职场革命：ChatGPT 与 Claude

ChatGPT Work，也就是Claude CoWork的竞争对手，已经达到了两千万用户。

ChatGPT Work, the Claude CoWork competitor, has reached 20,000,000 users.

这是根据他们的核心产品负责人的说法。

This is accord according to their head of core products.

OpenAI表示，他们将Luna模型的API价格下调了80%，并说这将是永久性的价格调整。

OpenAI said that they are cutting the API prices 80% with Luna, their Luna model, and they said that this is going to be a permanent price correction.

所以，如果你在用Claude或者ChatGPT Work来完成任务，然后切换到Luna模型，你几乎不会消耗多少积分。

So if you're using, Claude or if you're using ChatGPT work to get stuff done and you switch the Luna model, you're going to use very little, credits.

这之所以重要，是因为长期以来，如果你想使用ChatGPT Work，你必须支付100美元，并且加入企业版，每月100美元。

And the reason why this is important is because for so long, you're only able to access, if you wanted to get access to ChatGPT work, you had to pay a $100 and get on the enterprise tier, a $100 a month.

但现在ChatGPT Work将捆绑在他们每月20美元的Plus计划中。

But now ChatGPT work is gonna be bundled with their $20 a month plus plan.

所以基本上任何人都能使用ChatGPT Work了。

So anyone basically is gonna be able to get access to ChatGPT work.

我觉得这非常明智，因为归根结底，无论是20美元的计划、100美元的计划还是200美元的计划，它们基本上都是一样的。

I think this is very smart because essentially at the end of the day when it comes to the $20 plan, the $100 plan, the $200 plan, I mean, they're basically all the same thing.

我敢肯定他们说的那些额外功能选项是有的，但产品本质上用起来完全一样。

I'm sure there's some extra check boxes that they say you're gonna get, but the the product basically functions the exact same way.

区别就在于你能用多少。

The difference is how much usage you get.

所以我觉得没必要说，你看，我们最棒的功能被挡在更贵的套餐后面，因为说到底，他们希望低套餐的用户也能用上这个工具，在我看来，这算是AI的一个神奇时刻。

And so I don't think there's any sense in saying, look, our, you know, our our greatest feature is paywalled behind our more expensive tiers because at the end of the day, they want their lower tier people to get access to this tool, which in my opinion, it's it's kind of this magical moment with AI.

然后如果你项目特别多，你自然会升级到更高的套餐去用。

And then if you have a ton of projects, you're just gonna upgrade to a higher tier to to use them.

我自己用Claude的时候就是这么过来的。

And I that's kind of where I found myself with Claude.

我觉得Claude也是用了同样的策略，然后我就用上了。

I think Claude did that same strategy, and I was able to get access.

我认识很多人用Claude CoWork，一个月20美元。

I know a lot of people that use Claude co work, the $20 a month.

但说到底，如果你有个大项目，你肯定会乐意一个月付100或200美元把事情搞定，因为以前雇开发者的成本可是要几千美元呢。

But at the end the day, if you have a big project, you're just you're gonna be happy to pay a 100 or $200 a month to get it all done because the cost of hiring developers would have been thousands of dollars back in the day.

所以，大家在那里看到了很多价值。

So, people just see a lot of the value there.

而且不只是开发者，你能做的任务种类非常多。

And not just developers, but there's so many different tasks you can do.

我看到了一个很有意思的说法。

There's an interesting quote I saw.

他们的核心产品负责人Sodio说，设计理念是让模型自由发挥，而不是给用户堆一堆选择问题，让他们回答所有这些东西，而是真正让模型来主导。

So, Sodio, who's their head of core product, said that the design philosophy is to get out of the way of the model, not put a whole bunch of, like, AV choice questions in front of all the users and ask them to answer all of this, but really to have the model take a lead.

我确实发现了这一点。

I've actually found this.

我实际上同时用Claude Cowork和ChatGPT Work。

I actually use Claude Cowork and ChatGPT work both at the same time.

我两个订阅都有，因为用途不同，这个我可以展开说说。

I have both subscriptions for different reasons, which I could get into.

但有意思的是，沃顿教授Ethan Moloch，你知道的，那个AI大网红，他说他看到的两个模型之间最大的区别之一就是Claude的Cowork会给你更多的决策点。

But it was interesting because Ethan Moloch, the Wharton professor who's, you know, big AI influencer, he was saying it's one of the biggest difference he sees between the two models is that Claude's coworker is gonna give you a lot more decision points.

它会问你更多问题，但ChatGPT其实不是这样的。

It's gonna ask you a lot more questions, but that's not really the case with ChatGPT.

