今日概览
7
节目总数
7
订阅源
9h23m
总时长
Top Story
【All for Today】🎙️ 每日播客精要全景图
首条总览
42章经
(1)
锐评国模与 Neo Lab,畅谈 RL 人生与 RLaaS | 对谈 Mind Lab 首席科学家骁腾
1h 2m
2026/11/07
📝 AI
总结
.🎼There's something there.啦啦啦。🎼我们今天很开心请到了my lab的联创和手机科学家肖腾。hello大家好,我是my lab的联创和手席科学家马肖腾。然后大家可以叫我ponny。我是17年在清华自动化系直播。然后23年毕业,然后继续在清华自动化系做播后,然后25年10月的时候和我的两位好朋友陈凯杰和ang一起创办了 lab。现在我一直都是在做强华学习和后训练相关的工作。然后今天也非常开心可以。来到42张金,见到去凯老师本人。好的,非常官方。你看你待过公司都还算是有点争议的。我不知道你怎么看这些争议啊什么的。😊其实我觉得大家对于my lab的争议在于之前马卡龙那一段,对吧?其实你们现在的业务已经跟当时很不一样了。这也是为什么我们今天会有这个播客的原因之一嘛。可以先简单讲一下大家的这些争议和你的。你或者你们内部的一些看法。嗯,首先我感觉就是呃有争议是好的,就是创业肯定是不要回避大家的讨论跟争议嘛,就是只有。有人在讨论才说明就是被大家看见了。其实我想做的事情一直都是关于模型训练,然后关于。post train的一些事情。然后因为我自己就是学青蛙学习嘛,然后我是非常理解RO,然后我一直在找可能能通过R去做的一些真正有用的一些事情。所以后来我加入了man lab,然后去做后训练的posting lab。对我来说呃,无论它表征出来是马卡龙这样形式的app。还是嗯我们现在在做的R service的一些to B的业务,它的实质都是做超大的这个模型的RL,然后来让他在某一个垂语的这个performance上面去有效果的提升。我更希望可以把post string理解成有点像互联网时代的推荐系统一样,它是一个fuamental的技术。至于它上面这个长出来的是。西瓜视频还是抖音,还是。今日头条我觉得可能不是一个现阶段需要特别去考虑的事情。然后尤其是我作为一个reseer,我不是特别关心,就是最后表现出来的样子是什么样。嗯。😊所以你追求的还是说模型训练这个效果本身。他最后dber结果怎么样啊,就得。对对对对对,你的选择其实也是都是奔真目标啊。所以就我特别好奇,你为什么没有首选去AIL。嗯,因为我觉得首先我自己是。非常喜欢创业的人,就我认为创业是一种生活方式。这个也和我自己做很多L的research有关,就是。做多Lre你就发现就是。限制这个模型能力的完全是exspiration。和他的标彰能力。呃,这两件事这个能不能在。讲细一点,就是大家该怎么理解这两件事。嗯,就是L里面有一些。非常基本的概念。然后这些概念都现在已经变成了我一种人生的哲学,在指导我做我的各种人生的规划。对我们之前跟吴意录过一期博客,那些博客他里面就提到了一句话,然后大家也特别喜欢,就是人生本身就是一场强化学习。是的是的是的,然后我觉得在大的Lpro里面,然后我 learned到了三个对我来说非常有帮助的事情。第一个就是。😊exploration跟exloitation就是这个探索与利用。其实是RO非常入门级的一个概念,就是所有RO能work的算法。里面都有探索和利用的平衡。很早期的时候这个。我们做阿瑞小游戏的时候,他就是随机选10%的随机动作。不要看这10%的随机动作,是看上去非常低效的一种选择。就是如果你不做随机。这个动作的探索,你是无法达到这个最优解的。这里面有很多的pro。包括今天我们在训大模型的时候,我们也会控制模型的商。很多情况下就是训的过猛了,然后他的商探索了,然后他的泛法性就会变差,这也是exoration的一种体现。所以研究里面我就看到了,必须要有一部分的ex。那我觉得我的生活里面其实也是一样,就是如果我不跳出我的既定的思维去做一些我自己觉得有意义的一些事情,我就永远会困在那个locopop里面去。所以我一直都是1个90分主义者,就是做到90分,我就会觉得OK我应该去再花剩下10%的精力去explore something new。那在模型训练跟强化学这个事上,你自己定义你是90分还是多少分?嗯,我觉得现在显然我们应该是90分的一个状态吧。就我觉得很多东西我们还没有做到最好的一个状态。对,但这里还是有矛盾点的嘛,对吧?我可以理解你个人生活当中是一个。90分的状态,但是在创业这件事上。肯你还是追求在某一个点上能做到100分甚至更高的分数。我觉得你说的非常对,这是为什么我会跟。安卓跟凯杰一起来创业,就是我。do the part of exploration,然后他们来去。一起让整个team变成1个100分的状态。OK某种程度上弥补了我在做research探索上的。这个商业化的不足吧。明白明白,对对,这是第一个点。就还有两个点。对,第二个点就是知行合一L里面其实是有这个policy和。ベル。就是策略和值。比如说我们在做一个决策的时候,然后他可能有两个逻辑。第一个逻辑呢是我知道应该选哪个东西,然后他选择的概率是什么?第二个逻辑是它够好。比如说他这个A选项是个5分,然后B选项是个3分。这个时候我可能这个选。A选项对吧?😊那选择什么东西的这个策略的概率和。我们给他做的评分,这个事儿他是对偶的。就是说你知道他的评分以后,你就知道他的概率是多少。这也对应着就是LO的。两种不同的算法跟。流派吧,算是一个就是。policy itition,一个就是 valueition。他们都可以得到很好的这个策略的结果。但是。啊,无论是poit还是valit,它里面都一定要有一个要求,就是。你的值和你的策略要是一致的。那如果他要不是一致的那一定是有一个东西做错。要不然就是你的策略更新。晚了,对吧?要不然就是你的值更新的不够。所以就是。而有交给我应该成为一个知行合一的人,就是我应该。不停的在update我自己的价值观,然后我自己的判断。然后当我做出这个判断的时候,我就一定要去。colick experience,然后我一定要去。做一些事情,然后去看我的这个判断是否是正确的。反过来再去修正我的价值观。其实就是一种这个实践论嘛,嗯嗯。然后第三个其实就是长期主义,这也是我觉得现在。目前的清华学习里最受诟病的一点就是目前的大模型的这个训练是没有长期主义的。我们都是在用GRPO去训练。GRPO是一个单部的这个。决策就是问一个问题,然后他就回答一次。然后无论他是enttic的还是。只是一个长的reasoning任务,他就是只回答一次。实际的我们的人生是一个single pass,它是一个非常非常长的。在L的这个基础的理论里面有一个概念叫做这个折扣因子。他就是说呃你要把你非常非常远的这个reward。做一个discount算到当前的re里面去。每一步这个discount的 reward加在一起。就是你现在对这个事的价值判断。那么比如说可能我明天发生的事候,我就肯定是我最重视的这个事情,对吧?但是十年以后发生的事情,那就有点没谱。那我给他一些dicount。那这个discount呢,他就决定了这个agent他的视野。其实他some号可以理解成是时间分之1的这样一个概念。如果这个discount是0.99,那大概就是我的time present就是100。这样子,然后未来看未来的100部。这个discount就是在L里面,你会发现它不能太大,也不能太小。然后如果太小,那就很短视,他看不到未来会发生什么,对吧?😊然后如果他太大,你会发现他不收敛。他一直震荡,因为未来有很多这个事情是很noise的,然后他就不收敛。这个也知导我就是要有一个比较远视的目光。但是可能你要。有一个好的plan。所以不去Iab,就是我觉得我需要一些exspiration。然后我觉得。我希望的人生是一个创业状态的人生。创业有非常多的好处。他是最fit,而有这个loop的。因为他可以不停的colect real experiments。😊然后你在一个大厂的环境下面,其实。拿到的不是全部的信息。你拿到只是一个局部的信息,这个时候是不利于你去。做你 value6你的这个policy更新的。