← 所有节目
EP222026年11月9日00:41:44

AI 音乐、同人作品与版权归属

Michael Lin & Billy Poon
IIPCC 会长;香港专利师协会理事

Michael Lin — 国际知识产权商业化委员会(IIPCC)会长,也是统一专利法院专利调解及仲裁中心的专家。他自 1997 年起在亚洲从事知识产权和商业工作,持有波士顿大学法学院法律博士学位。他曾是本节目 EP03 的嘉宾。 Billy Poon — 香港专利师协会理事。他在 GRST 任专利科学家期间,是一项水基锂离子电池回收专利的发明人之一,并处理中国、欧洲、日本、韩国和美国专利审查员的审查意见。他持有香港大学化学 (Intensive) 理学士学位。

IIPCC ↗
IPCopyrightPatentsAI Music

章节

  1. 0:00AI 能否霸占所有歌曲?
  2. 1:06片头
  3. 1:20欢迎,以及为什么我们总是谈到知识产权
  4. 2:46Billy Poon:专利科学家与发明人
  5. 4:30专利团队放在哪里:宝洁
  6. 5:30中国专利代理师考试,以及 Billy 的发明
  7. 8:02AI 音乐与版权
  8. 10:29YouTube Content ID 与 AI 生成的歌曲
  9. 12:42小频道遇上版权警告
  10. 14:11霸占权利,以及独立创作作为抗辩
  11. 16:30Marvin Gaye 与无意识抄袭
  12. 16:55Copyleft
  13. 17:31用 AI 逆向工程 PlayStation 游戏
  14. 18:54同人作品:起诉他们还是合作?
  15. 19:50Louis Vuitton、Luigi 与乐高电影
  16. 21:54愤怒的小鸟:把侵权者变成授权商
  17. 25:20烧给祖先的纸扎 Gucci 手袋
  18. 28:28执法者与品牌的诱因不一致
  19. 29:15中国法律与香港法律:有没有可参考的模式?
  20. 30:58中国如何建立专利法
  21. 31:53IIPCC 与为政府提供意见
  22. 32:56AI 新动向:云端智能体与 Opus 5.5
  23. 34:12Claude 对 OpenAI:风格与代码审查
  24. 34:51OpenAI 是 Microsoft,Claude 是 Apple?
  25. 35:27现在的限制是你的想象力
  26. 36:53AI 实验室在内部模型看到什么
  27. 38:08亿万富翁在新西兰的地堡
  28. 40:03道别,以及纹身展

关于本集

Michael Lin 自 1997 年起在亚洲从事知识产权工作。Billy Poon 是专利科学家和发明人,正在准备中国专利代理师考试,合格率不足 10%。在香港中国国际纹身展之后的一顿饭上,Tanya 和 Augustin 问他们:AI 给知识产权带来什么改变? 有人能否用 AI 生成数以千计的歌曲,在 YouTube 的 Content ID 登记,然后把其他人的视频下架?Michael 说理论和实际很不同。AI 生成的作品未必受版权保护,而在版权法中,独立创作是一种抗辩。他们也谈到品牌应如何对待使用其知识产权的粉丝(乐高、愤怒的小鸟,以及香港人烧给祖先的纸扎 Gucci 手袋)、中国如何建立专利法,以及 AI 的新动向:云端智能体、Claude 对 OpenAI,还有新西兰的地堡。

文字稿

机器转录,经人工校对。人名与术语已更正,个别字词或仍有误。时间可链接至视频。英语部分保留原文。

AI 能否霸占所有歌曲?

0:00And so I think the problem is with AI, you can algorithmically generate sequences of tones or music and chords, and so you could basically squat and, like, algorithmically generate all of these things, all these songs, register them, such that you have a Content ID, register with YouTube, and then if YouTube hears something that's similar to what you've registered, you have the right to take it down, right? There's the theory and then there's the practice, and they're wildly different. So in theory, yes, you're right. In practice, How much is it really going to happen? I mean, the other thing is that if you did that, if you say you registered a thousand songs, like you squat on them, and then I come and I use something which is very similar or even the same, then you say, well, this is my song, copyright infringement, blah, blah, blah. But I would say it's not copyrighted.