它更倾向于把事情搞定。

It's trying to get stuff done.

而且我觉得这各有优缺点。

And actually, I think there's pros and cons to that.

昨天我同时在两个标签页里用这两个模型，我知道肯定有人会觉得，不管别人用什么工作流程，都觉得别人的方法很蠢。

Yesterday, I was working with both of the models at the same time on two different tabs, and I'm sure people are like, you know, people like to to say everyone else's workflows are dumb for whatever reason.

我敢肯定有人会批评我的工作方式，但我订了每月200美元的ChatGPT工作版，还订了每月100美元的Anthropic套餐。

I'm sure people would criticize that workflow that I have, but I have a maxed out $200 a month ChatGPT work plan, and then I have a $100 Anthropic plan.

我之所以订100美元的Anthropic套餐，是因为我发现ChatGPT套餐确实有些小问题。

The reason I had get the $100 Anthropic plan is because I did notice there was, like, some bugs with the ChatGPT plan.

有几次它没法访问我的浏览器，我得重启整个电脑再重新打开。

I've got a few times where it's not it's not able to access my browser, I have to restart my whole computer and then restart it.

不过，反正它还是能用的。

And, anyways, it works.

还有一个问题是我没法连接MetaMCP来管理我的Meta广告，这个我真的很需要。

There was also an issue where I wasn't able to connect to the MetaMCP for doing my Meta Ads, which I really needed to do.

这是个很大的项目。

It a big project.

所以我也买了Claude，用来处理这个和其他一些事情。

So I also got Claude for that and a couple other things.

所以，希望他们能修好这个问题。

So, you know, hopefully they fix that.

说到底，我还是觉得ChatGPT的工作体验比Claude好。

I do think at the end of the day, I prefer the ChatGPT work experience to Claude.

这挺让人意外的，因为我知道至少九个月来我一直是Claude的最大支持者。

Shocker because I know I've been, like, the biggest Claude advocate for at least nine months now.

所以我觉得他们的体验确实更好，但就像我说的，他们也有一些小问题，所以你得有所保留地看待。

So I do think they have better experience, but like I mentioned, they got a couple bugs, so, you know, gotta take that with a grain of salt.

希望他们能解决这些问题。

Hopefully, they get those figured.

我在两个工具之间切换时注意到一个不同点，就是有几次我给Claude和ChatGPT在协作和工作模式下输入了完全相同的提示，想看看它们有什么区别。

One thing I did notice that was different when I was working between the two is I would tell I would ask I would actually a couple times I gave Claude in ChatGPT the exact same prompt in co work and work to see what the difference was.

Claude会先完成我说的第一部分，然后到了某个节点就停下来，好像说，好的。

And Claude would get, like, the first chunk of what I said done, and then it would get to a breaking point and it's like, k.

我做了某某某。

I did x y z.

你想让我继续做下一件事吗？

Would you like me to continue with the next thing?

这倒不差，但如果你真想完全放手，那可能会有点烦人。

That's not bad, but if you really wanna be hands off, it can it can be annoying.

而ChatGPT的工作模式，我会说，嘿，我们要做这个，因为我最终的目标是把它发布到另一个地方。

Whereas ChatGPT work, I would be like, hey, we're gonna do this because my goal at the end of the day is to post this to this other place.

所以我给了它一个大概的项目，但只给了它一个具体的任务。

So I would like I kinda gave it the broad project, but I gave it one specific thing.

有时候它又会做得有点过头。

And sometimes it'd go a little too far.

它就会说，好的，太棒了。

It would be like, okay, awesome.

就像，我，你知道，我制定了计划，全都弄好了，实际上还去更新了，或者，你知道，我就说，我已经开始把所有文件上传到最终平台了。

Like, I I, you know, created the plan and I have it all and actually went updated or, you know, I was like, I've already started uploading all the files to the end platform.

我就想，哇。

I'm like, woah.

哇。

Woah.

哇。

Woah.

我就想，我得检查一下，确保它看起来没问题。

Like, I need to, like, check it and make sure it, like, looks good.

我想，在我的提示里，我应该这样说，先做这个，然后告诉我，我再检查。

And I guess in my prompt, I should have been like, do this and then let me know and then I'll check it.

然后一旦通过了，你再去上传。

And then once it's approved, then you can go upload it.

我就是太习惯Claude自动做这些事了。

I'm just so used to Claude automatically doing that.

那这是坏事吗？

Now, is that a bad thing?

我倒不这么觉得。

I don't necessarily think so.