然后同时你在大肠显然也没有办法做ex,你只能做。一个比较local的事情。所以我一直把创业当成一种生活方式,而不是当成一种。投这个目标我一定要干成什么事情。这个过程中我很享受这个带来的快乐。哎,但我在想件事,就是你看整个AI包括像智普啊kimi等等这些公司,其实他成立也就34年的时间嘛嗯。那个时候你应该在读什么博二左右。呃,那没有。智朴成立的时候,我应该也。博三博四了。对Pmi成立的时候,我应该博士最后一年。对,所以据我了解你的过程中,其实也没有去这些公司实习啊什么的嘛,是吧?对我觉得这个事情就是一个反过来的,就是我已经我不太看,就是我。在短期的外流是什么?然后我首先决定就是我的人生的plan就是。第一选择可能是成为一个创业者。然后其实我在这个过程里面是尝试过去找教职,然后成为一个教授。当教授其实原因也很简单,就是第一是我发很多论文。然后第二是。我很喜欢跟学生打交道。帮助一个 junior的学生,然后他探索出来很好的一些。成果的这样的感觉。呃,但我发现就是当professor。的大部分时间都不是在处理这些事情,都是在处理一些基金啊,在处理一些。😊比较枯燥的事情。所以我后来就。放弃了这个路,然后我就选择了。自己创业嗯哼。然后我觉得大家对于。能讯模型测时呢一般会觉得说他应该是出自于某一个现存的AI app。至少在里面做过,但你没有这些经力的话,市场上。其实是对你的认可啊,口碑什么都还挺好的对吧?对包括你们做的模型其实也还挺好的。你说这个来自于什么?就为什么。你没有真正的去在这些厂子里面试复模型,但是安宁做的很好。首先我肯定不知道,因为我没在场子里面修模型。😊然后但是我觉得就是我是有一些。在更大集群规模上做模型的经验。这个事情也源自于我痛苦的读博经历。在我读博期间,我导师让我去做了1个5000万的一个比较大的project。大概在18年开始。那个时候才刚刚有open eye,然后当时是希望可以在一个呃有1万核CPU和100张V100的这个集群上面去做一个大型的对战式的训练。我其实很早就了解整个L训练的全排版,然后包括英法的这些东西。那个时候。瑞在这个0点。20.3的版本,然后排套池也是在这个0.30.4的那个非常早期的版本。我我那个时候就是在用锐啊KYS这些东西去做大规模集训的训练。这是第一。然后第二我感觉虽然说大的模型训练它会带来一些新的挑战,但是。他的本质其实跟我在读博的时候做的事情完全没有变化。就是。我也是每天在面对reward hack的问题,然后也是在面对这个怎么。去做更好的similulator。然后怎么去设计各种各样的reward跟benchmark。嗯,只不过是网络变得更大了,然后他的表彰能力变得更强了。做实验变得更贵了而已。哎,你觉得你自己跟别人比有什么特别的学习方法吗?呃,其实我不是那种特别典型的那种ner,就是。像竞赛出身或者是怎么样,但我觉得。我自己感觉我自己的这个泛法能力比较强,然后通用性比较强。我对各种东西都非常感兴趣,然后。我喜欢在各的事情上面做一些尝试,这接跟我那个。90分的哲学标有关系。所以我其实在博士期间,我阴差阳错的做了非常多的research的话题,就是。离线强化学习啊,多智能体强化学习这些比较。现在是。彻底震的RU了,就是现在是经典的RU了,然后也做过那种非常老那种tableler的RU呃,只有在影印版的那种六七十年代的书里面才能看到的各种。😊markf的一些性质啊什么的这种seory我也做过,然后我还做过一些。现代的整各种棒的RO,然后也做过一些。实操性的一些项目。所以我觉得。因为我在R的这个做的project比较多,然后我也跟很多不同的。老师去打过一些交道,就是。很多不同老师,他带来很多不同的inset。我发现他们的inset其实汇总在一起,其实是有一个可能更mattter的一个抽象。然后这个东西对我帮助非常大。对,这个就是我刚才听你说你那个901分的理论的时候,我想讲,其实他就是。我一直觉得说你在某个领域,比如说真的做到90分以后,再往上涨,其实那个难度和编辑效应。都很低。对了,但是你。...(已截断)
Modern Wisdom
(1)
Your Home Might Be Destroying Your Health (& how to fix it) - Michael Feldstein - #1173
1h 22m
2026/11/07
📝 AI
总结
So I've got a bunch of products here. Okay, that a lot of people will have in their homes. that I want to get your opinion on. But first, before that, before we get to this, you walk into a hotel room. There's a street hooker. There's a bag of white powder might be pure, might be cut. You don't know and a potential mold problem. Which one are you most worried about. I guess I'm most interested. I'm most worried about the mold man. It's what you're most interested in that might be something different. It's the thing I could least.Control, if I have nowhere else to sleep that night, I could throw the bag away. I can ask the hooker to leave, but the mold is a little bit more difficult. Yeah, I think this air quality thing is one of the big three areas in health that everybody's going be interested in over the next few years. air quality, mold toxicity, light pollution, Those are the big three for me. I think water quality was a big deal. And now we're kind of getting to the peak of that, People understand any clean water. How did you get started working in air quality.😊What's the story. Also, I totally agree with that. And I feel like.Some things are trends.But some things, once the awareness raises.They don't go back the other direction. Yeah, like some things come and go. Red lights are kind of popular now, ish.Coal plunges. But like in 10 years, I'm not so sure agreed.Water filtration.Once you start drinking clean water.And then, you taste.Once you get the water awareness and you can taste.The chlorine and the lead, and you become water aware. There's no going back.To.Not drinking filtered