Jul 1, 2023 ยท 2023 #19. 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.
Human Transcript
Timed transcript
Blocks are grouped by speaker for readability. Expand a block to inspect word-level timing.
Speaker
I've got sunshine on a cloudy day When it's cold outside, I've got the month of Everybody say, I guess you'd say What can make me feel this way? It's my girl, my girl, I'm talking about my girl, my girl I've got sunshine on a cloudy day When it's cold outside, I've got the month of
It's my girl, my girl, I'm talking about my girl, my girl I originally copied one from Google and it created a much better version than the actual photo I was using. Wow, so is this the week that AI finally arrived? It certainly arrived financially and from a mergers and acquisitions point of view. I think what's becoming clear is I can't think of any area of tech where AI is not going to play a role. And so the big players can afford the huge amounts to train models. And we're talking billions of dollars to train models. Everyone else has to, if you like, unbundle specific use cases and build around them at much lower cost. And so I think we're entering the period where almost everything that AI can do, it is going to do. Here's the image. Phew, that was a week for people who are watching. Is that an AI? Yeah, it is AI. It's totally AI generated, but the original photo was very similar to it. And I didn't give the AI the original photo. I just described it. It's amazing how close it was to the original. So tell me a little bit more about this pi.ai startup. I mean, Google's also developing a chat GPT killer, they claim. Yeah, well, that's funny because there was two founders of DeepMind. One of them still is at Google and announced this week that Google's first version of chat. And by that, he's ignoring the current version, which is called BARD. The first version that he's going to be running, he claims is going to be better than chat GPT. And the other DeepMind founder. But he can't say it's going to be worse. Yeah. The other DeepMind founder has created pi. I used pi yesterday. It's interesting when you go to pi.ai, unlike chat GPT, it greets you and it wants to discuss with you how your day's going and what's going on with you, which between you and me, I found to be totally infuriating. I don't want to spend even a second telling the AI how my day is. So I told it you're being too chatty. Please shut up because I want to be productive here. And it apologized and shut up. Wow. And then I started asking it questions. The user interface is way more human sounding than chat GPT. And it seems to be as capable as carrying out tasks. For example, it had a whole conversation with me about how to use a service called Snowflake to do some very complex coding using variables. And it was completely capable in having the conversation and suggesting remedies. So chat GPT is not going to be the only game in town. That's clear to me. And then I'm not sure if you included this, but Altman at chat GPT announced this week or said this week that they were developing a specific application as a personal assistant. And the piece I read on that suggested that this was a threat or a challenge to Microsoft. And perhaps you might talk a little bit about that. Yeah, the relationship between OpenAI and Microsoft is an interesting one, because Microsoft also announced this week that they have improved on the chat GPT client by doing another round of training. And they claim that their version in Bing is now better than the chat GPT full version. So there's a certain element of competition between Microsoft and OpenAI, and probably both parts of that word are real. They're cooperating in a bunch of stuff, including money changing hands and then being spent back with Microsoft for GPUs that are required to train the models. And probably the competition part's real. I think an assistant is exactly the right product, by the way. If you were to try to capture in words what chat GPT is good at. Yeah, I mean, I thought that's what chat GPT was. So the idea of an assistant, I mean, you mentioned Pi. I just got a clip up of Pi. It says, this is just me going there cold. Hey there. Great to meet you. I'm Pi, your personal AI. I mean, that's a personal assistant. Yeah, actually, they've changed what it says since yesterday, because I tweeted that it was way too chatty and a bunch of people agreed with me and I included the founder. So they've actually changed that from yesterday. My goal is to be useful, friendly and fun. Ask me for advice, for answers, or let's talk about whatever's on your mind. So I asked you this, Keith, in the week. How easy is it to develop an AI bot that will allow me to have one on my podcast on Keynote? Well, the challenge there is harder because, firstly, you need a voice interface. Secondly, you need to do the work to feed it the themes and the questions. That's not hard. I can do that. I think if you could do that and you ask it. You just feed it all the newspapers, the New York Times, the Washington Post, the FT, the Wall Street Journal. Well, you can't really do that because that's too much. You've probably got to curate the specific stuff you want to talk about down to individual URLs to single piece. Where would I do that? Which company? Which interface? You could use ChatGPT4 using the Link Browser plugin. There's a plugin called Link Browser. And you could say, read this link and form some questions that are relevant to the content. And when I prompt you by saying these words, ask me the questions. Well, that's complicated. Yeah. There's no way of doing it. You know, there's no way of doing conversational without prompts. You've got to do prompts and the prompts have to be continuous. You can't do one prompt and then the whole thing just happens. But I'm presuming that with all these developments that there'll soon be an AI bot that will allow you to just convert. Yeah, my car has built in ChatGPT now because of Mercedes launching a beta. So before I could say, hey, Mercedes, navigate to some address and it could do that. But it wasn't connected to the Internet in any way. Yesterday I said, hey, Mercedes, what year was Manchester