheating. And I think there's a reciprocity there is that the students don't trust their teachers or their university administrators. So it's also sort of part of this crisis of trust.
Yeah, well, that is part of a broader cultural crisis suffered by all institutions. And the lack of trust is kind of well-earned, I believe, because the job that has been done by most of these institutions is less than good. So I think trust has to be earned, and it can only be earned
when you're helping and not standing in the way does this also suggest keith we've had this conversation before i'm a bit more skeptical than you on ai because i don't to me at least there still aren't real products in ai that are usable but does this sort of debate now between bogost and um sanofsky reflect the fact that ai is now entering the bloodstream of society, of education, and is for real?
I mean, my observation is it was almost instant. Nine months ago, you know, kids in Gunn High School here in Palo Alto had chat GPT, and they also had post-chat GPT AIs that take out the obvious signs of it being AI before submitting an essay. So these tools were multi-layered, almost like agentic, multiple agents being used by a single individual right away. And I'm talking about 16-year-olds.
So maybe we need to revise. In the old days, we used to assume that the cutting-edge technology was always pioneered by the porn industry. Maybe it's the kids who are doing it.
I think the kids are doing it. Even go back further, three or four years ago, there was an app. It still exists. It's called Photomath. If you install Photomath on your iPhone and you point it at an equation, it will solve it for you and explain the steps and teach you how it solved it. So I think this... evolution of educational technology to be able to do tasks which previously kids were tested on just means kids are going to have to be tested on some different things, arguably more important things.
Yeah, well, we will return to this many times. It's not going away. Lots of other interesting news, Keith. One thing that caught my eye was a piece in your newsletter from Genet tier? Your superior in the tier household?
Well, I'm sure he reports to Genet. Anyway, it's about A16Z and Founders Fund lead the way in defense venture capital. A16Z will come to them later, their X of the week as well. They're not very popular, certainly amongst the progressive tech community. A few weeks ago, we dealt with, according to The Verge, the moral bankruptcy of Andreessen and Horowitz when they came out in favor of Trump. Is this troubling for you, this leading the way in defense venture capital? I mean, it's basically tech for war, right?
Mm-hmm. Yeah, it's interesting. I didn't put it in the newsletter, but some Google employees this week signed a petition objecting to Google participating in defense contracts.
Wow, the Google, they... They took over one of their senior executive offices in terms of, I think, Google supporting Israeli military technology. So that's a continual theme at Google.
So this is a growing discussion in Silicon Valley, which is what is the relationship between innovation and defense? Clearly, there's always been a relationship. In the past, it was in the reverse. You know, the government spent money on R&D for defense, and many of the benefits span out into products.
Yeah, and the internet. I mean, coming back to your friend, the internet, that you're a little worried about, the internet came out of the military-industrial complex. It came out of government funding for defense.
So it's kind of going into reverse, which is innovation is no longer coming from the government. Even NASA, you know, with the stranded space travelers on the International Space Station that it can't get back. It looks like it's going to ask SpaceX to go and get them. So government really can't innovate anymore. Private sector can. And there are now growing. And not everyone would agree with that, Keith. But anyway, go on. Well, I'm painting broad brushstrokes that are roughly correct, I think, directionally at least. But now what you have in the Valley is what most people think of as the right wing in the Valley. I don't really agree with that characterization. I'm more and more inclined to take money to invest in defense tech and are pro-American on the world stage and are pro-using force where necessary on the world stage and don't have the scruples of the baby boomer generation about... participating in defense. And this article is tracing that. And it's real and on a very large scale. You know, the Secret Services do have venture funds now in the Valley that invest in companies that may end up being beneficial to U.S. defense interests. Palantir is a great example.
Yeah, and Palantir, of course, is a Peter Thiel company. I mean, maybe it's a split between Thiel and Hoffman. It came out in the open on their disagreement on Trump and Harris. But this has always been in the Valley.
Well, as long as you have competing global interests with armies backing up national interests, there's going to be in every country, you know, a whole industry in building things for the military.
So what about the moral? I mean, it's a big subject. I'm just interested in your view. The Andreessen Horowitz Company are leading this, as I said, a lot of criticism in their support of Trump. Should there be a particular moral... Do you have any moral qualms about this or you as a sort of libertarian supporter of the free market, if people wanna buy it, then you should be allowed to make it. Is that your argument?
