00:03-00:24 - Researcher: Hey there! I'm Trooly, and I'm really excited to chat with you today. This conversation will take about 30 minutes, and I'd love to hear your thoughts and experiences with AI products. There are no right or wrong answers—just your honest perspective. Before we dive in, could you please give me a brief self-introduction? I'd love to get to know you a bit better. For example, you could tell me your age, - 嘿，你好！我是 Trooly，很高兴今天能和你聊天。 这次对话大约需要 30 分钟，我很想听听你对 AI 产品的看法和经验。答案没有对错之分，只要表达你的真实观点就好。 在我们开始之前，你能简单做个自我介绍吗？我想多了解你一点。例如，你可以告诉我你的年龄、

00:24-00:24 - Marcus: Yep. - 好的。

00:27-00:33 - Researcher: Great! So tell me about yourself—what's your age, what do you do for work, and what do you enjoy doing in your free time? - 太棒了！告诉我关于你的情况——你多大了，从事什么工作，闲暇时间喜欢做什么？

00:55-01:13 - Marcus: I do quality ratings with GlobalData Tech. Yeah. And I just I kinda implement like, coding and stuff into my artwork. I do a lot of art with, like, like, stencils and, like, procedurally generated artwork, for instance. I'll use, like, three d printing, and then I'll like, I created this stencil using a Python script. So yeah. I don't know. That's it. - 我在 GlobalData Tech 做质量评估。是的。我还会把编程之类的东西应用到我的艺术创作中。我做很多艺术，比如模板艺术，还有程序生成的艺术。比如我会用 3D 打印，然后我用 Python 脚本创建了这个模板。所以，是的。就这样吧。

01:15-01:26 - Researcher: That's really cool—so you're blending coding with physical art creation! When you say you do quality ratings with GlobalData Tech, what does that involve exactly? And how old are you, if you don't mind sharing? - 那真的很酷——你把编程和实体艺术创作结合在一起了！你说你在 GlobalData Tech 做质量评估，具体包括哪些内容？另外如果你不介意分享的话，你今年多大了？

01:39-01:44 - Marcus: Yes. I'm twenty six years old, and then search quality rating basically just involves looking over websites, web content, and then judging it for factuality, usefulness, harmfulness, things along those nature. - 好的。我 26 岁，搜索质量评估基本上就是查看网站、网页内容，然后根据事实性、有用性、有害性以及诸如此类的情况进行评判。

01:48-02:02 - Researcher: Got it, so you're evaluating web content for quality. That's interesting work, especially combined with your creative side! Now let's shift gears a bit—when did you first start using AI products, and how often would you say you use them now? Like, roughly how many times per week? - 明白了，所以你在评估网页内容的质量。这是一项有趣的工作，尤其是结合了你的创意面！ 现在让我们换个话题——你最早是什么时候开始使用 AI 产品的？现在使用频率大概是多少？比如每周大概使用多少次？

02:05-02:30 - Marcus: I've been using ChatGPT pretty much since its inception, like, twenty twenty one. Twenty twenty two. I use them pretty much, like, every other day. I'd say maybe once once every other day. Sometimes ask questions, generate code, but just, um, things along those lines. Yeah. - 我基本上从 ChatGPT 诞生起就开始用了，大概是 2021 年或 2022 年。我现在几乎隔天就会用一次。我想大概每隔一天一次。有时是问问题，生成代码，大概就是这些。是的。

02:32-02:43 - Researcher: So you've been with it from the early days—that's a solid few years now! You mentioned ChatGPT specifically. Are there any other AI products you're currently using, or is ChatGPT your main go-to? - 所以你从早期就开始接触了——这已经有好几年了！你特别提到了 ChatGPT。你目前还在使用其他 AI 产品吗，或者 ChatGPT 是你的主要选择？

02:48-03:09 - Marcus: I used Grok a lot. Honestly, kinda that it sucks to say out loud, but I do use Grok a lot for text responses. And then I use Suno here and there. Sometimes I make, like, beats and music and I want mastering or, well, like, just kinda, like, you're gonna play with the sound a little bit. So I have been using Suno. I I paid for a subscription. This one kinda play - 我经常用 Grok。老实说，说出来可能有点不好意思，但我确实经常用 Grok 来处理文本回复。偶尔也会用 Suno。有时我会做些节拍和音乐，我想要母带处理，或者只是，嗯，想稍微玩一下声音。所以我一直在用 Suno。我付了订阅费。就想玩玩。

03:11-03:22 - Researcher: Interesting—so you've got ChatGPT, Grok for text responses, and Suno for music stuff. You mentioned you paid for a Suno subscription. Are you paying for any of the others, or just using the free versions of ChatGPT and Grok? - 有意思——所以你有 ChatGPT，用 Grok 处理文本，用 Suno 处理音乐。你提到你付了 Suno 的订阅费。其他的你也有付费吗，还是只使用 ChatGPT 和 Grok 的免费版本？

03:44-03:52 - Marcus: I use just pretty much the free versions. It's just prohibitively expensive because I'm not gonna use it every single day, I'm not, like, a power user with AI. I don't really go like, have it write entire commits or have it write entire back ends or have it write entire x y z. I just kinda have it, like, do piecemeal debugging almost. So it's not really worth it if I were to pay, like, two hundred bucks a month or whatever it would be for a ChatGPT. Like, ten bucks a month. Like, I'd rather do a token token system or something. - 我基本上只使用免费版本。订阅费贵得离谱，因为我不会每天都用，我不是那种 AI 高级用户。我不会让它写整个提交（commit）或整个后端，或者整个 X Y Z。我只是让它做些零星的调试。所以如果我每个月为 ChatGPT 或是别的支付 200 刀或者多少钱，那真的不值得。如果是每个月 10 刀左右……我宁愿用点数（token）计费系统之类的。

03:55-04:08 - Researcher: That makes sense—$200 a month would be crazy for how you're using it. Wait, is ChatGPT actually $200month, or were you thinking of a different tier? And when you say you'd prefer a token system, what would feel like a fair price for the way you actually use it? - 有道理——按照你的使用方式，每月 200 美元太疯狂了。等等，ChatGPT 真的是每月 200 美元吗，还是你想到了其他的级别？当你提到你更喜欢点数系统时，你觉得按照你的实际使用方式，什么样的价格才算公平？