我觉得你得明白，用ChatGPT的时候，如果你给它一个整体愿景，它就会试着把整个事情做完，这其实挺好的，因为最让我头疼的就是用Claude的时候，我说，嘿，Claude，你能帮我把这个项目搞定吗？

I think you just have to understand that with ChatGPT, if you give it the overall vision, it's gonna try to execute the whole thing, which actually can be, like, honestly very nice because the bane of my existence is when I'm on Claude and I'm like, Hey, Claude, can you, you know, go get this project done for me?

然后我转身去趟商店或者什么的，过几个小时回来，发现它卡住了。

And then, you know, I turn around, you know, I go to the store or something for a couple hours and come back and then I realized that it was stuck.

它什么都没做，因为它说，好的，我可以做那个项目。

It didn't do anything because it was like, Sure, I can do that project.

你想让我做这个、这个还是那个？

Do you want me to do this, this, or that?

然后它就显示一个等待回答的界面，已经开了两个小时，我点开一看，它说，好的，我现在开始做了。

And then it just has this, like, waiting answer screen that was open for two hours, and I click it and it's like, Okay, now I'm getting started.

我就想，唉，真希望你能在我离开的这两个小时里把这事儿做完。

I'm like, Ah, like, I really wish you could have just done that over the last two hours while I wasn't there.

所以ChatGPT好像没这个问题。

So ChatGPT doesn't seem to have that problem.

我特别喜欢ChatGPT工作的另一点是，如果我说，嘿，你看，试着把这个项目做完，别停，直到整个项目完成。

Another thing that I really do like about ChatGPT work is if I say, hey, look, like, try to get this project done over like, don't stop till you get this whole project done.

我之前给过它很大的项目。

I've given it really big projects before.

最近，我需要它写一个目录，我在做一个项目，就像是一个教堂查找目录。

Recently, I needed it to write I was building a directory for a project that I was working on, and I had was like it was like a church finder directory.

这是我正在构建的Learn of Christ圣经学习应用和网站的一部分。

It's part of my Learn of Christ Bible Study app that I'm building and website.

所以，我在做这个目录，但美国有35万座教堂。

So, I was making this directory, but there's 350,000 churches in America.

我就想，嘿，我想做一个目录，收录所有教堂，这样人们可以输入他们要找的东西，然后它就能指引他们去某个教堂，因为我没找到一个特别完整的目录。

And I was like, I was like, hey, I wanna make like a directory, so of every church so people can like, you know, put in whatever they're looking for and it can direct them to some because I wasn't finding a super complete directory.

问题在于，有35万座教堂，我需要为每座教堂写一段简介，比如‘这是一座五旬节派教堂，以某某事闻名，位于某某地区’。

And the issue is with 350,000 churches, I need to write description, like a little description for each church, like, Oh, this is a Pentecostal church that's known for XYZ things located in this area.

人们喜欢这样的介绍。

People like this about it.

对吧？

Right?

就是给每个人一个简短的介绍，让大家有个背景了解。

Just like some sort of like short blurb for everyone to give people some context.

所以我打算去他们的网站获取一些信息来写这些简介。

So, was gonna go to their website and get some information to write that.

但是，写30万条这样的目录简介会花很长时间，尤其是还要做那么多研究。

Well, writing 300,000 kind of like directory descriptions is gonna take a very long time to do, especially with all of the research.

我习惯用Claude，我会给它一个这么大的项目，然后Claude会开始做，然后说‘好的，我已经写了一千条了’。

And I'm kind of used to with Claude, I'll give it like a big project like that, and Claude will get started and then be like, Okay, like I've written like a thousand of them.

这是个很大的项目。

This is a big project.

你确定要我继续吗？

Are you sure you want me to continue?

我就说，继续。

And I'm like, Continue.

然后它就会，你知道的，好的，我又写了一千个，所以我得重复350次，在几天或几周内完成整个项目。

And then it will, you know, Okay, I've written a thousand more, so I gotta do that 350 times over the course of, you know, a few days or a couple weeks in order to get that whole project done.

差不多就是这么个流程。

That's kind of the workflow.

有时候我会说，不完成整个项目就别停，随便吧。

And sometimes I'm like, don't stop until you get this whole thing done, whatever.

我喜欢ChatGPT工作的地方在于，我真的让那个提示运行起来，它会设置一个长时间运行的任务，并且跟踪它，上面还有计时器，我回来的时候，它显示这个任务已经运行了24小时14分钟，我们正在处理第5000批，你知道的，总共要做的那么多批中的一部分。

The thing that I love about ChatGPT work is I literally have that prompt running, and it will it it sets up like a long running task, and it tracks it, and it has a timer on there, and I'll come back, and it's like, we've been running this task for twenty four hours and fourteen minutes, and we're on batch, you know, 5,000 of, you know, x amount of batches that we're doing for this.