欢迎,以及为什么我们总是谈到知识产权

1:22Welcome to the Hong Kong AI Podcast, and we're doing a slightly interesting version with food. I think we should always do it like this. And so I'm Tanya, and this is Aug. And I think today it's interesting because we are actually seeping through to our other podcast. The more we do the Hong Kong AI Podcast, the more we realize that we're always talking about one other thing which is the IP. So I think... And earlier today was culture. Yeah but you know that's like thank you all for throwing in the red tokens but I don't know I don't know. You completely derailed your train of thought. That's what the audience wants. I don't know about that. The audience wants some derailing. If you want, like, if you like opposite derailing or you want to beat the shit out of him, let me know. Leave a comment. I'm always willing to do whatever you guys want. For viewership, I'll do whatever. So I think the more that we talk about AI, the more the other subject come up is, like, IP. So I think we will branch out into a Hong Kong IP podcast. so this is why we have Michael and we have Billy and you're aspiring. Billy was telling me he's an inventor, he's also an inventor.

Billy Poon:专利科学家与发明人

2:53Is that why you got into patent? Not really, I was helping a company to do a patent lab like two years ago and doing that time because as a As a patent practitioner, you need to do lots of search on patents, you need to read lots of patents. Somehow I think, well it's a great opportunity to do an improvement on some kind of project and then I just contact the R&D team that I'm working with.

3:30and somehow we have established a clear road map on how to do the research. So you were in the research team? I'm collaborating with the research team because as an in-house patent scientist, I was an in-house patent scientist. I didn't even know there was such a thing. Yeah, I didn't know that patent has an in-house patent, but scientists, I didn't know there was a patent scientist. Yeah, okay.

3:59I mean, it's very broad spectrum, our IP world, so I think Michael is especially clear about that. Well, I think, you know, each company organizes themselves differently. Sometimes the patent people are part of the legal group, sometimes they're part of the R&D group, sometimes they're part of, you know, other groups, somewhere in between. So it really depends, each company will do it differently. And you were with Colgate or something, right?

专利团队放在哪里:宝洁

4:31I was with Procter & Gamble. Procter & Gamble, and how did they organize it? Patents and trademarks were part of the legal group. That's the traditional way. That's traditional. But you would think because they're all, well they're CPG but it's all formulas, you would think it would be more research. So we had the patent group and the trademark group which were part of legal. But within R&D we also had some patent engineers inside the R&D group that was embedded there.

5:06And a lot of those were senior scientists who also knew the business and the science and then they would help us in the legal group get the information we need out of R&D. Because they're embedded with these guys, they probably worked with them for 15 years. That's sort of like what Billy is doing. And so that's kind of what, so he's a patent scientist, but same name for the...

5:26You know, patent engineer, patent scientist. Our role is similar.

中国专利代理师考试,以及 Billy 的发明

5:32And you're studying to do the Chinese... I self-study to do it. Which is crazy. It's pretty crazy. The passing rate is around 10%, or even below 10. So back to the thing I'm working with: at that time I discovered a very interesting way to do battery recycling. We communicated a lot with the R&D team and we found a possible solution.

6:13Let's do it, let's try it. I went back to mainland China, because the lab was in mainland China, and I had only two days to verify the experiment, to verify the solution, and it worked. And we immediately filed the patent. I assisted in drafting it, collaborating with a legal practitioner, like Michael.

6:46Wonderful. So what does that have to do with AI?

7:03Well earlier we were just talking about it, like creation, AI helped with creation and now the issue is with judgment. and I think for that example Billy he was embedded with the experimental the experiments team and so he had to use his judgment is this thing patentable I guess that's part of what a patent scientist or engineer has to decide and so and you just decided based on the experiment like whether that whether it was feasible or not and then you were also telling me some some maybe things about the shady side how maybe people maybe patent things that shouldn't be Right? Maybe like, you know, there are lots of papers and some inventions are not valuable than the others.