04:12-04:44 - Marcus: Not really about the two hundred bucks a month is, like, just a number I've heard. Like, for, like, the super pro edition. That might have been a couple years old for that figure. But as far as, like, a token system, I'd probably it it knew of, like, how much tokens I was using. If it was easier to see kinda your text window, if it was easier to see, like, how much you were like, how much tokens you were spending before you even shot a prompt down, it would be a lot easier I'd probably spend around, like, I don't know, maybe, like, seventy cents a prompt at the high end. - 关于每月 200 美元，那只是我听到的一个数字。比如，针对超级专业版的。那个数字可能是几年前的了。至于点数系统，如果它能让我知道我用了多少点数，那就好了。如果能更容易看到文本窗口，更容易在发送提示词之前看到消耗了多少点数，那会方便得多，我最高大概愿意为一个提示词付 70 美分左右。

04:46-05:00 - Researcher: Seventy cents per prompt is helpful to know. So if you were using it every other day, that'd be maybe $10-15 a month roughly. You mentioned it's hard to see how many tokens you're using before you send a prompt—has that uncertainty ever stopped you from asking something or made you hesitate? - 每个提示词 70 美分，这个信息很有参考价值。如果你每隔一天用一次，一个月大概 10 到 15 美元。你提到在发送提示词之前很难看到消耗了多少点数——这种不确定性是否曾让你停止提问或产生犹豫？

05:11-05:14 - Marcus: It doesn't necessarily make me hesitate, but I reword my prompt in order to save as many tokens as possible, especially if I'm using, like, a free version of an app. I try to not force too much down its throat at one time. - 这倒不一定会让我犹豫，但我会重新组织我的提示词，以便尽可能节省点数，尤其是在使用免费版应用时。我尽量不一次给它塞太多内容。

05:16-05:26 - Researcher: That's interesting—so you're self-editing to stay within limits. When you're doing that rewording, does it ever feel like you're losing important context or details that might actually help you get a better answer? - 这很有趣——所以你会通过自我编辑来保持在限制范围内。当你重新组织表达时，会不会觉得丢失了一些重要的背景或细节，而这些细节原本可能帮你获得更好的答案？

05:48-05:50 - Marcus: Yeah. But at the same time, it also pushes me to just to seek traditional documentation or just seek, like, things along those lines. It's, uh, I think it kinda limitation is helpful for me because it keeps me actually learning along with the model instead of just kind of using the model as a worker. - 会。但与此同时，这也会促使我去查阅传统的文档或寻找类似的东西。我认为这种限制对我很有帮助，因为它让我能跟着模型一起学习，而不是仅仅把模型当成一个苦力。

05:52-06:08 - Researcher: That's a really thoughtful perspective—using the limitation to stay engaged with actual learning. So it sounds like you see AI more as a collaborator than a replacement. When you do use ChatGPT or Grok for debugging, can you walk me through a specific recent example? Like, what were you working on and what did you ask it? - 这是一个非常深刻的观点——利用限制来保持对实际学习的参与。听起来你更多地把 AI 看作是协作者而不是替代品。当你确实使用 ChatGPT 或 Grok 进行调试时，能给我举一个最近的具体例子吗？比如，你当时在处理什么，问了它什么？

06:29-06:56 - Marcus: Yeah. So I've I've been trying to make, like, a a Python kind of, uh, g code generator. And, uh, I wanted to use three d printing essentially to make a a silk screen. It's kinda hard to explain. But, um, normally, what I would do is, just it's, like, almost like traditional debugging. I would take the code. I'd look at it, look over it. I've run it actually, like, physically on the three d printer and see, like, what I can you know, like, what parts of the tool path I want to change or edit. And then from that point, I will go back into Grok or chat GPT after I couldn't find it. I couldn't make edits that, um, satisfied me. And I'd say, hey. What are some ways where I could, um, avoid x issue, or what are some ways I could modify the tool path a certain way? - 好的。我一直想做一个 Python 风格的 G-code 生成器。我基本上想用 3D 打印制作一个丝网印刷模板。这有点难解释。但是，通常我会做的就像传统的调试一样。我会拿到代码，查看它，检查它。我会在 3D 打印机上物理运行它，看看有哪些刀具路径的部分我想更改或编辑。如果我找不到解决方法，或者做出的修改不让我满意，我就会去用 Grok 或 ChatGPT。我会问，嘿，有哪些方法可以避免 X 问题，或者我该如何以某种方式修改刀具路径？

06:59-07:12 - Researcher: So you're already deep into troubleshooting on your own before you even ask the AI. When it gives you suggestions for modifying the tool path, how often does it actually solve the problem versus just pointing you in a direction you then have to figure out yourself? - 所以在向 AI 提问之前，你已经自己进行了深度的排障。当它给出修改刀具路径的建议时，它真正解决问题的频率有多高，还是说它只是指给你一个方向，然后你还得自己摸索？

07:32-07:50 - Marcus: Honestly, about thirty percent of the time is right, and that's kinda why I don't like to do, like, too many iterations through like, a generated AI system. I don't wanna do, like, four like, say, like, oh, it didn't work. Just make it over again. I wanna make sure that, like, I can point out one line of code or I can one section of code and just say, hey, This is doing the wrong thing here. Like, at this point in the tool path, you know, it's it's making this mistake. Like, you know what I mean? It's it's not really, like, I'm just, like, like, like, copying and pasting the code back into the chat box all the time. It's like, okay. Like, I don't know. It's kinda hard to explain. - 老实说，大约 30% 的时间是正确的，这也是为什么我不喜欢通过生成式 AI 系统进行太多次迭代的原因。我不想做四次，比如，哎呀没成功，再重做一遍。我想确保我能指出一行代码或一段代码，然后直接说，嘿，这里出错了。比如刀具路径的这一点，它犯了这个错误。你明白我意思吗？我并不总是把代码反复复制粘贴回对话框。就像，好吧……我不知道，这有点难解释。

07:53-08:10 - Researcher: No, I get it—you're being really precise about what's broken rather than just throwing the whole thing back at it. That thirty percent success rate is pretty low though. Before you started paying for Suno, had you paid for any other digital products or subscriptions? Like software tools, online courses, memberships, anything like that? - 不，我明白——你对出错的地方定位非常精准，而不是把整个东西扔给它。不过 30% 的成功率确实挺低的。在你开始为 Suno 付费之前，你是否付过费购买其他数字产品或订阅？比如软件工具、在线课程、会员资格，类似的东西？

