本集简介
双语字幕
仅展示文本字幕，不包含中文音频；想边听边看，请使用 Bayt 播客 App。
欢迎回到《医疗创新播客》。
Welcome back to the Medical Innovation Podcast.
我是普拉纳夫·桑格维。
My name is Pranav Sangvi.
今天，我们请回了一位几个月前在拉斯维加斯健康大会上采访过的嘉宾。
Today we're bringing back a guest that we spoke with a few months ago at the Health Conference in Las Vegas.
过去五年里，这位医生创业者把一个想法变成了影响全球数百万患者生活的公司。
Over the past five years, this physician entrepreneur has turned an idea into a company that's impacting the lives of millions of patients all over the world.
我们非常高兴再次邀请到海迪健康公司的首席执行官兼创始人汤姆·凯利医生。
Well, we're excited to be speaking once again with doctor Tom Kelly, the CEO and founder of Heidi Health.
汤姆，很高兴你再次做客。
Tom, it's so great to have you back.
非常感谢你，普拉纳夫。
Thanks so much, Pranav.
是的。
Yeah.
这六个月可真够忙的。
It's been busy six months.
我可都看在眼里。
So I've seen.
上次我们聊的时候，你刚完成B轮融资。
So last time we spoke, you had just closed your series b.
从那以后已经过去五个月了。
It's been about five months since then.
从去年十月到现在，Heidi最大的变化是什么？
What's the biggest thing that's changed, Heidi, since last October?
是啊。
Yeah.
我们现在成了一个多产品线的团队了。
I think we're a family multi product now.
我们已经推出了Evidence和Comms这两个产品，用户采纳率都特别高，令人兴奋。
So we we've launched Evidence and Comms and getting, like, really, really exciting adoption across both.
所以说，海蒂的产品一直以来对临床医生来说都挺深入的。
So, yeah, I think I mean, Heidi's always been, like, a pretty, like, deep product for clinicians.
你可以上传文件、来回聊天，还能总结多次会话，功能特别多。
You know, you could attach files and chat back and forth or summarize multiple sessions and do lots of different things.
现在好了，侧边栏里有了独立完整的产品，用户可以直接用，这感觉真不错。
But, yeah, it's nice to have, like, fully fledged separate products in in the the sidebar now that people can use.
另外，我们整体的影响力也增长了很多。
Otherwise, yeah, I think our our scale of impact has grown a lot.
我记不清上次跟你说的时候我们做到什么程度了，但现在每周支持的问诊量快到两百五十万次了。
I think I can't remember where we're up to when I spoke to you last, but we're now supporting almost two and a half million visits every week.
全球越来越多的机构开始采购临床AI工具，不再只是个别医生自己用。
Lots more organizations around the world are are now buying clinical AI tools rather than it being just individual clinicians.
医院的采用率在上升，电子病历系统也在陆续对接，各种进展都有。
So seeing a lot of hospital adoption and medical record integrations and all sorts of things.
总之，产品越来越多，客户越来越多，希望患者们也越用越满意。
So, yeah, just just like more, more, more, you know, more products, more customers, and and hopefully more happy patients.
这太惊人了。
That's incredible.
我们稍后会逐一深入探讨这些部分。
And we'll dive into each each of those parts in just a few minutes.
但我想知道，你提到你们从一家仅做记录的公司，发展成了如今的产品组合，这个转变是什么时候发生的？
But I'm curious, as you mentioned, you went from just being a scribe company to sort of a suite of products now.
你们一开始就有这个打算，还是后来出现某个关键节点，促使你们扩展了业务？
When did you first make a decision to move beyond the scribe?
是某个转折点推动你们走出记录这个范围的吗？
Was it always the intention, or was it a sort of a tipping point that pushed you to expand beyond the scribe?
是的。
Yeah.
我一直都有这个目标。
I think it was always the ambition.
我创办公司的时候，我们的使命就始终是提升整体医疗能力。
So back when I started the company, the mission we had was always about health care capacity overall.
所以我们该怎么把全球医疗资源翻倍呢？这可不只是记笔记那么简单。
So, you know, how do we double the world's health care capacity That's not just documentation.
你明白吧？
You know?
笔记写得快确实重要，但光靠这个，没法让病人更快离开急诊，没法加快出院速度，也没法更有效地解决实际问题。
Like, it's one important element, but if you write notes faster, that doesn't get your patient out of the emergency more quickly or or get them discharged faster or or get that issue actually managed more effectively.