7:55Some are completely worthless, yes. Like experts said, some are completely worthless.

8:01I think right now like from what I see is that there was a big discussion in the music industry.

AI 音乐与版权

8:08because there's a lot of AI music software and there's a lot of artists that claim that the AI software somehow really steal their work because they actually come up with new music based on the past songs that is in the market, right? So I think there's one audio site that already come up with their solution is to let the artist who the music got, like the new music that got the inspiration, to get the artist, to let them have a little bit of the earning with the new creation that's created.

8:56But the problem is that people argue if certain code can be patentable, then what if some company just go and patent all the unregistered code? It would be copyrightable, not patentable.

9:16This is a debate that's been going on since the days of Mozart and it's a bigger issue now because of things like sampling and AI yeah but it's a debate that's been going on for a long time I mean how long of a chord or sequence of notes do you need before it becomes copyrightable and not part of the normal musical scene right the whole purpose of the copyright is has to be original so these three notes four notes together original and maybe not probably not but is a three minute song original probably so there's there's whether or not it's copyrightable or not as a basic question and then what constitutes copyright about so that is a separate question from whether or not it's being generated by AI and whether or not artists are being properly compensated it's a different question is it don't get them all confused because otherwise you're talking in two different principles. I learned something new recently because I've been using free music for my shorts and there YouTube has something called

YouTube Content ID 与 AI 生成的歌曲

10:31Content ID and basically it's like a watermark or a fingerprint of the song that's playing so you know I don't know if you watch YouTube but sometimes youtubers they'll say oh I have to like mute for like a while while this background music is playing otherwise I'll get like a copyright strike or something I have to take down my video and so I think the problem is with AI You can algorithmically generate sequences of tones or music and chords, and so you can basically squat and, like, algorithmically generate all of these things, all these songs, register them such that you have a Content ID, register with YouTube, and then if YouTube hears something that's similar to what you've registered, you have the right to take it down, right? There's the theory and then there's the practice, and they're wildly different. So in theory, yes, you're right. In practice, How much is it really going to happen? I mean, the other thing is that if you did that, if you say you registered a thousand songs, like you squat on them, and then I come and I use something which is very similar or even the same, then you say, well, this is my song, copyright infringement, blah, blah, blah. But I would say it's not copyrighted.

11:56Because you generated it with AI, only people can hold copyright. Therefore, you didn't create it, your AI created it, therefore it's not copyrighted. How can you prove that it's AI created? If he sues me, I get access to all of his records and find out that this is copyrighted by AI. like these platforms are so powerful like it might not be copyrightable but YouTube is just going to look and just compare notes and their system will just make its decision and then you'll have to fight but it'll be a pain in the ass yeah that's true it's true and so that's what I mean there's the theory and then there's the practical aspect and they're two completely different things because right now there's a lot of content creator on like not just YouTube but on a lot of platforms

小频道遇上版权警告

12:43they get a lot of strikes and then they are stuck in the rock of the hard place because they might not be powerful enough to be able to have direct access to the company. Most platforms have very, very well defined methods to take down offending material and for them to contest it. They've been working on this for 20 years. But most people would say that their experience with quite a lot of platforms is that, like, it's okay if you are someone with a larger viewership, but if you are like a small standalone, like, let's say, under 50,000, then you don't get your dedicated line. And it's quite a tough battle to fight. Speaking of which, Hong Kong AI Podcast, we're definitely under 50,000. Yeah, so we need, just barely, just barely. So like and subscribe, please. Yes. Yeah, we need to get a lot stronger.

13:44And we were on a break for a while. Well, we were on a break, I had COVID. You got sick. Yeah, yeah. Summertime, kids, whatever. Yeah, so. We're back in business.