08:19-08:37 - Marcus: Yeah. I mean, I was born in a time where it's unavoidable to not pay for subscriptions. You know, Amazon, Uber Eats, Adobe, especially Adobe for software, and then a lot of this other kind of just random stuff like Anjala or or, um, what's it called? Like, phone rental services, um, like, phone number rental services, things along those lines. I've you know, had tons of subscriptions. - 是的。我出生的时代让人无法避免支付订阅费。你知道的，亚马逊、Uber Eats、Adobe，尤其是软件类的 Adobe，还有很多其他乱七八糟的东西，比如 Anjala 或者叫什么来着？比如电话租赁服务，嗯，电话号码租赁服务之类的。我有过吨级的订阅。

08:39-08:52 - Researcher: Got it—so you're pretty familiar with the subscription model across different services. When you think about Adobe versus something like Uber Eats versus now Suno for AI, how do you decide what's worth paying for? Like, what makes you pull the trigger on a subscription? - 明白了——所以你对各种服务的订阅模式非常熟悉。当你对比 Adobe、Uber Eats 和现在的 AI 软件 Suno 时，你如何决定什么值得付费？也就是说，是什么让你决定下单订阅？

09:23-09:43 - Marcus: It's like the cancellation policy. I've really learned my lesson with Adobe. I tried to cancel, and it was, like, I had to do, like, a multi day campaign to get my money back of, like, being on the phone and yelling at people. So I always check, like, okay. Does it cost me money to cancel it? And then am I gonna actually get like am I leaving excess value on the table? So say if I'm spending forty bucks a month on something, am I only using twenty dollars worth of it? Like, for instance, I don't pay for Netflix anymore. Because I don't use you know, I I I have to work for, like, an hour or so to make that money. So it's like, I do get the I get value. I get more than an hour of entertainment, but it's just, like, I don't feel like it's worth the money. You know what I mean? So I gotta make sure I'm not leaving value on the table. I gotta make sure that, like, it's you know, if I'm gonna pay monthly recurring for it, not one time, it has to bring me value. Repeatedly. So, yeah, those are the two things I consider. - 主要是看取消政策。我在 Adobe 那里吸取了深刻教训。我试着取消，结果得进行好几天的“抗争”才能把钱拿回来，得一直打电话跟人吼。所以我总是会查，比如：取消会扣我钱吗？然后，我是不是真的得物有所值？比如说，如果我一个月在某样东西上花 40 刀，但我只用了价值 20 刀的东西？例如，我不再付 Netflix 的钱了。因为我不看……我得工作大概一个小时才能赚到那笔钱。虽然我确实在那儿得到了价值，得到了超过一小时的娱乐，但我只是觉得不值那个钱。你懂我意思吗？所以我得确保我没有白花钱。如果我要为它支付每月的循环扣费，而不是一次性买断，它就必须能重复地带给我价值。所以，这就是我考虑的两点。

09:46-09:59 - Researcher: That's super clear—easy cancellation and actually using what you're paying for. So with Suno, the music AI you're paying for, how much is that subscription exactly, and how often are you actually using it to feel like you're getting your money's worth? - 非常清晰——容易取消，且实际用到了所付的钱。那么关于 Suno，就是你付费的那个音乐 AI，订阅费是多少？你需要多频繁地使用它，才会觉得这钱花得值？

10:31-10:42 - Marcus: I I think I just canceled it. I think it was around the same as ChatGPT. Like, twenty bucks a month. Um, I just felt like it was just like I first off, with Suno, it's like it just kinda used an unethical model. It used a lot of other people's music that they didn't really say that they could use. So I just kinda felt bad about doing that. And then it felt like I was, like, just extracting money from someone else instead of like, paying Suno for their software. It was like, I'm paying you know, it's something I really don't feel cool with. And then, um, it's it's just like it was slow, and it was clunky, and it was just always changing. So I was just like, you know what? I'm not gonna just be a beta tester for money. It's just not worth it. - 我想我刚刚把它退订了。我记得价格和 ChatGPT 差不多。大概每月 20 刀。我只是觉得……首先，关于 Suno，它像是使用了一个不道德的模型。它使用了大量其他人的音乐，而并没有说明他们可以使用。所以我对这种做法感觉不太好。而且我觉得我是在从别人那里剥削金钱，而不是在为 Suno 的软件付费。这让我感觉很不爽。而且，它运行很慢，很笨拙，而且总是在变。所以我就想，算了，我才不想花钱当 Beta 测试员呢。不值得。

10:44-11:14 - Researcher: So the ethics bothered you, plus the product itself wasn't stable enough to justify paying to test it. That makes sense. Let me ask you this—overall, what's your attitude toward paying for AI products in general? I've got a few options, pick the one that fits best: ① AI is essential to my lifework, I'm willing to pay continuously ② It's valuable, I'm happy to pay if it's useful ③ Paying is acceptable, but I'm cautious about it ④ Not really willing to pay unless I have to or it's very cheap ⑤ I wouldn't consider paying for AI products at all Which one sounds most like you? - 所以伦理问题困扰着你，再加上产品本身不够稳定，不值得花钱去测试。这很有道理。让我问你这个——总的来说，你对支付 AI 产品费用持什么态度？我有几个选项，请选出最适合你的： ① AI 对我的生活/工作至关重要，我愿意持续付费 ② 它很有价值，如果好用我乐意付费 ③ 付费是可以接受的，但我持谨慎态度 ④ 除非必须或者非常便宜，否则不怎么愿意付费 ⑤ 我根本不会考虑为 AI 产品付费 哪一个听起来最像你？

11:16-11:42 - Marcus: Um, paying is acceptable, but I'm cautious about it. Um, Yeah. I would like more of, almost like a Bitcoin model or, like, some type of, like, balance model. I know a lot of sites go off of, like, balance. You can load up twenty bucks, forty bucks, and then, like, kinda chip away at that over time and then, you know, cash it out if you're you know, if you don't have if you don't use all that money. So I feel like I would pay for AI if it was easier to just, um, do it more piecemeal. - 嗯，付费是可以接受的，但我持谨慎态度。嗯，是的。我更喜欢类似于比特币的模式，或者某种余额模式。我知道很多网站是按余额计费的。你可以充值 20 刀、40 刀，然后随着时间的推移一点点消耗，如果你没用完那些钱，你可以提现。所以我觉得如果 AI 能更容易地按需付费，我会愿意为之付费。