这确实是其中一部分，但只是很小的一部分。
It's definitely a a piece of it, but it's it's only sort of a small part.
一开始，我们的产品线要多得多。
So, in the beginning, we had, like, a lot more product.
我们做过远程医疗、预约系统，甚至还有面向患者的App，各种各样的功能都有。
You know, we had telehealth and bookings and even, like, patient facing app and all sorts of stuff.
但最近几年，我们才真正找到方向，决定把重心全放在病历记录上，其他功能都砍掉了。
And we stripped it back a lot when we were we finally were finding that product market fit with scribing really in the last few years.
不过，我的愿景一直没变：未来会有一个AI助手，从患者出发，逐步协助医生，自动处理越来越多的事务，重复执行那些重复性高、风险低的医疗行政工作。
But, yeah, I'd say we our vision has always been that there will be this AI partner that maps from the patients to the clinician over time doing more and more autonomous things with our patients and then, you know, repeats and, like, some of the more administrative low risk parts of health care.
而对于医生来说，使用这套系统——当我不在的时候也能支持我的病人——这其实是我自己思考和工作的延伸，本质上是把那些值得接手的任务接过来。
And then for doctors using that same system that supports my patients when I'm not there, just some more of my own reasoning, my own work, and basically taking the jobs that are worth taking away.
所以，我想做的任何事，依然由我主导。
So anything I wanna do is still mine.
但如果我真的想把一部分工作交出去，我也可以做到。
But if I if I do wanna delegate some amount of work, I can.
所以，我觉得这些转变都是很自然的。
So, yeah, I think I think these are just natural transitions.
我们推出这些产品的原因是，想想医生每天要做的事，其实有一长串任务。
And the reason we launched these products is that if you think of what a clinician does, there's effectively, like, there's a long list of jobs.
但归根结底，大概可以归纳为四种核心能力。
But basically, there's, I'd say, like, four capabilities.
你需要能读懂病历，或者说是了解关于病人的各种数据来源。
You need to be able to read read the medical records or, like, read the source of data about your patient.
你也需要能写回信息。
You need to be able to write back.
所以这就像是病历记录、医嘱和检查这些事。
So that's like, you know, notes and orders and investigations.
通过助手功能，我们正好能同时搞定这两点。
So really with scribe, we achieve both those those two things.
你还需要能够参考循证依据。
You need an ability to reference evidence.
比如第三方的研究证据，或者你所在机构的临床指南，这些都会影响你对眼前患者的决策。
So there's third party evidence studies, maybe your your organization's guidelines that guide your decision making for the patient in front of you.
最后，你还需要能够有效沟通。
And then, ultimately, you need to be able to communicate.
无论是发短信、用语音跟患者交流，还是跟其他医护人员、医疗系统、拒付机构、热线电话沟通，方方面面都得能应对。
So you need to be able to text, talk to patients with your voice, talk to other providers, health systems, denials, hotlines, you know, all of the above.
有了这些功能之后，当然，每个单独拿出来都有用。
And then once you have those components, yeah, sure, they're good separate products and there's utility.
但要是把它们整合在一起，就能处理任何临床工作了。
But when you combine them, you basically can do any arbitrary clinical work.
你可以设置它们来避免手术取消，开展流感疫苗接种活动，或者进行患者自报指标的术前核查。
So you can set them up to do avoid surgical cancellations or to do flu vaccination campaigns or to do surgical check ins for patient reported measures.
这主要看具体情况。
Just it kinda depends.
现在限制它的不再是能力，而是想象力。
It it's limited by imagination, not by capability anymore.
所以我觉得，总的来说，未来的发展方向就是让海蒂为临床医生承担更多工作，这应该是大势所趋。
So, yeah, I think, generally generally, I think the the path is always gonna be, like, just Heidi doing more work for for clinicians is the the general trend trend, I think, where we'll we'll go on.
没错。
Right.
这样一来，医生就能专心当医生，多陪陪病人，而不是花时间在你刚才说的那些行政事务上，比如看病历、写病历、读各种文档之类的。
So physicians can be physicians and spend time with their patients instead of doing all these administrative tasks you mentioned, you know, reading notes or writing notes or reading documentation, etcetera.
这确实让
That makes a
好多了。
lot Yeah.
当然
Of
对
Yeah.
没错
Yeah.