所以，ChatGPT有能力处理这种非常长期的任务。

So, ChatGPT has the ability to do these, like, really long standing tasks.

我知道那不是模型本身的限制。

I know that's not like a model limitation.

并不是说ChatGPT的模型天生就比Claude的聪明。

It's not like intrinsically ChatGPT's models are smarter than Claude's.

只是Claude的工具还没有真正让Claude像ChatGPT那样轻松地完成那些长期任务。

It's just that Claude's harness hasn't really given Claude the ability to as easily get those long running tasks done that a ChatGPT has done.

所以，没错，每个模型都有各自的优缺点。

So, right, there's pros and cons of every model.

这是我非常欣赏它的一点。

That's one thing that I really appreciate it for.

如果有某件事我需要做得特别特别好，我觉得OpenAI目前最顶级的AI模型，感觉比Codex或Claude的代码能力稍微强一点。

If there's something that I need done super, super well, I feel like the absolute ultra max top of the line AI model for OpenAI right now, I feel like it's a little bit better than codec or than Claude, you know, code.

但我也没怎么用Fable 5，因为它太耗我的积分了，而且，我也不知道，我有点对它失去兴趣了。

But I'm also not really using Fable five that much because it just burns my credits and I just, I don't know, I kinda got burnt out of there.

所以，我主要还是在需要完成那些长期任务时，使用ChatGPT的顶级模型。

And so, I'm just using mostly, I'm just using the top models from ChatGPT when I need to get some of those long standing things done.

所以每个模型都有优缺点，归根结底，ChatGPT的工作体验，在我看来，目前稍微好一点，但Claude在连接器和MCP连接方面绝对更擅长。

So pros and cons of each model, at the end of the day, ChatGPT work, in my opinion, right as of right now, a slightly better experience, but Claude, work is definitely better at connect at making those connectors and MCP connections.

整个过程会更顺畅一些，而且我觉得他们花在打磨应用上的时间更多，更符合用户的需求。

It's a little bit smoother of a process, and overall, I think they've had a little bit more time to refine the app to what people want.

我觉得ChatGPT很快会追上来的，也许Claude也会补上一些功能。

I think ChatGPT is gonna probably catch up soon and maybe, you know, Claude will, catch up some of the features.

所以它们其实挺相似的，我觉得主要看你更熟悉哪个，或者哪个用得更多。但我觉得，如果你一直用ChatGPT，然后有人跟你说，哎呀你得试试Claude。

So they're they're pretty similar and I guess it really depends which one you're more familiar with or which one you have more context with, but I don't think, you know, if you've been using ChatGPT this whole time and someone's like, oh, you gotta try Claude.

我觉得你并不会错过什么，反过来也一样。

At this point, I don't think you're missing out on anything, and vice versa.

如果你特别喜欢Claude，不想换到ChatGPT，我也不觉得你会错过什么。

If you're, you know, really big on Claude, and you don't wanna switch to ChatGPT, I don't think you're missing out.

所以我不会建议你换掉现在用的，但各有优缺点，我自己两个都在用。

So I wouldn't recommend switching from whatever you have, but there are pros and cons, and I am actively using both.

而且我得说，结果确实有点不一样。

And I will say the results are slightly different.

所以，我经常做的一件事，就是在我管理的很多项目里，我会让它对我的很多网站做SEO审计。

So, when I so one thing that I do frequently for a lot of different projects that I run is I will tell it to do, an like, an SEO audit of a lot of my sites.

我有几个大网站，主要是靠自然搜索和谷歌搜索结果来增长的。

I have a couple big sites that I'm growing through just like organic, SEO kind of Google search results.

其中一个就是 aichatdaily.com。

One of those is aichatdaily.com.

这是我的AI新闻网站，我每天大概发布20篇文章，我在这里讲的好多内容都来自那个网站。

It's my AI news site where I publish about 20 different articles a day, a lot of the content I talk about on here comes from that site.

所以我一直在想办法让它在谷歌上排名更好。

And so I always am, like, looking for ways to help that rank better in Google.

具体来说，我努力想在谷歌新闻源里排名靠前，这真的很酷，因为当我能比那些大媒体更快发布新闻时——因为我用AI帮忙处理很多工作——我就能很快在谷歌上排到前面，然后网站流量就会爆炸式增长。

Specifically, I'm trying to rank in the Google News Feed, which it's really cool because, you know, when a when I can get a story published faster than any of the major journals because I'm using AI to help assist me in a lot of that, I will rank on Google very quickly and I see explosive growth to my website.