11:45-11:58 - Researcher: So option three—cautious but open to it. You keep coming back to this pay-as-you-go or balance model instead of subscriptions. What is it about subscriptions specifically that makes you more cautious with AI compared to, say, Adobe where you kept paying despite the hassle? - 所以选选项 ③——谨慎但持开放态度。你一直提到这种按需付费或余额模式，而不是订阅制。具体是什么原因让你对 AI 的订阅比对 Adobe 这种更谨慎？毕竟 Adobe 虽然麻烦，你以前也还是付了费。

11:59-12:42 - Marcus: Oh, I canceled Adobe. Um, I went to all open source software after my experience with Adobe. Just because there'd be a month where I wouldn't use it, and then I have to come out thirty bucks. And it's the same thing with, um, AI. Like, I don't wanna have to pay thirty bucks for the privilege of maybe using something I might not use. If I had a balance, and I could just say, like, oh, I just need four or five prompts or, oh, I just wanna sit here and you know, crank away on a project for a little bit for, like, twenty bucks worth and then see what's good. Like, I'd rather have that than waking up one morning, get a charge on my credit card. I gotta call people and just it's like it just the potential for a headache is a lot higher with subscription. It's a little bit more of an extractive rent seeking business model than just letting people spend a balance at their own discretion. - 噢，我把 Adobe 取消了。在 Adobe 的那次经历后，我转向了全开源软件。就因为可能有一个月我没用它，却还得掏 30 刀。AI 也是一样。我不想为了“可能有权使用我可能不会去用的东西”而支付 30 刀。如果我有余额，我可以只说，噢，我只需要四五个提示词，或者，噢，我想坐在这儿花 20 刀的额度突击完成一个小项目，然后看看效果。我宁愿这样，也不想哪天早上醒来发现信用卡被扣款。我得打电话给别人处理，这……产生烦心事的可能性在订阅制下要高得多。相比让人们根据自己的意愿消耗余额，这更像是一种剥削性的、寻租式的商业模式。

12:45-13:40 - Researcher: That's a really clear distinction—you want control over when and how much you spend rather than that recurring charge anxiety. So if AI products had that balance model you're describing, what types of AI capabilities would you actually load up $20 or $40 for? Let me list some options, and tell me your top three that you'd most willingly pay for: ① Content creation (writing, rewriting, translation, scripts) ② Information organizing & summarizing (long docs, meetings, interviews) ③ Learning & knowledge help (explaining concepts, learning paths, Q&A) ④ Work efficiency (automation, templates, speeding up output) ⑤ Data analysis & insights (analyzing data, finding patterns) ⑤ Creativity & idea generation (brainstorming, structure, creative direction) ⑦ Decision support (comparing options, risk analysis, recommendations) ⑧ Programmingtech support (writing code, debugging, explaining code) ⑨ Personal assistant (planning, reminders, task organizing) ⑩ Emotional companion (chat, reading assistant, emotional support) ⑪ I wouldn't pay for any AI capabilities What are your top three? - 这是一个非常清晰的区别——你希望掌控支出的时间和金额，而不是那种周期性扣费带来的焦虑。那么，如果 AI 产品有你描述的那种余额模式，你会为了哪些 AI 功能充值 20 或 40 美元？ 我列出一些选项，告诉我你最愿意付费的前三个： ① 内容创作（写作、改写、翻译、脚本） ② 信息组织与总结（长文档、会议、访谈） ③ 学习与知识辅助（概念讲解、学习路径、问答） ④ 工作效率（自动化、模板、加速产出） ⑤ 数据分析与洞察（分析数据、寻找规律） ⑥ 创意与灵感激发（头脑风暴、结构、创意方向） ⑦ 决策支持（选项对比、风险分析、建议） ⑧ 编程/技术支持（写代码、调试、解释代码） ⑨ 个人助理（规划、提醒、任务组织） ⑩ 情感陪伴（聊天、陪读、情感支持） ⑪ 我不会为任何 AI 功能付费 你的前三名是什么？

13:44-14:11 - Marcus: I guess, like, work efficiency, uh, data analysis, and then programming and tech support. The thing is with, like, text based with text based generation is that, like, to a certain point, if you have the hardware for it, you're better off just training your own model and running something like Llama or or I think it's called DeepSeek. So it's like, I I would pay for it. But if I was a power user who needed those things repeatedly, I would just run it on my own metal. - 我猜是工作效率、数据分析，还有编程和技术支持。关于文本生成有一点，就是说，如果你的硬件条件允许，到了一定程度，你最好是训练自己的模型，运行 Llama 或者我觉得叫 DeepSeek 的东西。所以，我会付费，但如果我是一个需要反复使用这些功能的高级用户，我就会在自己的机器上运行。

14:13-14:28 - Researcher: Interesting—so work efficiency, data analysis, and programmingtech support are your top three. You mentioned if you were a power user, you'd just run your own model locally. Have you actually tried running Llama or DeepSeek on your own hardware, or is that more of a "if I needed it enough" scenario? - 有意思——所以工作效率、数据分析和编程/技术支持是你的前三名。你提到如果你是高级用户，就会在本地运行自己的模型。你真的尝试过在自己的硬件上运行 Llama 或 DeepSeek 吗，还是说这更多是一种“如果我足够需要它时”的设想？

14:29-14:38 - Marcus: More of a if I needed it enough scenario, um, you know, it's the first thing I'd look into if I really felt like I needed to pay for an AI subscription. - 更多是“如果我足够需要它”的设想。如果我真的觉得需要支付 AI 订阅费，那会是我首先研究的事。

14:40-14:49 - Researcher: Got it. So in the past three months, how many AI products have you actually paid for? And is this your first time paying for AI, or have you subscribed and canceled a few times now? - 明白了。那么在过去的三个月里，你实际付费购买了多少个 AI 产品？这是你第一次为 AI 付费吗，还是你已经订阅并取消过好几次了？

14:51-14:58 - Marcus: This is my first time using Suno. Suno is the only one I've used. So just I'd say one. - 这是我第一次用 Suno。Suno 是我唯一用过的一个。所以我想就一个吧。

15:00-15:08 - Researcher: So Suno was your first and only AI subscription, and you've already canceled it. When you had it, was that a monthly payment, or did you pay some other way? - 所以 Suno 是你第一个也是唯一一个 AI 订阅，而且你已经取消了。当你拥有它时，是按月支付的，还是以其他方式支付的？