water.And error is the same way.I say, the only thing that.Is rough for people about breathing clean air as You might become an air snob.And that might make it a little bit more difficult to navigate Airbnbss and hotels and people's homes.So I got into this by being in the disaster restoration business. So floods.Fires, hurricanes.Modern mediation.So I wasn't your, like, local.Flood guy that you call, if there's a.A toilet leak.We were the ones who would fly to.You know, Hurricane Harvey or Alberta, Canada or wildfires in California.Going to the biggest disasters in the country.And remediating.And.The big moment for me.Well, first of all, when you remediate from mold, what does it even mean to remediate.You need to.Issolate the contaminated space.You need to remove the physical mold.And scrub the air.Those are the steps.And you would always measure.The air quality before, during and after.And some of the cleanest, freshest air I would ever breathe was after we were finishing a moldern mediation job.Cause we had scrubbed the air for four or five days. And then I'd go home. and I'm like.it was cleaner at that job site.And.The big moment for me, my aha air moment was in 2016 in Fort McMurray, Alberta.Biggest wildfire in Canadian history.All hundred000 people in the entire town were evacuated for a month.We went there.And.Everyone's homes were so smoke drenched.So it wasn't fire damage. It was smoke damage.But this.Such significant smoke damage.That your entire home had to be gutted.And cleaned.And what was unique about this was when the it wasn't just a..It wasn't just smomores in campfire.When you had the 5,000 homes that Bt.And the cars, every chemical paint.Drywall.The materials that we're combusting in those.Created a very toxic smoke.嗯.😊So we go into people's homes.We'd clean them. We'd restore them. It could take weeks. It could take months.But this one time we finished cleaning a home, we got a call a week later.Baby was in the hospital.We go back to the house, we test the airre like, ah, it's actually really recateaminated again.Then we go whenever you're testing air, you always test outside and inside.If anyone's ever testing air just inside, they do not know what they're doing.You always need to take a control sample of the outdoor ambient air and compare that to the indoor air in that moment. That's how we have the Delta, because otherwise, you could just have bad air.In the local environment, like if you'd taken an air reading during the fires, you'd have been like, the air insides horrible. Its like, yes, so is the air outside. It's just the same. You know, you're not supposed to do mo testing.Like 48 to 72 hours.After a rain.But no company adheres to that because it would.So.😊A lot of people get, they get mold testing.The results are in the millions.You look, it rained the day before.Like, did they do a control sample.If they didn't, of course, there's multiples in the air.Doesn't mean your home is contaminated.It's just what was in the air that day.So we go in there, we take the outdoor air very contaminated. after a bad wildfire, the air can stay bad for 4,5,6 months.Gets in the soil.It's the homes that are burnt, it's in the ash.So we called the insurance companyre like, hey, guys.Tanya's home is recateaminated.I suggest.She moves, leaves. She heard her family have to leave town.Wait in a hotel room.Let's wa...(已截断)
Decoder with Nilay Patel
(1)
AI broke the cybersecurity world. Now what?