13:56No, but this is my question, right? Like, so I explained a way that someone could algorithmically generate music and squat. is there like an equivalent in the IP world in other areas like just

霸占权利,以及独立创作作为抗辩

14:14squatting on concepts? Of course, trademark squatting, patents, submarine patents, all this stuff, it's all happened before. But the other thing about copyright which is unique to copyright is independent creation is a defense. So if you're... I didn't see your stuff yeah so copyright requires copying right and so yeah and so copyright requires copy copyright violation requires copy if you've gone and generated all this stuff even by yourself not even talking about AI and then I come up and generate similar or identical stuff but I've never met you I've never seen your stuff I've never you know there's no way you can prove that I even ever saw or heard your stuff. Yeah. I have no distribution. I have no reach. Independent creation is a full defense of copyright. I see. So this is again distribution is so powerful. It's an affirmative defense. I mean this you know and so unless you like the thing about digital. It's going at the essence of what copyright is defending against.

15:25It's like, I need to copy you for the copyright that you're also taking back. The way that the copyright litigation has developed nowadays in most places, it's big artists fighting against big artists. And so, you know, what was it, you know, if you're a multi-million selling artist, you're number four on Spotify or whatever, right? And then I'm an artist, and as a music artist, I'm listening to music all the time. I'm trying to find inspiration for my own music, but I'm also listening to other people's music. Then there's this unconscious copying type of mentality that goes on, which is of course very difficult to unprove or disprove. But you've basically said, oh, you must have heard my music somewhere, and that's why you pick this chord progression because it's so similar to mine and mine stuck in your head even consciously or unconsciously and is that a good and people have lost millions of dollars on it so it is right the whole

Marvin Gaye 与无意识抄袭

16:33Marvin Gaye thing from like a couple years ago it wasn't actually Marvin was a Marvin Gaye's co-writer, his estate brought, you know, the thing against, what's it called, the "Blurred Lines" thing, and it's not they saying you copied our stuff it's like unconsciously you heard this and this progression was so similar that

Copyleft

16:55therefore... Now, are you familiar with copyleft? Yes. I heard just a couple, like, there's a couple variations of it. One is basically saying we're completely disclaiming any copyright in this material. Okay, so you can freely copy what I've done. Yes. That's the essence of copyleft, I guess. Yeah, basically. It's like an open source thing, maybe. Yeah, it's open source. Yeah, open source for creative works. And that's the other thing, like, I think with software now, because, have you seen this? Like, people are reverse engineering PlayStation games

用 AI 逆向工程 PlayStation 游戏

17:36This is this is this goes back to the enshittification again We don't own anything anymore Yeah, because we're always like it's always it's tied to a subscription But we wanna but gamers want to have physical copies, but then so Sony says no more physical copies And so now people are like I'll just use AI To crack or reverse engineer all the games and just share and so now Sony is like, oh my god what's going on people are like running bootleg ps5 reverse-engineered classic games and so now now no one owns anything and no need to buy anything you don't want us to own anything okay fine We're just copying and re-reading the news. So this is what's happening in the gaming world.

18:20so what's the question the question is what can Sony do? this is the real world like what can Sony do now they're screwed right well I mean there's a lot of things that they can do, they can make things more difficult to copy, they can also just say, hey, we're stopping to make new games, you guys have screwed our business model, so, no new games. Well, here's the other thing, right, there's IPs, there's everything in between, like Pokemon and Disney and there's like people making AI generated content on

同人作品:起诉他们还是合作?

18:56Harry Potter. And they can still go after these guys for copyright infringement. Because it is using the Harry Potter character. It is using a Pokemon. These are copyrighted IPs. So whether or not they're being generated by AI, or whether or not they're being used in the original context or a different context, you can still say, oh, well, hey, you're using my IP. So, there are alternate ways to go after these things, if they wanted to do that.

19:32Yeah, and that was my question. Do you think, like, because there's social capital. Like, let's say I go after a creator who has used my IP, but the people love his stuff. Would you counsel going after that person, or would you say, hey, maybe you can work with them?