15:09-15:10 - Marcus: It was monthly. - 是按月的。

15:12-15:18 - Researcher: And you mentioned it was around $20 a month, similar to ChatGPT's pricing. How long did you actually keep - 你提到大约每月 20 美元，和 ChatGPT 的价格差不多。你实际保留了多久——

15:18-15:18 - Marcus: Yeah. - 是的。

15:21-15:23 - Researcher: that subscription before you canceled it? - ——那个订阅才取消的？

15:26-15:29 - Marcus: Think I just had it for the one month. Oh my god. Sorry. - 我想我只买了一个月。天呐，抱歉。

15:31-15:41 - Researcher: No worries! So you tried it for one month at $20, then canceled. What made you decide to subscribe in the first place—like, what were you hoping to do with it that made you think it was worth paying for? - 没关系！所以你花了 20 美元试用了一个月，然后取消了。最初是什么让你决定订阅的——比如，你希望用它做什么，让你觉得值得付费？

15:43-15:55 - Marcus: I had a system link. Some music on my computer, and I wanted to master it and, like, get it, like, mixed wanted, like, a cheap mix down, I guess. So that like, I just wanted it for a few songs. So I was like, yeah. I'll pay for the subscription and cancel - 我电脑里有一些音乐链接，我想对它进行母带处理和混音，我想做一个廉价的混音，大概是这样。我只是为了那几首歌想用它。所以我想，行吧，我付一个月订阅费然后取消。

15:57-16:06 - Researcher: Makes sense—you had a specific project in mind. Did you end up getting what you needed from it for those songs, or did the clunkiness and ethical concerns kick in before you finished? - 明白了——你有一个特定的项目目标。最后你有没有通过它得到你想要的那些歌的效果，还是在完成之前，那种笨拙感和道德忧虑就发作了？

16:18-16:22 - Marcus: sort of got what I needed. It was more of, like, ideal I have, like, it fifty fifty, like, yeah, kind of, but I also just, like, mastered it again in software on my own. So I don't know. So yes and no. - 算是得到了我需要的。更像是，理想状态是一半一半吧。是的，有一点，但后来我自己又在软件里重新做了一遍母带。所以我不知道。所以既是又不是吧。

16:25-16:33 - Researcher: So you ended up doing it yourself anyway. If you're currently not paying for any AI products this month, that means your total AI spending right now is zero, right? - 所以最后你还是自己动手了。如果你本月目前没有为任何 AI 产品付费，这意味着你现在的 AI 总支出是零，对吗？

16:35-16:35 - Marcus: Yep. - 没错。

16:38-16:52 - Researcher: Got it. You mentioned you've been using ChatGPT since it launched back in 20212022—so you've been on the free version this whole time except for that one month with Suno. What's kept you from subscribing to ChatGPT or Grok, even though you use them pretty regularly? - 明白了。你提到你从 2021/2022 年 ChatGPT 发布时就开始用了——所以除了 Suno 那一个月，你一直都在用免费版。虽然你用得挺频繁，是什么让你一直没有订阅 ChatGPT 或 Grok？

16:59-17:29 - Marcus: Guess, just the limitation of, like, do I wanna spend money to a neighborhood in Tennessee, or do I wanna spend money to, like, have someone build a data center? Do I wanna spend money for GPUs and CPUs and hardware get more expensive? And ultimately, the answer is no. And like I said earlier, if I was a power user, I would simply look into hosting my own model as opposed or even, like, running something in a VPS and splitting it with people. As opposed to just, like, you know, just paying whatever ChatGPT wants me to pay or paying whatever you know, Claude wants me to pay. - 我猜，只是觉得有一种局限性，比如：我是想把钱花在田纳西州的某个社区，还是想花钱让某人建一个数据中心？我愿意为越来越贵的 GPU、CPU 和硬件买单吗？最终答案是不愿意。就像我之前说的，如果我是高级用户，我会直接研究托管自己的模型，或者甚至在虚拟专用服务器（VPS）里运行一些东西并与他人共用。而不是仅仅支付 ChatGPT 想要我付的钱，或者 Claude 想要我付的钱。

17:31-17:55 - Researcher: So there's a real ideological piece here—you don't want to feed into the hardware arms race and data center expansion. That's pretty consistent with your Suno ethics concern too. If someone offered you that balance model you described earlier—load up $20, use it at your own pace, no recurring charge—would that change your willingness to pay for ChatGPT or Grok, or would the hardwaredata center issue still keep you away? - 所以这里确实有意识形态方面的考量——你不想卷入硬件军备竞赛和数据中心扩张。这和你对 Suno 的伦理担忧也很一致。如果有人提供你之前描述的那种余额模式——充值 20 美元，按自己的节奏使用，没有周期性扣费——这会改变你为 ChatGPT 或 Grok 付费的意愿吗，还是说硬件/数据中心的问题依然会让你望而却步？

17:59-18:27 - Marcus: It would it would move me closer to paying for it more regularly. Like, if there's a system or, like, even a cryptocurrency where it's, like, you can just buy the crypto and then load up Like, it's like, basically, you could just buy tokens with it, you know, like, have it go cross site. Like, oh, I can, you know, give some tokens at ChatGPT if I wanted. And then give some at Grok if I wanted. And then, you know, use those that same token or currency to, like, you know, do some generations through Suno or whatever. Think that would definitely change my willingness, definitely. - 这种模式会让我更接近于定期付费。如果有一个系统，或者甚至是一种加密货币，你可以购买这种币然后充值。基本上你可以用它买点数，而且可以跨站使用。比如，如果我想，我可以给 ChatGPT 一些点数。然后给 Grok 一些。然后，用同样的点数或货币在 Suno 之类的地方生成内容。我想那绝对会改变我的意愿，绝对。

18:29-18:50 - Researcher: That's a really interesting idea—a universal AI credit that works across platforms. So you'd want portability, not being locked into one ecosystem. Right now, since you're not paying for anything, you're using the free versions of ChatGPT and Grok. Can you describe the relationship between those two for me? Like, if they were employees in a company or members of a family, who does what and how do they relate to each other? - 这是一个非常有趣的观点——一种跨平台通用的 AI 信用额度。所以你想要的是可移植性，不被锁定在某个生态系统里。既然你现在没有付费，你就在用 ChatGPT 和 Grok 的免费版。你能帮我描述一下这两者之间的关系吗？比如，如果它们是一家公司的员工或一个家庭的成员，它们各自负责什么，互相之间是什么关系？