1h 3m
2026/11/07
📝 AI
总结
🎼I'm in a garden. and each week on H hour, I invite new and old friends for a drink and a fun conversation at my kitchen table in New York City. On this week's episode, I'm joined by Jennifer Garner. We talk all about what it means to be a good friend, Jen's encyclopedic knowledge of my recipes. and the Paris tour I put together for Jen and her daughter. New episodes of Happy hour are available every Wednesday, You can watch on YouTube or listen wherever you get your.😊Podcasts, cheers.🎼 Hello and welcome to decoder. I'm Ne La Patel editor in chief the Verge. And decocoder is my show about big ideas and other problems. Today, I'm talking with Alex Stamos, chief security officer cognition and a longtime cybersecurity expert about what a mess security on the Internet is right now. This episode is part one of this special series, looking at AI in real life. I always enjoy talking Alex, because he's so direct.😊And you're going to hear him say that the multibiion dollar cybersecurity industry, as we've known it.🎼Is essentially over.Due to the unrelenting AI boosted waves of hacks, breaches and attacks.🎼And all of those are only the beginning. Both script kitty hackers and also nation states like Russia and North Korea are breaking into systems all over.Thanks to AI, helping them both detect and exploit vulnerabilities.🎼All at the same time, it's got a real frontier feel to it. You're going to hear Alex say it feels like the 90s all over again when the whole idea of an always online world was brand new. That much disruption means there are a lot of places where bad guys can get in while everyone else is left scrambling. Alex isn't a domer in the long run.But he does think we're in for a quote, dark age. while we try to figure out how to adapt this new era. There's a lot going on in this conversation. and leaving all the way to how AI powered misinformation and Slo.🎼Might be managed on our social networks before we get going.Doive me a favor and subscribe to Dakota on YouTube where we put out new episodes every Monday and Thursday. Okay, Alex Damos and how AI is upending cybersecurity.Here we go.Alex Damos, you're the chief security officer at cognition. You're also a lecturer at Stanford University's Cyber Policy Center. Welcome back to Dakoder.😊Hey, thanks for having me man.I'm excited to talk to you. It's been a long time since you and I chatted in the.😊What you might call the content moderation era of Internet policy discourse. I feel like you and I talked a lot.About how all that was working, about election security, all those sorts of things. We're in a new era of big problems on the Internet, particularly around cyber defense, cybersecur.え?You've been thinking about this stuff a lot. You referred to.The past few weeks is Security Awareness Month.Tell me what tell me what it's going on.Just give the, give me the way of land right now.October is, you know, traditionally security Aware monthth, right, Like lots of companies.Run.Events and, and deal, both internally.And then tell people, you know, change your password, use multiple passwords, use a password manager.And what has happened over the last month as we've had.You couple of these big bugs, P P.Note and such.What's been less discussed is.A huge number of companies have been reached.あ。You have a series of.AI related companies, the neoclos.They have been breached into.Succession.You have the Binance hack.Which is the largest heist of anything in human history. You have the cursor incident.Where., one of the things that's going on here is these hacks are being chained together.So.It looks like from what we know right now.One of the companies it was broken into was fireworks, which is a.Prominent provider of AI services to a number of companies.From the fireworks hack.APpiI keys were stolen.That allowed people to break into cursor.Once they gone to cursor, they then pushed.A bad plugin, which went to 5000 more companies. And that caused.5000 potential breaches.So it's just this kind of.Permanent.Continuous breach that is going on throughout the tech industry.And one of my friends.Said.😊ACil friend of mine said to a bunch of us, is anybody else tired?And, and everybody said, yes.Because effectively, we have been at Deffcon, too.For all of October and now going into November.So I'm going to ask you some really dumb questions here.Who is.Behind all these attacks is it a Russia, Is it North Korea. Do we, do we know.It's a variety of different groups. It is.It is not.Just.😡Because adversaries have decided. But what has happened is.From my perspective.We are at the start.Of the takeoff of adversary capability.Enabled by AI that the open weight models.Have now gotten to the point.At which you can automate most of the kill chain. so this kill chain term is something we stole for the military, right?あ。In the military, they mean to actually kill something there are the steps you do to blow something up.In cyber.The kill chain is the things you do to have an effect on ...(已截断)
Y Combinator Startup Podcast
(1)
How To Build The Future: Paul Graham & Amjad Masad