Louis Vuitton、Luigi 与乐高电影

19:51I mean, look what happened with Louis Vuitton and the Chinese motif of the tea company in China. Yeah, they messed up. Yeah, they messed up. Right? If they were smart, they would have figured out a way to work with them or give them a license so that every time they sell a cup of tea, they get a cent. Or even say, we'll license you, but no charge. Or whatever, it doesn't matter.

20:15I mean, anything would have been better than the debacle that they ended up doing where Louis Vuitton lost 20% of their business in China overnight. And I looked into it a bit more. Because it was ridiculous. Well, that's not the point. The point is that his question is, what would I have done? What would you have counseled them? If I've got an IP, you know, Luigi or something like that from Mario Bros., and Billy's making a great game out of it, and everyone loves it, you know, It's not like Mario porn or something. That's okay But it's a new new form of you know you make Luigi the star instead of Mario and blah blah blah and everyone loves it because they're so bored of Mario and What I would do is I'd go to him and say hey, let's do some partnership together Let's let's I'll give you a license to it when you give me some royalties or whatever or let's collaborate together on the news That's what happened to us.

21:17That's the way you want to do this And everyone will love both of us though. The LEGO film started with a guy that just made LEGO videos, like making videos of LEGO characters online. And LEGO did try to ban this guy, but in the end they go, oh, well, we'll just buy him out then. Buy him out or give him a budget to create better ones. Because it's building brand equity, actually, if people love what they're doing, it's building the brand. They never decided to make a film to start with, you know, it was really, this guy was playing with Lego characters.

愤怒的小鸟:把侵权者变成授权商

21:54I mean the same thing happened with Angry Birds, right? So you remember the company Rovio, Rovio I think, the Angry Birds was a Finnish company or something like that, some, you know, and so, you know, they came up with this Angry Birds game. Yeah. What, 15 years ago? Yeah. They only had that one game. Nothing else was going on. They had all these grand ideas to make a movie, and do this and that, and blah blah blah. But in the meantime, what happened? All these backpack makers, shoemakers, you know, folder makers and stuff, started putting Angry Birds on every single school supply you could ever imagine. Folders. Yeah, folders and things like that.

22:36Rovio made a lot of money from Angry Birds, but they didn't have anything else. And so, you know, the next game took like five years to make, which was ridiculous. It should have taken like two months.

22:50What did Rovio do? Rovio went and licensed all these guys and said, hey, you've been robbing me, but let me give you an extremely low royalty rate. That then funneled the money to Rovio, the royalties to Rovio, which allowed them to then build a freaking theme park, make two really crappy movies, also do Angry Birds 2, 3, 4, things like that. I mean, there's literally an Angry Birds theme park now. No one goes there, but it's there.

23:21So this is the question, or I think I know the answer, but it's like how much value do you assign to the mindshare and sort of building the brand? So even though they're stealing the IP, they're building something good. They actually kept Angry Birds relevant to an entire populace that would have forgotten it. So there's value in doing that, even though they were stealing it. Well, yeah, right and that's why they converted them from stealers to official licensees, which then allowed them to earn from it So for them, that was a great Great great move. Yeah now There's also the question of What exactly are they doing with it again going back to the Mario Brothers? Yeah, if it's if they're doing more if you're doing a new Mario Brothers game You know and I'm all for that. I'm you know, I'm like, hey great you know, go ahead, keep doing that, let's work together. But if it's something which hurts my brand image.

24:23Then you have to enforce it. That's much more serious. Like Mario porn or something like that. Yeah, I was about to say. Mario porn hub. You don't really want your kids to be playing this game, like, hey, that's the guy I saw on, grrr. So, you know, so it depends on what you're doing with it, how it helps or hurts your brand image, that determines what your reaction should be, right?

24:44And I think with sort of AI tools, and this is what I see on Twitter, people are much more, you see a lot of things go viral now, which have decent production quality.