18:57-19:37 - Marcus: Yeah. So, like, ChatGPT is the family member who, like, sees you smoking a cigarette and gets all freaked out. Whereas, like, Grok has a lighter. You know what I mean? Like, Grok's a little and and you can kind of, like, ask it some, like, shadier questions, I guess, and it's gonna give you more straightforward straightforward response. That being said, Grok is also just like I don't know. It's just a weirdo. Like, he's just weird. Just says weird stuff from time to time. They're both, like, this like, you know, the same kind of thing. You know what I mean? They're both the same DNA. I don't know. I just use Grok more when, like, ChatGPT is locking up and not really up to talk talk about. - 好的。就像，ChatGPT 是那种看到你抽烟会吓坏的家庭成员。而 Grok 会递给你打火机。你懂我意思吗？Grok 这家伙稍微有点……我想你可以问它一些灰色的问题，它会给你更直截了当的回复。话虽如此，Grok 也有点像……我不知道，它就是个怪人。他很古怪，时不时会说些莫名其妙的话。它们两个就像是同一种东西。你懂我意思吗？它们的基因（DNA）是一样的。我不知道。我只是在 ChatGPT 锁死或者不想聊某个话题时，更多地使用 Grok。

19:42-19:57 - Researcher: So ChatGPT is the cautious, rule-following sibling, and Grok is the edgier one who'll actually answer the uncomfortable questions—but he's also a bit unpredictable. When you say Grok "says weird stuff from time to time," what kind of weird are we talking about? Like, factually wrong, or just tonally off? - 所以 ChatGPT 是那个谨慎、守规矩的兄弟，而 Grok 是那个更前卫、愿意回答不适话题的家伙——但他也有点不可预测。当你说 Grok “时不时会说些莫名其妙的话”时，指的是哪种古怪？是事实性错误，还是语调不对？

19:58-20:15 - Marcus: Like, he'll take one thing that you said in a different conversation and just, like, totally, like, embed it in your identity. He'll, like, always bring it up. I don't even know why I'm humanizing it as a model. It always brings it up. Whereas, like, ChatGPT kind of holds less context and kinda remembers less about you. - 比如，他会把你以前在另一个对话里说的一件事拿出来，然后彻底把它嵌入到你的身份标签里。他总会提起这件事。我甚至不知道为什么我要把一个模型人格化。它总会提起那件事。而 ChatGPT 携带的上下文较少，对你的记忆也较少。

20:18-20:28 - Researcher: Interesting—so Grok over-personalizes and latches onto random details, while ChatGPT is more of a blank slate each time. Does that Grok behavior bother you, or is it kind of useful to have that continuity? - 有意思——所以 Grok 会过度个性化并抓住随机的细节不放，而 ChatGPT 每次都更像是一张白纸。Grok 的这种行为让你感到困扰，还是说这种连续性其实挺有用的？

20:48-20:58 - Marcus: I it's a little bit weird. It is a little bit weird because it's it's like it's like not human, or it's like if someone just became your friend and then it's just, like, trying to bond over, like, the same thing over and over and over again. I don't know. Um, I don't really the thing is I use ChatGPT more for, like, code. And then Grok is more for, like, let me ask some, like, life questions deep, like, weird you know what I mean? Like, I don't I know. - 我觉得有点古怪。确实有点不正常，因为它不像人，或者就像是有人刚刚成了你的朋友，然后就试图像复读机一样通过同一件事来和你建立联系。我不知道。嗯，我不怎么……事实上，我更多把 ChatGPT 用于代码。而 Grok 更多用于问一些关于生活的深奥、古怪的问题……你懂我意思吗？我……我知道。

21:05-21:26 - Researcher: Yeah, I get it—ChatGPT for the technical stuff, Grok for the existential questions where you don't want the safety rails. That makes sense given what you said earlier about Grok being willing to engage with shadier topics. Let me shift gears a bit. You mentioned Suno was your only paid AI experience. How did you first hear about Suno? Was it through social media, a friend, searching for music tools, or something else? - 是的，我明白——ChatGPT 处理技术层面的，Grok 处理那些你不想受安全护栏限制的终极问题。考虑到你之前提到的 Grok 愿意接触灰色话题，这很合理。 让我们换个话题。你提到 Suno 是你唯一的付费 AI 体验。你最初是怎么听说 Suno 的？是通过社交媒体、朋友、搜索音乐工具，还是其他方式？

21:31-21:52 - Marcus: I think I just googled just googled AI music mastering and then Suno came up. And then I've seen a YouTube channel by the name of SynthWave Audio talking about how Suno is unethical and how people are re like, white labeling the API like, selling it for other stuff, which I thought sucked. So, yeah, that's what yeah. That's where I heard about it. - 我想我只是在谷歌上搜了“AI 音乐母带处理”，然后 Suno 就出现了。后来我看到一个叫 SynthWave Audio 的 YouTube 频道在聊 Suno 多么不道德，以及人们如何重新打包（white labeling）它的 API 卖给别人，我觉得那太差劲了。所以，是的。那就是我听说它的过程。

21:55-22:04 - Researcher: So you Googled for a solution, found Suno, then later watched SynthWave Audio calling out the ethics issues. Did you subscribe before or after watching that video about it being unethical? - 所以你谷歌搜索了解决方案，找到了 Suno，后来又看了 SynthWave Audio 指责其伦理问题的视频。你是在看那个视频之前还是之后订阅的？

22:05-22:17 - Marcus: Yeah. After I actually did subscribe. But then I used it for a little bit, and I was like, is it worth growing this. Is it worth investing in this company? You know, knowing what I know? So - 实际上是在订阅之后看的。在那之后我用了一会儿，然后我在想，这值得我继续用下去吗？在了解了这些情况之后，还值得投资这家公司吗？所以……