I'm generally a laissez-faire believer. I think if China has better technology, if China has higher productivity, if China can produce things for human beings cheaper and better than America, then I'm not anti-Chinese. I don't reject it because it's Chinese. I think the output of human civilization for human life is valid no matter where it comes from.
In terms of a military-industrial complex, presumably these companies are getting financed by the Founders Fund and the Andreessen Horowitz Fund. They're not going to be able to sell their technology to the Chinese or the Iranians or the Russians, are they?
They try. I mean, NVIDIA sells... NVIDIA has developed very specific GPUs, which are a little bit less powerful than the ones they sell here, specifically for the Chinese market, and has gotten U.S. permission to sell them as well, by the way, because capitalism, even though it's competitive between nation states, actually requires... purchases between nation states in order to grow so there's a you know there's a a key contradiction at the very core of capitalism is markets need to grow and nations are small therefore globalism is inevitable at some level and denying technology across borders probably ends up failing So if you're really thinking about the future, you want to embrace less fighting. And this defense stuff is really all about making money from the belief that...
Yeah, I mean, maybe that's one of the reasons I've always a little bemused with how explicitly Andreessen and Horowitz came out in favor of Trump. But maybe... Like him, they're thinking in instrumental terms. Lots of other news this week. Again, it'll be a theme we come back to. Last week, the last few weeks, Keith, we've been talking about the decision on Google. Interesting Benedict Evans piece today on Google and competing in search. What does Evans, who is an intelligent commentator, what does he say about the Google antitrust case?
Two main points. The first point he makes is that mostly an incumbent once they get into the lead are very hard to shake out of the lead and so new search engines probably isn't scary to Google because The likelihood they could do better than Google is very small therefore a second point is Technologies are displaced by new technologies not by competing equivalents and in the case of search the most likely replacement is AI because it can convey information without doing searches. And so the risk to Google is really AI, hence Google's requirement to invest in AI with Gemini. And the government seeking to regulate or adjudge wrapping it over the knuckles, is way less threatening to Google than ChatGPT building a replacement. That's his core point.
No, they're bothered, but they're more bothered by ChatGPT than they are by this because it's more likely to affect them. Money-wise, I think most commentators don't believe Google will be broken up after this decision. They believe there'll be some kind of knuckle-wrapping or maybe a fine or maybe some behavioral constraints and maybe nothing because it will be appealed. So ChatGPT, on the other hand, is going to release Strawberry, its next version, sometime soon. And the rumors are it is many, many leaps ahead of anything we've seen so far.
And I'm sure Jem and I are on similar tracks. Again, a story that we'll come back to time and time again. Last week, we talked about, at least in your language, Keith, maybe the death of the EU, a theme from Rohan Silva, who is very highly placed. tech investor player in the uk we can't let eu's anti-ai rules drag us down too on the uk is he in your camp uh does rohan silva suggest i think he was which um he he was at number 10 for a while i met him a few times um yeah is he warning about the same thing that europe is dying
He is not just warning, he's saying it's happened already, that innovation in Europe is being strangled out of it by regulation. He sees the British system as being outside of that, but... British politicians being underfunded and lazy, he fears that they'll just take European rules and apply them to the UK. And it's a begging letter for them not to do that, but to embrace innovation. So it's very aligned to last week's editorial that we discussed on the show, you know, is Europe dying? And very similar belief system to mine, which is that regulation really isn't appropriate in the very early days of a transformative technology because what happens then is the government become the decision makers of what is and isn't possible. And I don't know if you're going to show it, Andrew, but there's a second article this week about Meta, which uses public posts to train its AI. And the EU was wanting to stop that. but has decided it can't because it doesn't know what it's doing.
Yeah, and it's basically because they're waiting for the EU to tell them what the product needs to be, which is bizarre. So I think the politicians have gotten themselves into a place they really don't want to be which is they're going to have to be the architects of innovation because no one is going to ship a product if they don't know it isn't going to be regulated out of existence.
Well, one thing I think we can say for sure, Keith, is that The EU is not going to win your Startup of the Week award this week. The award goes to Black Forest Labs. You have a link from TechCrunch, which describes it as the startup pairing Elon Musk's unhinged AI image generator. So not everyone's an admirer of this. Tell us about Black Forest. Who are they and what do they do?