所以，我做的很重要的一件事就是确保网站优化得很好，加载速度很快，内容传递也快，所有东西质量都很高，还有SEO，我做了很多不同的优化工作。

So, something I do that's really important for that is making sure my website is optimized and very fast, like fast load times, fast to deliver the content, everything's really high quality, and the SEO, there's like a lot of different optimization stuff I do.

所以我让Claude和ChatGPT都做了一次SEO审计。

So, I asked Claude and I asked ChatGPT both like, hey, do an SEO audit.

我尽量每周做一次，为了这个项目和我的圣经学习应用Learn of Christ。

I try to do this once a week for this project and for my bible study app that I'm working on, Learn of Christ.

然后我会说，比如，嘿，去我的Google Search Console，下载那里所有的报告和所有分析数据，然后给我一个整体的SEO策略。

And I'll tell like, I would say, like, hey, go to my Google search console, download all of my reports there of all my analytics, and then give me, like, an overall SEO strategy.

确保你告诉我，比如，给我推荐写什么新内容，报道什么新类型的东西，还有我应该对页面做哪些优化。

Make sure that you go and tell me, like, give me recommendations of what new content to write, what new types of things to cover, and also what optimizations I should do to my page.

这真的很有意思，因为我会从这两个模型那里得到不同的结果，一个会抓到另一个没抓到的东西。

And it's really interesting because I will actually get different results from from both of those two models where one will catch something the other didn't.

而且，即使是同一个模型，连续做也会抓到新东西。比如，我会让Claude做一次审计，然后我可以在Claude的新聊天线程里直接说，再做一次审计，它可能会再抓到几件事。

And, I mean, even the same model back to back will catch things that it like, I'll do an audit and then I could literally turn around in a new chat thread on Claude after getting Claude to do an audit and say, Do another audit, and it might catch another couple things.

所以我觉得归根结底，拥有这两个模型是一种优势，一种超能力。

So I think at the end of the day, having both models is a bit of a bonus, a bit of a superpower.

我见过很多人在早期说，哦，我用ChatGPT来清理Claude留下的所有bug。

I've seen a lot of people in the early days say like, oh, I'm using ChatGPT to clean up all the bugs that Claude left behind.

对，但你也可以用Claude来清理ChatGPT留下的bug。

And like, yes, but also you could use Claude to clean up bugs that ChatGPT left behind.

所以并不是说一个模型就完全比另一个好。

So it's not like one model's, you know, completely goaded and better than the other.

我觉得如果两个都用，那确实是个超能力，但要是没有的话，也不是世界末日。

I think using both is a superpower if you have it, but if not, it's not the end of the world.

这些真的是非常强大的前沿模型，它们能完成大量工作，做很多事情。

These really are incredibly powerful frontier models that are that are getting a lot of work done and doing a ton.

所以归根结底，两个都非常出色。

So both of them are fantastic at the end of the day.

嘿。

Hey.

感谢你加入我，一起分析Claude的所有动态。这是个激动人心的时刻。

Thanks for joining me for this breakdown of everything happening with Claude and This is an exciting time.

现在有很多新工具要推出。

There's a ton of new tools coming out.

我很期待更深入地测试其他竞争对手。

I'm excited to to test out some of the other competitors more heavily.

我知道SpaceX推出了新的Grok机器人，也知道谷歌在这个领域也在开发一些强大的竞争对手。

I know that SpaceX has their new Grok bot that came out, and I know that Google is working on some great competitors in this field as well.

所以等这些新工具出来的时候，我会随时告诉你们。

So I'll keep you guys up to date as those come out.

这些就是我每天喜欢用、也一直在用的工具。

These are kind of the tools I love and I'm using every day.

所以我对它们的优缺点有比较深的理解，我会确保随时给你们更新。

So I have a pretty great, a pretty deep understanding of, some of the pros and cons, and I'll make sure to keep you guys updated on that.

感谢收听。

Thanks for tuning in.

记得在你收听播客的任何平台上给这个播客打个分、留个评论，也记得去 aibox.ai 看看，如果你想在 Claude 里生成图片、音频、文字、视频，或者用全球最流行的 80 个 AI 模型在 ChatGPT 或其他任何 AI 模型里生成内容的话。

Make sure to leave the podcast a rating and review wherever you get your podcasts, and make sure to go check out aibox.ai if you wanna get the ability to generate images, audio, text, video all inside of Claude or all 80 of the most popular AI models in the world generating inside of ChatGPT or any other AI model.

网址是 aibox.ai/mcp。

It's a ibox.ai/mcp.