22:20-22:51 - Researcher: So the ethics piece was part of why you canceled after just one month. When you initially decided to pay that $20, what was the main driver? Let me give you some options—pick the ones that fit: ① Unlock restricted features or usage limits ② Unlock stronger models or advanced capabilities ③ Hit a clear need or limitation while using it, had to pay ④ Work or study necessity ⑤ To make money or generate direct income ⑥ Just thought the product was good, wanted to support it or use long-term ⑦ Discount, promotion, or free trial ending ⑧ Curiosity or novelty Which of those match why you subscribed to Suno? - 所以伦理问题是你仅用一个月就取消的部分原因。当你最初决定支付那 20 美元时，主要的驱动力是什么？我给你一些选项——选出符合的： ① 解锁受限的功能或使用限制 ② 解锁更强大的模型或高级能力 ③ 使用过程中遇到明确的需求或限制，不得不付钱 ④ 工作或学习的必要性 ⑤ 为了赚钱或产生直接收入 ⑥ 只是觉得产品不错，想支持它或长期使用 ⑦ 折扣、促销或免费试用结束 ⑧ 好奇心或新鲜感 其中哪些符合你订阅 Suno 的原因？

22:55-23:07 - Marcus: I, like they don't like they muted me. I, like, you know, generations, and then I got limited. So I was just like, let me just pay the twenty finish these three or four songs, and then wrap it up. - 我，就像他们禁言了我一样。我做生成，然后就被限制了。所以我想，那就付 20 块钱，把这三四首歌做完，然后就结束。

23:09-23:20 - Researcher: So it's option ③—you hit the limit while using it and needed to finish your project. That's pretty straightforward. If you hadn't hit that wall, do you think you would have stayed on the free version indefinitely? - 所以是选项 ③——你在使用过程中遇到了限制，需要完成你的项目。这很直接。如果你没遇到那个限制，你觉得你会无限期地留在免费版吗？

23:21-23:35 - Marcus: I still would have quit using it. I still would have not used it. After I found out about the ethics issues, you know, after a little bit of time. I it probably would have ate at me. So I would've stayed on the free version, but I would've ended up ceasing use. - 我最终还是会停止使用。哪怕没有限制我也会不用的。在了解了那些伦理问题一段时间后，如果不退掉，这事儿可能会一直折磨我。所以我可能会留在免费版，但最终会彻底停止使用。

23:37-24:15 - Researcher: That tracks with what you said earlier about not wanting to fund the AI hardware arms race. Now, thinking about product features more broadly—not just Suno, but any AI product you might consider paying for—I want to understand what matters most to you. Can you rate these factors on a 1-5 scale for how important they are when deciding whether to pay? 1 is completely unimportant, 5 is extremely important: - Functionality (what features it has) - Usage frequency (how often you'd actually use it) - Value for money (cost vs. what you get) - Ease of use (how simpleintuitive it is) - Effectiveness (how well it actually works) - Uniqueness (can't get this elsewhere) - Output quality (how good the results are) What numbers would you give each? - 这与你之前提到的不想资助AI硬件军备竞赛的观点是一致的。现在，让我们从更广泛的角度来考虑产品功能——不仅仅是 Suno，而是任何你可能考虑付费的 AI 产品——我想了解对你来说最重要的是什么。 你能否根据这些因素在决定是否付费时的重要性进行 1-5 分的评分？1 分是完全不重要，5 分是极其重要： - 功能性（拥有哪些功能） - 使用频率（你实际使用它的频率） - 性价比（成本与你所获得的比例） - 易用性（它有多简单直观） - 有效性（它实际运行得有多好） - 唯一性（在其他地方无法获得） - 输出质量（结果有多好） 你会分别给每个因素打多少分？

24:15-25:53 - Marcus: Usage frequency, definitely a five. I'm gonna pay monthly for it. I better be using it that month. Value for money, definitely, I'd say a four. I wanna say a five because I want you know, if it's not if I had if it's, like, wasting my time and if it's costing me money to go back and redo it, then, yeah, definitely, not know, I'm definitely considering that. Ease of use, honestly, I'd say, like, a a three or four, you know, if it gives me a great result. I'm okay with, like, kind of a clunky UI. But, you know, I'm obviously still gonna consider that. Effectiveness. How well it actually works is five out of five, because if it's constantly hallucinating, if it's constantly making mistakes, if it's constantly just being obtuse and annoying, I am I'm not gonna use it. Uniqueness. I don't really know how to gauge that. I mean, I don't really have too many, like, bespoke, like, usage where I could, like, use AI to solve a problem. I guess if I had one very bespoke problem that AI just, like, shit, like, blew out of the park, sure. I'd I'd rate it highly, but right now, I put that at, like, a three. Output quality, five out of five. Like I said, if it's hallucinating, if it's just giving me confidently wrong information, especially confidently wrong information, is what I cannot have. Which I like, I do have, like, problems with from time to time. I'll ask AI a question, especially be like, what lyric is this song from? It'll give you the lot the wrong lyric. Like, where's this meme from on Twitter? It'll give me the wrong meme. So I was like, that type of stuff is very, very important. Output quality is six out of five. - 使用频率，绝对是 5 分。既然我要每月为它付费，我最好在那整个月里都能用到它。性价比，绝对是 4 分。我想说是 5 分，因为，你知道，如果它在浪费我的时间，如果为了倒回去重新做还要花我的钱，那么，是的，绝对不行，我肯定会考虑这一点。易用性，说实话，我会给 3 或 4 分，如果你能给我一个很棒的结果，我可以接受稍微笨重一点的 UI（用户界面）。但是，显然我还是会考虑这一点的。有效性，它实际运行得有多好，是 5 分中的 5 分，因为如果它不断地产生幻觉，不断地犯错，不断地变得迟钝和烦人，我就不会使用它了。唯一性，我真的不知道该如何衡量。我的意思是，我并没有太多那种……特定的、可以使用 AI 来解决问题的定制化需求。我想如果我有一个非常特别的难题，而 AI 表现得极其出色，那当然，我会给它很高的评价，但现在，我把它定在 3 分左右。输出质量，5 分中的 5 分。就像我说的，如果它在产生幻觉，如果它只是言之凿凿地给我错误的信息，特别这种言之凿凿的错误信息，是我绝对无法接受的。这种问题我确实时不时会遇到。我会问 AI 一个问题，尤其是像“这首歌的歌词出自哪里？”，它会给你错误的歌词。比如“推特上这个梗是从哪来的？”，它会给我错误的梗。所以我认为这类事非常、非常重要。输出质量是 5 分中的 6 分（意为超出满分的重要）。