24:58And sort of the brands, I think it's actually too quick for the brands to react to. Usually it is. So it just happens. But that's, there's another separate issue. So they're just slow to react to, but I would say, yeah, I think, you know, counseling those brands I would say like hey if it's a net benefit or it's in keeping my user base engaged I would try to work with it. So we're in Hong Kong so in Hong Kong

烧给祖先的纸扎 Gucci 手袋

25:25every year you burn gifts for your ancestors and there was a big thing about like paper gucci bags and things like that which had the Gucci logo on And then people were burning them so that their ancestors will have Gucci bags in the afterlife and things like that. And they got all these warning letters from the luxury brand saying, we don't want you selling these, you know, blah, blah, blah.

25:52You're burning our brand. And from a pure trademark perspective, they were absolutely correct. Yes. Right. However, do you really want to make, I mean, there's no confusion that these are not real Gucci bags. They're little paper icons and things like that. And they are viewing it as something positive. They want their ancestor to have it. It's literally aspirational for your ancestor. And they're not making a lot of money either.

26:22They're not making a lot of money for it. But nonetheless, a bunch of these warning letters went out here in Hong Kong and it caused a huge row and people were really mad at all these luxury brands for sending out these warning letters to these small little shops that happened to be having a little paper handbag with some symbol on it. But what ended up happening was the same thing. The reason why these letters went out is because the luxury brands literally don't have the time to police the stuff and they can't police it on a local level. So they contract a law firm or a company in Hong Kong that is in charge of enforcing things. These are like the regional enforcers. They're not using common sense. The more letters they send out, the more money they make. It's like a bad system. It's like how people comment about Western medicine. In the past in India, the Ayurvedic doctors, they get a bonus if no one in the village gets sick. So it's up to them to make sure that nobody gets sick, right? But now with the modern like Western medicine. The incentives are different. The incentives is like the sicker you are, the more money they're going to make and it's the same in this like the system is broken.

27:43But the point is that these people who are in charge of the enforcement were sending out letters, most of them probably didn't check with head office.

27:54A smart marketing person in head office would say no, this is an opportunity, don't send out that letter because everyone's gonna get pissed off at us and we're going to miss an opportunity. I mean there's actually an opportunity to brand these things as official licensed products.

28:12Again get one cent but people are burning a hundred thousand of them every day so there's a lot of opportunity. I would like an official licensed version for my ancestors.

28:24Only the best. Exactly. Only the best for my ancestors. Even better. They can charge a premium. Whatever. But the point is that, you know, the incentives for the people filing those letters and sending out those letters was different from the incentives for the people who actually own the brands. Making sure that those incentives are lined up is a much better way.

执法者与品牌的诱因不一致

28:49Again, back to making the movie or something like that, or making a video game with Mario and Luigi. Billy, the talented game maker. The talented game maker over here.

29:04Make sure that our priorities are aligned. make sure our incentives are aligned. We'll all help each other and have a win-win situation. I had a question.

中国法律与香港法律:有没有可参考的模式?

29:16Switching gears a little bit. China, because Billy is studying for the Chinese patent law thing, is there, like, there's so many layers to it. There's Chinese law, there's Hong Kong law, Hong Kong law is based on British common law, and so on. Is there another area in the world or another time in history where you have this kind of mixture or combination of law sort of interacting?

29:48I mean, is there a model to understand how things will happen? Well, the world for a long time was moving towards harmonization in a lot of fields, in a lot of laws, in a lot of areas, environmental, tax, creative industries, things like that. Everyone's getting closer and closer together. So there's, like, in the copyright field, worldwide, 95-97% the same. In the trademark area, the laws of every country are 90% the same.

30:31In the patent area, maybe 87% the same. A lot of these countries' laws and jurisdictions' laws are moving closer and closer together. And have been for a hundred years. Now, there are certain areas that are not, but generally this is the truth, right?

30:54But China is sort of new entrants, right?