44m
2026/11/08
📝 AI
总结
🎼I think with AI, everyone will have to go through this period of denial and then depression and then eventually acceptance. Reples like this iceberg, where there's this visible bit above the water that's like, look, you just tell it in English what to do. And it writes your code and then underneath. there's all this stuff for deployment and sand. Like, I don't think we describe ourselves as5 coding now, we think repplelet is the platform for the transformation to soft driving companies, more than other companies, though, I think.You guys are have been living in the future. If you had to say, like, what's the real DNA of re It is living in the future, for sure.🎼So you guys basically invented vibe coding, right, And when you did it, like in 2024, this was, which it sounds, it's like two years ago, but it seems so long ago.😊Seems like a technologically.How did you even know this was going to work, Because it's obvious now that it did work, right, But at the time.It seemed crazy. It seemed crazy to me.Moveful for us, it was.Incremental, it wasn't like for the rest of the world, I felt like a.Big departure for how from how people did things. But we've always, you were already writing more and more of the code. Well, no, it was.It was sort of like the Y C.Kind of.Eths of.Look at what your users are doing and, and, and just that will give you ideas on what to build. because you were already doing auto complete. We're doing auto. And so how much of the code were you writing before you went to like full vibe coding, where you tell the thing in English what to do. Yeah, where you're just generating code with auto complete.How much of the code were you already generating then, I mean, you, you could, you could probably say like 50,60%.Really, so that's interesting. Yeah, even before vibe coding, you were already 60% vibe coding right.And that was how you knew it would work. with the user you thought we can make the 60%. Why not make the rest.Well.Kind of, I think that, the thing that was.Very enlightening.Is watching some of our users code.And so some, some users.Where.Doing 100 percent.Of their code using AI. But the way if you know, uses is you're talking about once you released5 coding before before that, before that, I'll tell you how are you doing.So, they would.嗯。😊Open chat in one window.They would open raplet in another window.This someone who's never quoted before.How could you tell that they were pasting code out of chat EptT, I invited some users to all. you would literally watch them and would watch them.😊Okay, how do you do this.And, basically.They would like, copy the code, put into Rplet.Reple look like produce some error.They would like, copy there, put in your chatGBT.あ?Yeah, and I copy the fix put.And it was such a manual process. And I looked at it. All we have to do is make an API.Well, we have to automate the entire process. It wasn't obvious. It wasn't easy, because.Back then.AI models would like do a couple of tool calls. Oh, you know what's different, The difference is, it's not just you have to make automate the back and forth.You have to understand the errors.Right.Exactly, yeah, that's hard. Yeah, and, and you have to loop.And the terminal condition should be that the program is working, I suppose. Yeah, that's still true, right.As opposed to kind of like.We'll do one term, right, because that's how ChaT sort of worked. That's how most AI tools worked.And so can you loop, So this is loop indefinitely go to completion. Yes, right, which is, that's the new thing. That's what Claude does. Yes, yeah, so that's what we demonstrated. That's what Replelic first demonstrated is. So we demonstrated a couple things one.Is that you can run.LLMs to completion.2, which I bet was news to the model companies. They did. They actually reach out to us and they were like, we're so surprised that our recplication works.How did you make it work? And the crazy thing, Paul, is that.The harness circuitca 24.Is 10 times more complex than the harness today.Because the models were less capable. Yeah, so as models get more capable.Like.Wite?The kind of the engineering you have to do.To make them.Become agents is less or less so. Does that mean it's easier for people to compete with you. Yes, but we have to do.Well, that's the thing about AI business today is that.You have to just accept.That everything that you're doing is like.Getting commoditized constantly.And you have to like be doing the next thing. Do doing the next thing, right.But the other big thing that we demonstrated is that AI can provision infrastructure. So up until that point, I, you actually have infrastructure to provision. Yes, that's the thing that's difference between you and all these competitors rightly. So we had built like data. You'd spent all these years suffering, doing system software, right.え外。So it was an literal like tool called a spin up a database.You know, spin up the Redis, like host this, you know.And so.What.People found so surpris...(已截断)
音乐进化论 MusicEvolution
(1)
画地为牢之外,重新认识音乐人许飞