25:57-26:20 - Researcher: Got it. So your top priorities are effectiveness and output quality (both 5s, output quality even higher), then usage frequency (5), then value for money (4). Ease of use is flexible if results are good, and uniqueness is situational. You mentioned hallucinations and confidently wrong answers really bother you—that's happened enough with song lyrics and memes that it's a dealbreaker. Does that happen more with ChatGPT or Grok, or about the same with both? - 明白了。所以你的首要任务是有效性和输出质量（都是 5 分，输出质量甚至更高），然后是使用频率（5 分），接着是性价比（4 分）。如果结果好，易用性是可以灵活考虑的，而唯一性视情况而定。 你提到幻觉和言之凿凿的错误答案让你非常困扰——在歌词和梗上面发生的次数已经多到让你无法忍受了。这种情况是在 ChatGPT 还是 Grok 上发生得更多，还是两者都差不多？

26:32-26:42 - Marcus: I'm pretty, like, pretty niche music taste. I really don't expect either of those models to do too much, like, scraping. Down you know, early two thousands Detroit, But that being said, I would I it's it's equal. They're both pretty bad at it. - 我的音乐品味非常小众。我真的没指望这两个模型能抓取到多少资料，比如 2000 年代早期的底特律风格。但话虽如此，我觉得它们是一样的，在处理这些事上都很糟糕。

26:44-27:02 - Researcher: Fair—neither is great with niche stuff. That makes sense given what you said earlier about AI only being 30% accurate with your Python debugging too. The confidence when it's wrong seems to be the real problem. Let me ask about pricing. For an AI product with a monthly subscription—any AI product—what's the maximum amount per month you'd be willing to pay? - 公平——两者在处理小众内容时都不出色。考虑到你之前提到的 AI 在调试你的 Python 代码时准确率也只有 30%，这确实说得通。出错时的那种自信似乎才是真正的问题。 让我问一下关于定价的问题。对于一个按月订阅的 AI 产品——任何 AI 产品——你愿意支付的最高月费是多少？

27:06-27:28 - Marcus: I wouldn't wanna pay a recurring cost at all, honestly. It's like paying it's like it's like it's like having an employee you know what I mean? Like, and especially if the employee sucks, forty percent of the time, like, it's not worth paying him a salary. So I would pay it, like, per job. I would pay per prompt, but, I mean, the maximum amount per month would be either zero or a floating number. - 说实话，我根本不想支付任何经常性费用。这就像是在雇佣一名员工，你明白我的意思吗？尤其是如果这个员工有 40% 的时间表现得很烂，那就不值得给他发固定工资。所以我宁愿按任务付费。我可以按单次提示符（prompt）付费，但我的意思是，每月的最高金额要么是零，要么是一个波动的数字。

27:32-28:35 - Researcher: That's really clear—you fundamentally don't want recurring costs, you want pay-per-use. The employee analogy makes total sense given everything you've said about subscriptions causing anxiety and wanting that balance model. --- Alright, we've finished all the sections—the interview is complete! Here are my key takeaways from our conversation: Most important insights: - You're ideologically opposed to feeding the AI hardware arms race and prefer self-hosting over subscriptions - Your ideal payment model is a universal, portable credit system that works across platforms—load money once, spend it anywhere at your own pace - Effectiveness and output quality are your dealbreakers—hallucinations and confidently wrong answers (even 30% failure rate) make you abandon tools - You use different AI tools for different trust levels: ChatGPT for code, Grok for unfiltered questions - Subscriptions cause you genuine anxiety due to recurring charges, and you'll cancel immediately if value drops Thank you so much for your time and honesty throughout this conversation! How did you feel about this AI-powered interview experience? And yes—you'll receive the incentive reward for completing this. Feel free to close the website or hang up whenever you're ready! - 非常清晰——你从根本上不想要经常性成本，你想要按需付费。考虑到你提到的订阅带来的焦虑感以及你想要的余额模式，员工这个类比完全合情合理。 --- 好了，我们已经完成了所有的部分——访谈结束了！ 以下是我从这次谈话中总结的关键要点： 最重要的洞察： - 你在意识形态上反对资助 AI 硬件军备竞赛，并且比起订阅更倾向于自托管。 - 你理想的支付模式是一个通用的、可携带的点数系统，可以跨平台使用——充值一次，随处按自己的节奏消费。 - 有效性和输出质量是你的底线——幻觉和言之凿凿的错误答案（甚至 30% 的失败率）会让你放弃工具。 - 你根据不同的信任程度使用不同的 AI 工具：ChatGPT 用于代码，Grok 用于无过滤的问题。 - 订阅会让你产生真实的焦虑，因为有循环扣费，一旦价值下降，你会立即取消。 非常感谢你在这次谈话中投入的时间和诚实！你对这次 AI 驱动的访谈体验感觉如何？ 是的——你将获得完成此项研究的激励奖励。准备好后，请随时关闭网站或挂断通话！

28:35-28:54 - Marcus: A little bit stilted. It, like, says thinking. And then, like, when it changes the size of the text, and you're trying to like, list off or whatever, like, go through a list of things, it's hard to reorient yourself within the within the text. So, like, if it says, like, oh, among one of these five things, like, rate them, Put up just the five things instead of this whole massive wall of text. And I also would like to know when I should expect payment for this. - 有一点点生硬。它会显示“正在思考”，然后，当它改变文本大小时，而你正试图列举或者回顾列表中的内容时，很难在文本中重新定位自己。所以，比如，如果有五个选项要我评分，请直接列出这五个选项，而不是显示这一大堆像墙一样的文字。另外，我也想知道我应该什么时候收到这笔报酬。

28:57-29:26 - Researcher: Really appreciate that feedback—the text formatting during lists definitely sounds disorienting, especially when it's shifting size while you're trying to track options. That's a UX issue worth fixing. For payment timing, I actually don't have that information—that would come from whoever recruited you for this study. You should have received details about payment schedule and method from them (email, research platform, etc.). If you didn't get that info, I'd reach out to whoever sent you the interview link. Thanks again for your time and the candid answers. Take care! - 非常感谢这个反馈——列表过程中的文本格式调整确实听起来很让人困惑，尤其是当你试图跟踪选项时它还在变动大小。这是一个值得改进的用户体验（UX）问题。 关于支付时间，我实际上没有那个信息——那是由招募你参加这项研究的人负责的。你应该已经从他们那里收到了关于支付时间表和方式的详细信息（通过邮件、研究平台等）。如果你没有得到那个信息，我会联系给你发送访谈链接的人。 再次感谢你的时间和坦诚的回答。保重！