中国如何建立专利法

30:58Well, yes and no. They didn't have an IP law until the patent law in 1984, I think, went into effect. In 1983 or 1984, which is relatively new. A couple hundred years behind everyone else, but they revised it every five years. And so actually, in the past 35 years, they've gotten very, very advanced, right?

31:21And so... I see. Yeah, so... So they're maintaining... That's interesting. They're iterating more on their laws because they're more top-down. They can do that. They have more flexibility. Right, right. But they also recognize that they're starting way behind. And so what they did in the early 80s was basically take a European-German law and said, we're gonna take it. I see. And then now they've adjusted that a bit and they've updated it and there's certain things that were kind of wonky and they fixed it, you know, and then they've been slowly improving their laws for 25-30 years.

IIPCC 与为政府提供意见

31:55So did IIPCC have a chance, were you guys able to advise or do some advisory for them? We've talked, we've given seminars, we've talked to government officials and things like that in many countries around the world to help them or you know think what about this policy is this a good policy or not we've talked to you know governments of Singapore, Malaysia, APEC, around the world, China we've talked with China on a regular basis other places around the world we've talked with US senators US House Representatives we've had actual events on Capitol Hill before so we've done a things around the world because we want to help SMEs have better policies, governments have better policies for SMEs and entrepreneurs and ecosystems. So we've done a lot of that. We're not lobbyists, but we give advice when we're asked about it. We put on programs to identify issues and things like that.

AI 新动向:云端智能体与 Opus 5.5

32:56So what's new in the AI world? Now, what's new in the AI world? Well, I've just been, you know for me, I tend to wait six months. Like, so for example, OpenClaw, OpenHermes, six months ago came out. It looked interesting, but I was like, okay, this is too much work for me. But now, Grokbots and OpenAI Dot, they're basically the same concept, and they'll have a computer in the cloud for it.

33:26So you don't have to have your own computer. And so now it's much more consumer friendly to use that style of agent that's connected to your emails, manages your schedule, browses the web, and so I've been using it to automate some of my stuff.

33:41But otherwise, I would say the models are getting much stronger. Opus 5.5 is a very strong coding model, but it also has a creative sense because I've been using it to create some shorts. It has a sense of timing. It has a sense of how things should be sequenced. It's a little bit scary. It's quite scary, yeah. If AI starts to have a sense of time.

Claude 对 OpenAI:风格与代码审查

34:12And the other thing is you can tell the difference between Claude and OpenAI. The Claude models have a better sense of style. The OpenAI models have a better sense of... they're more concise with how they talk and they are excellent coders and code reviewers. Opus is also a good coder but I would say for review I like for someone to look at and find buggy code I think I like the OpenAI. So in some sense that you would say maybe like like OpenAI is more like Microsoft and then

OpenAI 是 Microsoft,Claude 是 Apple?

34:53Claude is more like Apple. Yeah that's like that's a good analogy and actually OpenAI is pretty much funded by like Microsoft. They would have ran out of money already, right? So the analogy is Apple. And Apple is, yeah, Dario is a bit more like Apple, but yeah, I don't know. Well, the analogy can only go thus far, it cannot go 100%. And the other thing that I think that I talk about with these new models is before you

现在的限制是你的想象力

35:27were always hitting up into limitations more easily when using them, now it's more about exploring what they can do. You're not quite sure of their limitations now. You can try things. So maybe the limitation is your own limitations is not there yet. Exactly, your own imagination because now they, if you set them up to use the right, they can't do everything on their own. They have to have their own tools, so you tell them what tools to use or you ask them to do a search. Because I was like, look at that. To find tools that can do extra work. Exactly. And then see how far they go and I've been very impressed. You look at look at Kimi and I think it's Microsoft that you look at them the way they built our slides. Kimi does a much better job. Yeah, that's what Bruce uses for his slides.