1h 15m
2026/11/08
📝 AI
总结
🎼好久没有见到你。😔我在猜想是因为天在下雨,我用时欢迎收听播客音乐进化论。我是最近跟大家一样,都在综艺以后会有驱离可生科死的沟沟。相信在过去的一两周呢,大家跟我都一样,就是开启了一种对利蜡体上头,对画地为牢的大文豪许飞,发起像素级的考古当中,真的越看。🎼看越有。因为最近大家考古出来的物料实在太多了,所以我真的。营养均衡。我发现我很喜欢看。许飞写的文字。他真的有一种非常惊人的表达能力。就是人怎么可以把那些如此难以言说的情感和情绪。变成一个如此具体的可以被知的画面呢。有人开始考古他早年间的采访。生活经历社交媒体上的常文,包括在粉丝群里给大家的手写信。也有人重新听起了他的歌,就比如我我这次重新听了他很多歌,才发现,其实原来有很多歌。是我自己听过。但听的那个时候并没有意识到这是许飞唱的。但是啊。不管怎么考古啊,我发现其实他都是一个非常言行一致。知行合一。十分经得起推敲的人。所以呢我也很想先感谢一下这档综艺就是。😊不管未来两个人是否重修于好。至少他做了一件非常有意义的事情,是让一些。原本可能已经被市场或者互联网记忆遗忘的。许飞的音乐作品重新获得了一些被听见的机会。很多人对于许飞的印象可能都始终停留在一些关键词上。比如说父亲写的散文诗。那年夏天。包括中间和电视剧。一起来看流星雨的合作曲,成为了很多人的青春记忆。所以好像许飞呢他是一直被放在一个很固定的位置上的。是那个抱着吉他唱着民谣,有一些文艺和清冷的女生。但是如果我们真的把他过去20年的作品从头听一遍的话,就会发现。这个印象其实相当不完整。他写过城市和故乡,写过远行和相遇。也写过爱情里的犹豫和笨拙。还有成年人的思路。和他的社会观察。更写过一个人如何重新找回自己的梦想。他甚至还在自己的音乐中尝试过grange。浩朋克这样的摇滚表达。从早期的清新民谣到中期的流行摇滚。再到后来更加个人化的独立创作。他的音乐其实和他的生活一样,都是在发生着变化的。在先论过天价赔偿之后,他开过了吉他私塾。在争议满天飞的同时,仍然坚持作着巡演。他跑过马拉松。坚持读书、写作。重新理解人和人之间的关系。再到最近,他和诗人余秀华合作完成了一张。诗歌音乐专辑。对此,我对他的印象有进一步的升华了。我觉得他是一个音乐诗人。而不是一个被诗歌所加持的歌手。如果说有的音乐人他是在不断寻找一个属于自己的标签的话。那许飞给我的感觉。他是相反的。他好像一直在努力摆脱一些别人给他贴上的标签。所以这期节目我们就从那些。可能你还没有听过的许飞订阅出发。去重新认识一个在热梗画地为牢之外的。音乐人许飞。听听他是如何从一个被时代记住的选秀歌手慢慢的。变成了一个拥有自己音乐世界的创作者。也想跟你聊一聊,一个人如何能够在20年的时间中始终保持着一种。重新出发的。🎼能力和勇气。的是。🎼日前。我了下。🎼有谢。是不会怜悯我的。😔あ?あ?🎼 🎼 う。🎼 🎼 🎼う。 🎼Yeah.う。🎼 うん。うん。🎼う。 .🎼 う。う。あ?🎼 我们。あ。ちゃ。🎼社長。あ?如果要用一首歌来认识早期的许飞。我会推荐淡淡的歌,或者说他的这张专辑。这首歌收录在他2008年的首张个人专辑。恰许同学年少中。🎼这张专辑的制作人阵容很豪华。陈生、袁惟仁、周志平啊都是非常大拿的选手联手。徐飞本人呢参与了6首歌的制作,而专辑当中的intro就收录了。许飞和陈生两个人的呢喃。后面接着是火车启动的声音,然后马上就过渡到了下一首歌。在我听起来是一首。天作之合。而其中呢,飞碟与大头、痊愈等等三首词曲是许飞自己独立完成的。许飞海首次尝试了rap。就在飞碟与大头这首歌当中。🎼同学。Babyさー.おいたくよ。べ事けっては。🎼同学出了名的老城牌。他们都说他像个京剧老师。🎼他同学最喜欢读书哦,彬彬最爱漫画书那个张同学,漫画都不会读。拿他没辙,像个米老鼠小。🎼talk about。你说着说下。好用。不し。うん?🎼表达出就是一样的话。不够小花。🎼大大的。정大。🎼会た。飞到飞利浦中永远带你起飞。🎼飞翔开。而为什么我会首先推荐这首淡淡的歌呢?其实2008年对于华语流星乐来说是一个。很有意思的年份。因为那个时候互联网音乐正在快速的改变人们的听歌方式。彩铃NP3音乐网站。还有各种的电视剧原声,它构成了很多年轻人的音乐寄遇。而同时,校园民谣的余韵还在。一把吉他,一段简单的旋律。🎼加上一些能够让人共鸣的歌词就足够了。许飞的演绎,他有一种很自然的松弛感。他并不是那种很强调技巧的歌手。或者说用高音去制造情绪高潮。他更擅长让自己的声音就放在这段旋律中。让歌曲本身像一段轻轻讲出来的话。这其实也是他早期月当中很鲜明的一个特点。有一点少年期,有一点不确定,还有一点对于生活的好奇。而且呢。淡淡的歌,这首名字就很有意思,我们总是习惯性的把音乐和某种强烈的情绪情感联系起来。比如说失恋就该悲伤,告别就该痛哭,青春就该热烈。但是生活。你会发现最终我们都活成了一个蛋人。因为绝大多数的日子都没有那么多戏剧性的情节。所以这些没有被特别命名的情绪,构成了我们生活的大部分。而淡淡的歌,他让我想到的。🎼就是这种非常超前的。日常感说我傻。🎼这一首淡淡的歌。给你要。。🎼说说心里话。播一首淡淡的歌给。🎼你よ。啦啦啦啦啦。😊🎼这一首淡淡的歌给你啊。让你回想他。😔🎼我会一直拥抱着你。告诉他们。🎼我也会牵挂。😔Yeah.🎼Yeah.说你想说的情话。🎼做你爱做的土。我。🎼我已不想再挣扎这份感情有多复杂。关于你的回答。🎼我在心里牵挂。😔这样煽情的歌唱到你想家唱情歌流眼泪。🎼不会。你想着些双人床的咖啡。🎼只少有人有个人来陪。写一首淡淡的歌给你啊,让你说我傻这一首淡淡的歌给你啊,说说心里话,播一首淡淡的歌给你啊。啦啦啦啦啦,这一首淡淡的歌给你啊,让你回想他我会忆。🎼一直拥抱着你,告诉他了,我也会牵挂。 接下来这首吉林到北京是他首张专辑当中很陈生的一首歌,就有很多有意思的小细节。因为他的曲子完全是陈生写的,而歌词是他和许飞一起完成。所以在听感上,我们既能听到那种很流浪的气质,也保留了许飞这种倔强和直白。这首歌里是有很多许飞真实生活的原型的。他曾经说在唱到吉林到北京的时候,。🎼脑海里闪过的画面是列车开动前,妈妈眼角的一行泪。徐飞出生在吉林四平,年少开始,她就离开家乡到北京,也因为追星的缘故,开始了自己后面的音乐生涯。而这首歌基本上就是他从故乡到异乡的成长主角。不知。🎼已经。多少事。🎼And.没有。🎼天动生明。的吃。🎼更醒了。就想要。🎼燃想。你说。🎼安全。年少时候。🎼梦想是。 🎼春天时候。おちゃ。🎼边张随。放。🎼飘散了6。都是。🎼我的。可是。🎼是谁都。そ?🎼一个人。不在。🎼就上。梦想问。🎼要去的地。你问我。🎼あま?たし定。🎼一生。下。🎼放弃。年と。🎼假着说你。分手。🎼就不。不许。🎼悲上。远走。🎼安装。走过吗?🎼的城市。我站在。🎼那你就。有。🎼不起。你说的。🎼我们一。きめ?🎼多少事。台上绚烂的灯光,忘了明天要去。🎼前是彩虹那一段,分手桀要带我去寻找我的梦想。😊在什么。🎼有些。不再大街。🎼上。想问。🎼是否。谁。🎼悲伤。的确定。🎼一し?想。🎼笑着说。分手。🎼就不。不许。🎼すべ?悲伤。🎼有幸悟。是个。🎼感想起。老城。🎼的知。天年都。🎼你笑着说。再见了,孩子。🎼是你。不的。🎼世界。で.🎼出てる。到2012年呢,许飞发行了自己的同名专辑,其中有一首歌是我其实之前听过真的非常多遍,但是直到这次,我才突然意识到原来寻水的鱼是许飞唱的。这首歌和早期他的青春民谣已经明显有着不太一样的气质了,他更像是一个成年人开始重新思考自己和世界的关系。鱼需要水,这似乎是一个。🎼再简单不过的常识。但是如果把它放到人的生活中,就会变成一个很有意思的问题,我们到底有没有真正的找到过适合自己的环境呢?我们常常把适应环境当成一种美德,认为一个人足够努力优秀,就应该能够适应任何地方,但是事实不一定是这样的。一个不太擅长竞争的人,其实不一定非要在激烈的环境当中去证明自己。而一个喜欢独处的人,也不必为了融入群体,强迫自己,变得非常外向。我觉得许飞的音乐中有一个很明显的特点。就是他并不热衷于去表达标准化的成功学。他很少把人生写成一个向上的过程。相反,他更愿意观察自己的状态。我现在在哪里?我想去哪里?我是不是正在做着自己真正喜欢的事情,这些问题看上去很简单,对不对?但是往往。他需要我们花很多年才能够回答。🎼所以寻水的鱼,它让我想到的不是怎样变得游的更快,而是我们要先找到属于自己的那片水。这首歌的歌名听上去有些忧郁。但是编曲上呢又用了轻摇滚的节奏。加入了明显的吉他和节奏感。听感上比那年夏天,淡淡的歌更加利落。🎼把那种可能在歌名当中比较忧郁的感觉唱的。非常的。🎼扩大。的爱。🎼没有答。你的香。🎼你。这种感觉到来。🎼你的?不可。🎼す。い?🎼没有余儿。的风。🎼你。没有星星的夜。🎼确定.你的湖面冰静如昔。🎼有些有。😊一个沉水的。🎼我飘。这寂寞长眠。🎼我忘记了自己,紧紧拥抱你给我的那天。我是一个悬水的。😊🎼我轻轻跳着。我自由的忧豫,寻找着自己。渴望。😊🎼一条顺水的。我成。🎼我忘记了自己,紧紧拥抱你,给了人。一条是水的鱼。🎼我轻轻飘进。我听。🎼寻找着自己。渴望。那片天地。🎼え?如果说寻水的鱼,它讲的是寻找自我,那么接下来这时我们终究会牵手旅行,就是许飞在音乐当中另一种我很欣赏的表达。关于履行关于关系。关于人和人之间的陪伴。我是非常喜欢许飞音乐里的旅行感的。就像是吉林到北京,他就是一首非常好听的公路之歌。而这种旅行感呢,他并没有去炫耀某个目的地。是一种非常柔软的生活想象。就两个人不必立刻拥有这个答案。但是只要愿意一起往前走,路就会慢慢的亮起来。这首歌是一位70后的男性朋友写的歌词。记录的是他和一位90后女生的恋爱经历。几年后,男生为了陪女生出国读书,辞掉了年薪百万的工作。许飞被这份行动打动,决定把这首歌收进专辑。还特意用了男生的视角来唱。这首歌表面上是在写旅行。其实在写一种非常具体的承诺。就是我愿意为了你。🎼去重新安排自己的人生。音乐中有非常多古早的旅行符号,比如凤凰稻城。这些地名我们如今听上去已经有很强的年代感了。但他代表的是两个人在那个时期共同想象未来的能力。哪怕现在还没有出发,但我们的心中已经有了一张地图。而有关这首歌,我其实还看到过有关许飞。谈到的一段很值得回味的感情经历。他回忆早年在恋爱的时候,自己最大的愿望就是有一天可以跟对方一起旅行。而后来他真的攒下了一些旅行的钱。银行卡里终于有了足够出发的存款。却发现那个。原本想要一起旅行的人已经不在了。我觉得这个故事里面的主角,无论是谁,我都非常被他打动。因为他好像也让这首歌有了一个不那么浪漫,甚至有一些遗憾的注角。我也重新理解了歌曲当中为什么会有终究两个字。他很像是一种期待,当然也可能是一种遗憾。旅行可以是一个人的事,也可以是两个人共同经历的过程。而真正长久的关系,或许并不一定是两个人始终都走在同一条路上。而是为了彼此愿意理解。为什么对方想要去远方?你会突然间意识到。旅行可能最珍贵的东西。并不是目的地。🎼而是那个曾经想要和你一起出发的人。 🎼あて?Yeah.🎼于是否。开始放肆。🎼不知。无数眼光。🎼总被。こ?🎼都是。伤口来回。🎼穿行城市。一直。🎼写的。存多温存。🎼在寂寞。と?🎼我们终究。不会牵手。🎼你行.疯光。🎼到车。我是。🎼巴黎。おめ?🎼终究会牵手。旅行。🎼日光倾斜。爱上。🎼这旋律。天空。🎼起太过。甜蜜。🎼只是我。说。🎼你听.不必害怕一。🎼有艰辛。只需爱。🎼ち。呼吸。🎼这良辰美景,任世间静止看云淡风轻。😊们终究会牵手旅行。🎼疯狂到这或是。我们终究会牵手旅行。🎼日光倾斜带上这旋律。空气太过甜蜜,只需我说你听,不必害怕,一路艰行,只是安静呼吸。这良辰美景。🎼人世间。感青楚。🎼The。的。🎼风轻。看云淡。🎼风轻。😔 🎼.接下来这首歌我想多花一点时间来介绍。因为少年去游荡这五个字的分量,我相信。在最近也好,或者是在之前也好。读过许非在16年写下那篇长文的人。都绝对。不会无动于衷。你会真的折服于这个少年。他明明经历过。那些无数的。哪怕是每一件压到一个人身上,都可以称之为苦难的经历。但他到现在仍然保持着极为旺盛的生命力。少年去游荡,它是一首歌,也是这张我们很值得重新去认识许飞的专辑。他发行在201...(已截断)
All-In with Chamath, Jason, Sacks & Friedberg
(1)