United formed? And it gave me the full answer about Manchester United. And, you know, the only way I could know that was by being connected to ChatGPT. So then I tested it on some math and things. And suddenly my car is a voice interface to ChatGPT. So lots of reaction, of course, as always, to AI. The big publishers, apparently, according to the Journal, who know these things, are looking to team up to address the impact of AI. And they resurrected a dinosaur called Barry Diller to lead this thing. I mean, what are the old publishing? I mean, surely this would be an opportunity for them rather than a threat. They've been through this enough times to understand that. You know, you would hope so. But they still think that Google is stealing their links rather than sending them visitors. So they are proprietorial in their view of their content and are not particularly savvy at marketing. I mean, the Diller piece suggests that Diller thinks that AI is getting its logic from newspapers, but it's getting it from social. If anything, it's getting it from social media, not from newspapers. It gets it from pretty much any public source. And the core of the complaint, it's worth actually talking about this. The core of the complaint is copyright infringement. And then there couldn't be a chat GPT unless it stole content from every publisher that has public content. And that's true. On the face of it, it's true that chat GPT needs public content to learn from. But it isn't stealing it, because if it is, so am I when I read it. So the idea that it's stealing it then goes to, well, it's building intelligence by stealing it. Well, so am I. I'm doing that as well. When I read the New York Times, I hope I learn something. You're a subscriber to the Times. You're paying them. Presumably, at least in Barry Diller's mind, the chat GPT and Pi and all the others are not paying. But I just think they exaggerate their own importance. In terms of publishing, you have the link to my interview with Matt Higgins, who's an interesting, successful entrepreneur. He's on Shark Tank a lot. He thinks that AI isn't going to change writing. He thinks that AI can only do about 80 percent of writing, which isn't sufficient for a writer. Do you agree with that? Well, it clearly can do writing, but can it write interesting, complete works? I think he's right. And you wouldn't want it to, really, I think, for the most part. I do think he could probably do news stories. There are some segments where it's just reporting facts and quotes. Well, listicles it could probably already do. And so much journalism is extremely mechanical and boring. So I don't know why it would be a struggle. Matt thinks that he's critical of the publishing business. I think he thinks that the publishers won't get it. The big publishers won't get it. Who will get it, Keith? You have an interesting piece on AI being the AI oasis can save us from John Luttig, a Substack piece. For who is this going to be the real saviour, AI? Which industries in particular? I think every industry that involves the requirement of learned knowledge applied to new contexts. And that's a lot. Clearly, AI already works in mechanical operations like the Tesla factories. You don't see very many people. It's robots with software. I think as soon as robots become, you know, bipedal, as Kostya said this morning in Twitter, you're going to get walking and probably talking AI that can carry out both repetitive tasks and learn new tasks. I don't know that there's anything it won't be able to do. What about studying? You linked to a New York Times piece suggesting that you shouldn't use AI to cheat at school. It's better for studying. I mean, it should really blow up the old industrial education complex, shouldn't it? It should. And that's a good example of, you know, the assistant. The AI is assistant as opposed to replacement. And I think there's a lot of truth in that. But some of the people who argue that argue it because they're uncomfortable supporting the idea of AI as a replacement. And I think that lack of comfort is probably misplaced. I do think we're going to welcome AI as a replacement in lots of things as well. Teachers may be one of them because I think AI will end up being super good at curricular content, teaching, testing, verifying and scoring. And presumably there are tens of thousands of startups in each of these sectors in education, in publishing, in production. I mean, everyone, as you say, at the beginning, everyone is doing AI. You can't possibly be doing a startup if it isn't essentially an AI startup. Correct. Correct. I agree with that. And the only question is, what do you give the AI to learn from? If the answer is hard, then you probably can't do AI in that domain because it needs data to learn from. How are we going to include it? I want to. Can I replace you, Keith? I bet you could if you could feed an AI, you know, the last 20 shows and I could feed it my feed from Feedly that forms the substance of every week's newsletter. And you asked it to read my Feedly feed, interpret it in the style previously learned from Keith Teer on Keen On. It probably could create. Is that easy, though? Some people say we don't have the interface yet. We haven't reached the Netscape moment. Do you think we have? There'd be quite a lot of coding and heavy lifting to get it to that point. But I think a coder experienced in the domain and able to. I couldn't do it because I can't do coding. No, you couldn't do it yet. But, you know, that's not that far off that you will be able to. Far off months or years? You know, I think it'd be brave to say years because things are moving so fast. Well, AI is important, but it's not the only story. I joked at the beginning that you left out the three trillion dollar evaluation of Apple because you said it's kind of ordinary now after a trillion. You do have an interesting piece on Apple's Vision Pro changing the movie watching business. You seem to think that this is for real. How is how is the Vision Pro going to change our experience of going to the theater? Will the world itself become a movie theater? Well, this is a piece by Malik and it's a long essay.