36:22Yeah, it's very good. Kimi is quite good. So it's not just Claude and OpenAI, but yeah I wondered are we like is next year the year that we realize it's like our limitation is the problem? Well, I think, because the big providers, they have their own internal models that they don't release, that they use to help them train their commercial models. So I think one of my conspiracy theories is

AI 实验室在内部模型看到什么

36:55the reason why all these people in these labs think the world is gonna end is they're seeing things from their internal models that we are not aware of. And so they're seeing capabilities. So they are like Skynet is coming. Yes, because they're in the middle of designing Skynet. No, Skynet is already here. So the example is... He doesn't know what Skynet is. Terminator 2. So basically, learn your history man. I'm a regular guy. I'm using Opus 5.5.

37:31I'm amazed at what it can do. But they, Anthropic, they have something called Model 2. It's basically an unnamed model that they use to train their Opus and Sonnet and Haiku. So if they have another stronger model that they're using to train what we see and we're already amazed, Can you imagine what they see? I can't. It's sort of hard to imagine. But what I don't understand is, so I only look at it from the sidelines, I don't know.

38:00First of all, there's this idea that all the super-rich...

38:06Bunkers. Like building bunkers in New Zealand.

亿万富翁在新西兰的地堡

38:10Yeah, yeah. The world's gonna end. If the world's gonna end, I'm not sure that I want to be stuck in the bunkers in New Zealand. Do you need to learn how to spawn something that is much better for you? I don't know. I just don't know whether, like, if the world is going to end, I might not want to survive in a bunker. You know, but maybe I just don't have...

38:34Have you seen their bunkers? Their bunkers are like 20 feet tall, very spacious. I don't care what bunkers it is, it's just the human nature that the moment you tell me I like if you don't tell me that I can't leave this place I might just get very happy living in a very small square but the second you tell me that this is the limit that way you can be mentally you go, oh man. You do have commitment issues. You can see that now.

39:13You can leave, you just can't come back. Our mind is the prison. I just I think especially with the people who are building the bunkers the people who have the ability to build the bunkers in New Zealand I don't see how they can just live in bunkers because they are the one that would have to jet around the world like six times a day. This is what I think is happening these billionaires they they have to compare they get around maybe they talk to each other they have like smoke cigars and they're like how big is your bunker, oh where is your bunker located, how much food does it have you know like yeah that's what I think it is yeah but anyway you know how do we get to We're talking about the world again.

道别,以及纹身展

40:03Speaking of the world ending, I need to go. So our world is ending. Now our world is ending. Well thank you very much Michael. Thank you, good to see you. Thanks for coming out today. It was fun, it was good. Not what you expected, I'm sure about that. No, I loved it, it was spectacular. When you said tattoo, I was not imagining this. You know, when you hear of tattoo, being from the East Coast, I think of a grotty, dark room in some basement. No, no, that's just you. That's what tattoo was like 50 years ago. Yeah, I'm an old guy from the East Coast, that's what I think of when I think of tattoo. You think of stuff from 50 years ago that you were not used to. It was interesting. It was very interesting to me. I really enjoyed it. It was like an art show sort of...

40:53It was. it was like an art show. It was a little bit edgier. It was a pleasure meeting Gabe. It's very interesting. He's a very enlightening guy. I think we'll definitely get him on. Yeah, we'll get Gabe. We can even get John for a sit down if he's free. No, we definitely will have John as a sit down. Yeah, I gotta sign up. Go check out his installation at PMQ, you'll be amazed. No, no, wait, wait. No, no, John is inviting us to go and try out the tattoo. I want to go, I want to go. PMQ. Whether you want or not, we're invited and we have to go. So the choice is never, like, whether you want or not, it already affects. You're going.

相关文章

我拿着 Cursor 的旧制方案一年多。它自己的文档说,这个方案在我下次续订就终止12 min read中国 AI 模型指南:DeepSeek、Qwen、ERNIE 及更多12 min read从 Kimi 到 GLM-5:在香港上市的中国 AI 公司6 min read
在 YouTube 观看 在 Spotify 收听 在 Apple Podcasts 收听 在 Overcast 收听所有节目