在我们深入讨论各个产品之前，你之前提到过，在医疗领域，那些早早做出AI选择的医疗系统，一旦发现很多产品其实平平无奇，就开始重新寻找替代方案。
Before we dive into the individual products, one thing you mentioned at health was that you expected health systems that made choices about AI early on, you know, choosing a product to start exploring alternatives once they realized how mediocre a lot of those products were.
你有没有已经开始看到这种情况发生了？
Have you started seeing that in action yet?
有的。
Yeah.
百分之百如此。
A 100%.
现在能看到很多医疗系统重新启动招标流程了。
Seeing a lot of different health systems going out for tenders again.
我知道，至少对我们来说，现在有更多我们以前根本没听说过的公司邀请我们参与。
I know, at least for us, we've been we've been invited to a lot more that we were never have heard of in the first place.
这挺有意思，因为我觉得临床实际和高管们在董事会里为了显得懂行而说的话之间，存在很大差距。
And it's interesting because I I think there's this there's a gap between, like, clinical reality from maybe what the c suite have to say to sound intelligent in their board meetings.
你懂的。
You know?
在这种环境里，大家都会说，‘哦，对，没错。’
The in that environment, like, oh, yeah.
文书工作已经变成大宗商品了。
Scribing's commoditized.
完全商品化了。
Commoditize.
这东西已经没戏了。
It's, you know, it's dead.
以后会免费提供，而且功能强大，但说实话，我同意这个看法。
It's gonna be for free and epic, and, you know, and, like but but I actually agree.
我觉得，这种特别狭窄的场景——比如创建一个SOAP节点，然后在问题导向的清单里把它优化好——迟早会被标准化、变成大宗商品。
Like, I think this, like, very narrow, like, creating a SOAP node and making it well optimized in a problem based list is largely gonna be commoditized.
所以我们才在2024年把它免费开放了。
Like, that's why we made it free in 2024.
我们一直就是这么想的。
Like, we we always thought that.
真正没法被标准化的，是能为医生和医疗系统带来实际价值的工作。
What isn't commoditized is being able to do, like, useful work for clinicians and health systems.
你现在就能看到，那些用Claude Code或者类似工具的机构，已经开始出现这种趋势了。
So, like, you're seeing that trend in organizations that use Claude code or co work.
想想任何一个行业、任何一个场景里，每天要完成的工作多得数不清。
Like, if you think of all the jobs that get done in any business, any setting, there's a huge list.
如果你把它们拆成一个个单独的功能，没错，每个小功能确实都容易被标准化，但真正重要的是整体的协同效应——这也是为什么我们一直管它叫‘AI伙伴’。
And if you dissolve it down to one feature, yeah, like any one feature is commoditized, but it's the, like, overall, like that's why, again, we always call it like an AI partner.
关键在于Heidi能为你完成的整个工作量。
Like, it's like the overall volume of work Heidi can do for you.
是的。
So, yeah.
我觉得，随着大家逐渐意识到，也许我需要一个能虚拟运行我桌面、自动移动鼠标、创建文档、幻灯片和表格，还能打电话给病人的系统，它几乎能代替一个人的工作。
I think I think, like, perspectives that will change as they realize that, okay, like, probably I want this the system that can run my desktop virtually and, like, move the mouse and can create documents slides, and sheets, and can call my patients and can basically like be a head count.
也许我该用它来做记录，因为这样能给系统提供足够的上下文，让它为我完成最多的工作。
Maybe I I should use that for scribing because that's gonna give that system all the context it needs to be able to do the most work for me.
对。
Yeah.
这其实就是我们Soapbox一直在跟大家说的事，不过我们已经看到越来越多的系统开始回流了。
So that's basically what we're we're this is Soapbox, you know, telling everyone about, but we're already seeing lots of systems coming back around.
但还早呢，他们签的合同大多是一到三年，甚至更久。
But it is early, like, a lot of the contracts they would have signed would be anywhere from one to three years or longer.
原来如此。
Oh, okay.
所以，我不觉得这会是短期内就能见效的事。
So so, like, you know, I don't expect it to be a short term thing.
未来几年里，我们会看到这个转变逐步发生。
It'll it'll be over the next few years that we see the transition.
这说得通。
That makes sense.
你提到的一点，我过去几个月和很多人聊过，这些平台其实大多可以互换，除非有企业级合同。
I think and one thing you mentioned that I've had a conversation with lots of people in the past even few months is a lot of these platforms are sort of interchangeable unless there's an enterprise contract.