Midterm Blood Bath, $4T AI Sell Off, Data Center Freeze, What's Next for Trump
1h 33m
2026/11/08
📝 AI
总结
All right everybody, welcomee back to the number one podcast in the world See all in podcast and I'd like to start by saying that all in recognizes that we are on the sacred land in Austin, Texas of the Cato, the Comanche, the Kickapoo and the Tokawata and we appreciate all their sacrifices and bring us here and we will be evaluating, I just talk to our HR department will be evaluating a fifth bestie Brad to white to male.🎼So we're looking for alternative DEi fifth Bessies to include in the round table. Wow, so some big changes coming to the Di is back as DEI is back S, did you have anything for the docket this week, I have an announcement, You have an announcement. Okay, I'm transitioning. from a douch back to a NAs your winner ride Ra man, David S.😊We open source.🎼And in the discovery.妈咪ご晒。🎼Quだ。い?あ。S, what do you want to you want to start with?I think there's some really important news around SaS.Out there, I mean, maybe salesforce launch a couple of new features. Defite top a docket type.Material, I'd like to start with science Cor.W nobody in our fan base I ever said. I'm ready to go. You guys want to talk about exponential time hypothesis. That's what the audience wants. They what.😊The SAS update from SS and the science update, and then we can get our market review and we'll wrap the show up no.In all seriousness, there was an election.This week, we had an amazing live stream over a million views so far.😊Now, crazy Shaath, I don't know if you knew this, but 400000 people tuned in live making us.I believe the number one election night.Life' should not true. It was 500000. And just to put it in perspective, the far right.Streamers got about 87000.The far left streamers got 250000 We got 500000, which was bigger than most cable as well, so.We were the biggest and bigger than, I think most, if not all cable as well.Apparently, America wants hot takes from tech bros. I. Here we go. Who would have thought would have it. And by the way, that was with only three of our all stars on the court at Schs had a board meeting.Thursday night couldn't make it Tuesday night. He was busy. Yeah, I was busy that night. I was.I think I was passed out in the bar.出什。😊やっか。Treatment.S is the only guy who goes to the bar with his own bottle of tequila.He's like, can you put this behind the bar and serve it to me had $50 a glasses?Special thanks to Ted Cruz Ra Emanuel, Shane Copeland of Polymarket man. Did he have a great night, I think.😊Poly market nailed it 99% of their prediction markets turned out to have it perfectly.Be hilarious,5th Besie Tim Dylan.Taylor Lorenz based Taylor Lorenz was fantastic and insightful.Brack Urst are just a solid job.Brought some charts.And.😊Ben Shapiro from the right, so it was a really great, great panel.And today, we're going to take a pretty systematic approach to what happened this week. We're going to start with a post mortum on the midterms. And then second.We're going to talk about the second half of Trump's.Terminal the next two years. So I'm sure data centers will come up in our recap, postmtem, et cetera, but.We're going to do a deep dive into the.AI trade, potentially cracking, so.Let's start with.Are.Post mortem on.What just happened in the midterms, obviously the Democrats sweeped.Both chambermbs of Congress.And they won at least 29 governor races.DEms will wind up with 54, 55 or 56 seats.Depending on how Alaska and Mississippi.Wind up. They have a lot of.Votes still coming in in the House Dems will have at least 250 seats, maybe even 260 the bigger sort is where the Republicans lost deeply red states that haven't gone blue in generations went blue including the great state of Texas please don't blame me I've onlyve been here for three years and you might be surprised that what my voting ticket look like Kansas, Iowa, South Carolina for the first time in decades Trump won all of those states.duble digits just two years ago compared to 2024 the GOP lost ground in every voter group even white men so let's go around the horn sacks people want to hear from you most of all I guess so what went wrong here what are the lessons the Republican Party can learn and and what did the Democrats do right in the last 18 months leading up to the midterms this was a slaing this is a blowout it was I believe the biggest midterm loss since.The Watergate midterm loss. So this was obviously a big defeat for Republicans. There's no sugarcoing it. What's the reason for that, I believe the vice president JD Vance explained it very well. it was this war and the only thing I would add to that is it was this war and the economic consequences that follow from the war. So what am I talking about price of gas It was around three bucks before the war went over four and higher price of diesel even higher. and then that factors into.the costs of food, all the goods that Fb transported, also fertilizer costs went up and so you saw the inflation rate at the beginning of the year back in January before the war it was down...(已截断)