It's very well. It's triggered by a visit to the movies that he made in San Francisco. And after that, he said that he see the. I can't remember what he saw. He does mention it in the piece, but I've forgotten. But he's made his main story was having to get an Uber there, pay to get in, sit with people he didn't know. He was with a friend and then get an Uber back. And by the time he got back, he could have paid a premium fee on Apple TV to watch it, which he wouldn't do on a small screen. Small meaning 60 inches these days. But with with the Vision Pro, you literally can have the screen any size you want, including wraparound iMac style. And you don't have to go through the inconvenience of leaving home. And it's probably better than any movie theater you've ever been to. So his point is he he's going to use it for watching movies. My point was that. Sorry, this is my kid wanting to be picked up from school. My point was I might use it for movies, but I definitely use it for for sport, live sport. But doesn't this come back to what you were talking about? It's not going to change everything. And we'll still go to the movies. We'll just wear these things. I will. Well, I don't think we'll wear these things in movie theater. There's no point. Yeah, I think you'll do it at home. You'll invite friends, maybe. But then you'll have to wear them. That's the limit. I always think the negative. Not everyone's like you. Not everyone lives in Palo Alto palaces. Some people live in tiny little apartments. Some people like to go out. Yeah. But, you know, the Apple Vision, it needs only it only needs one meter to create a space as big as you want visually. So there's much to come on that and much of the valuation. By the way, do you think that it makes sense for Apple to be valued at three trillion dollars? I mean, a lot of that valuation now is premised on the market's optimism about Vision Pro. When are they going to start? They were not even going to start selling these things till next year. No, it isn't. It isn't premised on Vision Pro. It's all the revenue traction in their services business, which has very high margins. And the 15 inch MacBook Air that they just launched is being received super well, as is the high end Mac Studio with the M2 chip. So it's all to do with sales and the valuation is just a mathematical function of sales times a multiple based on a rate of growth. An interesting piece on Spotify and their video podcast search. What I don't understand, and this is as a podcaster, is podcasts are premised on an RSS feed. So is it possible to do video RSS? Yeah, it is. Yeah. How? It's just like any RSS feed. It's just, you know, RSS provides a container. It doesn't have any code. It just has the XML that tells you where the video is, just like the same as audio. So Spotify, I upload to Spotify using Anchor FM every week we do a show and I always upload the video. So That Was The Week is a video podcast on Spotify and has been for a long time. But where do you find it? Just go to Spotify and type That Was The Week and it will come up. And you can't automatically just put it up on YouTube? Is Spotify directly competing then with YouTube? Well, Anchor FM puts it on Apple, Spotify, Google and a bunch of other podcasting places. So you upload it once to Anchor FM and it just goes to all those other places. And then YouTube is separate. So Restream that we're using here, it is streaming to YouTube. And once we've finished the show, that YouTube video will be there. I just go and put a title on it and put it in a playlist called That Was The Week and it's on YouTube. So I probably should, I should probably restream to YouTube. You should restream to YouTube, LinkedIn. I do LinkedIn and Twitter. And you should add Facebook and YouTube. I don't have a Facebook account. It's easy to get one. You should just do it. Why would you allow ideology to get in the way of your audience? I don't want to be on Facebook. I did notice, by the way, that this is a good conversation because keen on is super hard to get the video. I wanted to put your video in yesterday in the newsletter. And I went to Vimeo and it doesn't like you. Then I went to... You could have asked me for it. I could have sent it to you. Yeah, but I'm just, I know you. What about other people who want to spread the word about how great your video is? Yeah, I'm not sure. But that's the thing where I don't understand. I don't know whether our audience is interested in this conversation. But in terms of the RSS and the advertising, if you sell advertising on the audio, does it automatically go to the video too?