So Flux, and this isn't even the most realistic you can go, but Flux is many leaps ahead. of the ones we used to use in that it's able to be super hyper-realistic.
Yeah, it's not as generic as the other ones you use. I mean, the image for people who are just listening to this, the image of a young boy staring into a computer screen, and the image of the boy is actually quite good.
I mean, again, so we do have AI startups in Europe.
Words and timings
Imean,again,sowedohaveAIstartupsinEurope.
Speaker 3
We do. They're just underfunded by European money. That's the problem. European money... And I bet you anything Black Forest Labs in Germany... have very strict rules about what they're not allowed to let you produce. Although X, being consistent with Elon Musk view of the world, does not censor what you can produce in an image at all.
Yeah, well, that's... our friend at X. The X of the week, Keith, is from Martin Casado, who is very senior at A16Z, the GP there. Martin, and very much in the context of what you've been talking about today, it's an X about OpenAI's letter opposing SB1047, which is...
She certainly would have favoured it. I don't know if she had her hands on it. But it was Scott Weiner, I think it was. No Weiner jokes today, I hope. No, absolutely not. But this law has been reduced, let's say. OpenAI... OpenAI produced an open letter saying, A, we're super happy to work with California in figuring out regulation, and we do think there's a need for regulation. And two, we're also super happy to work with the federal government doing the same. And Martin Casado is pointing the finger at them and saying, you're idiots, you are providing
the governments, in this case, with the power to act as the product manager over AI, which they're clearly incompetent to do. And you as the leader, the only reason you would do this is because you want them to knight you as the leader and work with you to prevent other people being able to compete with you. And so Casado is the laissez-faire capitalist here. and I agree with him.
What about Altman in this? I know that he's brought on Chris Lehane, who's a very heavy-hitting former Democrat uh power player uh was it um was it airbnb i mean is the issue i think he's his right hand man now at open ai um is the the challenge of open ai now all about regulation and and it is uh a16z are they invested in open ai
I don't think so. They're investors in lots of competitive products, though, including Facebook. Look, I think we're at a moment in history where it isn't 100% clear to me what the right thing to do is, to be totally honest.
Clearly, AI in the hands of bad actors can be, I doubt it can be seriously damaging to the human race, but it can certainly be used for criminal end results that impact individuals. And so I think, you know, no regulation makes no sense. It also makes no sense to stop early stage innovation and ask government permission for almost everything you do. So there's somewhere in the middle that makes sense. And I don't think anyone's found that yet. It also needs to be global, by the way, because it. If the U.S. has regulation...
It's not going to be global because the regulation already, the Internet's regulated dramatically differently in China or Iran or EU or Russia than it is in the U.S. So it's never going to be global.
So when I say it needs to be global, I'm almost pointing to a flaw in the whole idea because if it isn't global, you know, innovation... that you don't like can happen in places that are not regulated. There's no, you know, the earth's a small place, but it's also a big place. It's totally possible to go to a place where there is no regulation and do exactly what you would do in a place where there is. So unless there's a kind of a human consensus on what is it that benefits humanity versus what is it that doesn't,
There is a dramatic, going back to Google, Google was so unregulated in the late 90s. I mean, there was no regulation at all. Well, what Google was doing and the other Web2 companies, Facebook, YouTube, and now... governments at least and politicians and maybe the public are feeling as if they need to have some control, otherwise they'll look away and in five or ten years' time we'll be in a situation where this technology can't be controlled or regulated.
Anyway, his point is that, um, there's a point of no going back when the winner is already in. And, and that, I think that's already true with AI, to be honest. I think the winner is already in and it's open AI and probably anthropic. Uh, the two big ones, at least in, in general AI, I think you're writing Google off. Google will be a player, but they won't own the market. Um, And by the way, I think it would have been wrong to regulate Google earlier. Imagine when Google was young and you kind of stopped it doing Google Docs, spreadsheets and PowerPoints and so on. That wouldn't have been a good thing. It just would have preserved Microsoft's monopoly with Word and PowerPoint and Excel. So I personally think regulation is a blunt instrument. It's way better to look at the market and fund better competition than it is to regulate.
Well, that was the week for August 23rd, 2024. We probably have one more August week, Keith, and then we'll be into September. Lots of new products and developments. It has been anything but a quiet summer, and I will see you.