The Joe Rogan Experience
(1)
#2578 - Calley Means
2h 24m
2026/11/07
📝 AI
总结
やろ。🎼The Joe Rogan experience train my day. Joe Rogan podcast by night, all day.Hey, good to see you. Good to see you. What are you doing, Well, I just recently resigned for my job. How was that.😊You know, two years ago when Bobby Kennedy joined President Trump it was one of the greatest, I think political miracles in history, I think we were also excited, it showed that there was a real political animation about a Democrat dealing working with the Republican I think it showed the country that the Trump administration was going to fight not just partisan stuff but special lenderers and the swamp I helped set that up, came in, dropped everything I was doing, came in was a senior advisor at the White House led.Implementation fought really hard. And I was optimistic, because.I'm a realist, and.Things get done in politics.Two ways. One is, there's.Special interest money.That's how a lot of things get done.That's why we spend three times more than any other country in healthcare.That's why.We have such high housing prices.think a lot of the things we're complaining about dealing with it because the systems have been rigged on the inside, we all know that.But the other way things get done is real public pressure and real public shame on the political system. And what Bobby Kennedy and President Trump did is showed thats a lot of voters They care about American health.And I was optimistic we could use that coalition to bust through the.Dc swamp to get some things done. And we did get some things done. But all of them were through public shame.You know, the psychedelic thing.That was being totally blocked.Full directive.From the top that we were going to slow that down and not do it. And your call to President Trump.You know, President Trump.Whenever there was an issue which was on his radar, he's like, let's get it done, let's go hard.He said we need to get this duet that's on my fucking desk.In the next week.And it busted through the system.You know, I was able to engender a coalition to get that.Same dynamic happening with things like dietary guidelines, The food dies.Getting those out of the system. These were all just this system of Dc does not want any change to happen and.The fact that we had.A voter coalition and could use that and say this coalition, you, other people are going to be pissed off if we don't get things done. We were able to get things done, but.The simple reality and why I left.嗯。Is that there is a very concentrated group of people in the Trump administration.That are antagonistic to what people voted for.And are, frankly.诶。I think it's very clear.Doing things for other reasons.Particularly in many cases, to enrich their families.Ohof.So.Walk us through this.So, you get in there.You, you have your.Your first day on the job.When did you realize it was going to be an issue, And you were very hopeful at first, right, It's very exciting.A lot of people were very excited about the coalition.They felt like it was one of the main things that I was excited about with Trump getting into office was that.Bobby Kennedy was going to be in there, and there was a there was a lot of these things that you just discussed.ALo lot of these things about peptides and stem cells.For sure, the psychedelics.And then just making people aware of like additives and food that were completely unnecessary.They were ruining people's health.So it's all this promise. everybody was very excited about it.It was electric two years ago, I mean.We remember it. It was absolutely electric. And that compared to what we felt this week, right, that's the real issue. Like, there was a real sense.Every single person in the first day of the Trump administration thought we were working for a presidency that would be.Absolutely historic. We thought we were that Trump was going to be better than George Washington. We the energy.Was.We felt like we were in the most consequential moment in American history. We were going to go hard, and we were going to be able to channel.What voters voted for, I cannot express you how electric it felt in there.In early days.So.What was the process like where you felt like it was off the rails,.So.I want to get into specifics. We all know things are corrupt and bad, but I'll, I'll just get right into specifics, What I learned.And I don't think anyone would dispute this the way the White House, this White House is set up.Is the most centralized control.Up to the Chief of Staff office.Of any administration in American history. So I'll just get into details about who runs the White House. And there's a picture I have of the key people.But.The White House, Susie Welles is most powerful chief of staff in modern American history, and she has taken a strangle hold on any policy decision. No policy decision.Gets done without her office.Giving it the okay. And it's absolutely systematic. So every single cabinet.Secretary.There's spending meetings constantly worrying about how to get on Susie's good side and how to win ...(已截断)