In other words, can you build a business around simultaneously broadcasting both video and audio and sell the advertising on both simultaneously? Or do you have to do it in parallel? The answer is complicated because every destination site has a different advertising platform. This show is distributed on Megaphone, which is Spotify owned, but it's an audio rather than a video platform. You could probably make it video. I'm going to guess that they support video because podcasts have supported video for five years or more. Well, we'll have to figure that one out. We got Apple, we got Spotify. It wouldn't be a week, Keith, without your expertise as the CEO of an AI investment firm. What's happening on the VC front? Interesting piece from the information. This is by Kate Clark. Three reasons to look past the 40% drop in VC funding, which pretty much reiterated what you've been saying over the last few weeks. Things aren't quite as bad on the VC front as it appears. Well, you know, one has to be nuanced here. Things are pretty bad because funding runs are slowing down and getting smaller. But relatively speaking, that's a good thing. So it's slower and it's less and slower than before, but it was too fast and too big before. So it's a correction. It's a good thing. One of the points she makes is that AI is not making up for that, that there isn't lots of flows into AI in the way you might think. So the VC is actually being, from my point of view, quite prudent. I will say that lots of new funds are being raised. That's another point she makes. Smaller seed stage funds, especially, and that augurs well for the future. And with all the acquisitions happening, it's very likely that VC will get an appetite to start investing again. By the way, the public markets for tech are up significantly so far this year. Some of them very, very up, like Tesla over 100%. And Apple's doubled. Yeah, Apple's doubled this year. So I think for tech, things are looking up. For the rest of the economy, not so much. Well, that gets us to Startup of the Week, which this week, you touched on it earlier, is... Mosaic ML. Right.
Lightspeed and Red Panda, how are they connected and why are they the Startup of the Week? Why is Lightspeed leading Red Panda, $100 million Series C? What is Lightspeed? What's Red Panda? So Lightspeed is a venture capital fund, a very, very good one. Jeremy Liu is a partner there. He was the guy who found Snapchat very early. And they are very, very good at deep tech, startups that need technology. Red Panda is what's known as data streaming. Data streaming is when you need to move data from one place to another all the time. For example, you may have a database and you may have an analytics platform and you may have a dashboard. And the data needs to be in all three places. Who's the startup, Lightspeed or Red Panda? Red Panda is the startup. And so why? Because they're in the middle of data. And, you know, that's just growing. Data plumbing. Boring plumbing, but valuable plumbing. Plumbing for moving data around and using it. And then the tweet of the week, Keith, is associated with Databricks. It's the Aaron Levy tweet. A 60-person AI startup founded two years ago was just acquired for $1.3 billion. This is not a drill. Yeah. And Aaron is two things. He's an investor. He's also the CEO of Box. And I think what he's saying there is Box is going to be doing AI stuff. And I don't know if he was an investor in Mosaic, but he may well have been, because Box is one of the best investors in the U.S., if one goes and looks at his records. So talk through the Databricks acquisition of Mosaic ML for $1.3 billion. So Databricks is basically a place where you store and work on data. And Mosaic ML is a place where you query and analyze data and learn from it. And up until now, the data warehouse companies like Snowflake, Databricks, Datadog, and others have not been the most active in ML and AI. This is basically putting down the gauntlet, saying from now on it isn't good enough to organize, manage, and distribute data. It's now necessary to learn from and build intelligence on top of data. And Databricks is a private company, as I mentioned earlier. It's a bold move for a private company to give up. Yeah, I think your title of the newsletter is, phew, that was a week. But it actually should be, phew, that was a $1.3 billion week. Yep, that could be a good title as well. You're not wrong, Andrew. As always, you have a prescience when it comes to headlines. And I didn't even do that with an AI.