基本上，点一下按钮就能切换了。
You can, you know, switch with a click of a button, essentially.
但我觉得，语音助手这个功能更难替代，因为你一旦习惯了它的工作流程，记笔记的方式也定型了，再想换平台、从头开始就特别麻烦。
But the scribe, in my opinion, is a stickier feature because once you have the workflow and the notes written how you like them, it's very hard to switch platforms and try to start from scratch essentially.
所以我特别欣赏你这种做法，先从语音助手入手，免费开放给大家用，让大家先适应。
So I think that's one thing I admire about the way you approached it was, you know, starting with scribe, making it available to everyone for free essentially so that they can get used to it.
等大家用顺手了，再慢慢加入其他能提升效率的功能。
They can get comfortable before introducing these other features that could add to the workflow.
是的。
Yeah.
是的。
Yeah.
很好。
Great.
是的。
Yeah.
而且我觉得，从很多方面来说，这个更难一些。
And it's also, I think I think a harder one in many ways.
原因有很多，我们以后可以细聊，但说白了，这确实是个更复杂的使用场景。
So, like, for for a million reasons we can get into, but I think the practical ones is it's actually a harder use case.
要让语音转文字稳定运行，工程上难多了。
Like, it's way harder engineering to make scribing stable.
做个演示反而特别简单。
It's super easy to launch a demo.
随便谁都能上Lovable，说一声‘帮我用Whisper和某个大模型做个demo’，然后就跟朋友吹自己搞定语音转文字了，好像特别容易似的。
Like, anyone can go on Lovable and be like, yeah, you know, Vibe code me, whisper, and like some LLM, and then like go tell their friends they've solved scribing and it's easy.
但当你每天处理上百万次会话时，每一次会话丢失都像是一位怒火中烧的医生。
But when you're doing like a million sessions, every single lost session is like a raging doctor.
这真的会对病人的照护造成实质性的伤害。
Actually, like like real real damage to patient care.
你知道吗，如果我一天要依赖这个系统三十次，却没记下笔记，最后笔记没出来，那就是一场彻底的灾难。
You know, if if I'm relying on the system 30 times a day and I'm not writing notes, if I don't get my note at the end, it's it's like an abject disaster.
因为我在会诊时根本没做笔记，完全指望这个系统能准确还原我说过的话。
Because I I didn't make notes during, so I'm, like, completely relying on this system to represent what I said.
结果我就得拼命回想：我到底想做什么？病人刚才说了什么？这真的会让你陷入非常被动的境地。
And so then I'm trying to remember, like, all the things I wanna do and, like, what did the patient say again, and it it can put you in a really difficult spot.
所以每次有人推出一个语音转录工具，我都会不自觉地耸耸肩。
So this is why I always have kinda just showed a shrug, like, whenever anyone launches a scribe.
我们真的花了快十八个月，动用了我认识的最顶尖的工程团队，还有我和Moe最出色的产品思维，就为了做出一个靠谱的东西。
Like, we've literally spent almost eighteen months with, like, the best engineering team that I know and, like, the best product thinkers in myself and Moe, just trying to make good.
这真的太难了。
And it's really hard.
但问题是，这事儿根本没人觉得酷。
And the problem is, it's just like not a sexy thing.
大家都喜欢随便吐槽那些文书公司，觉得这事儿太简单了，一解决就完事。
Like, everyone loves the idea of like dunking on like all the scribing companies and it's all easy and it's all over.
但我认为，实际结果只会是用户接受度很低。
But I just think I think in practice, all this will show up as is just bad adoption, essentially.
人们可能会用一阵子。
Like, they'll people will use it.
但系统老出问题。
It'll flake.
偶尔写个乱七八糟的笔记，然后就干脆算了。
It'll write shit note once in a while, and then just like, oh, stuff this.
我再也不想用它了。
Like, I'm not gonna use it anymore.
所以我们得让大家明白，在医疗领域做临床AI，跟别的地方不一样，要求高得多。
And so it's our job to establish that, like, clinical AI is like, building this in health care is gonna be a different thing with a higher burden.
一旦你建立起这种信任，让患者在就诊时真正依赖你——这本身就很不容易——那你就自然有权去主导其他由就诊衍生出来的各种流程。
And and once you build that trust as, like, the thing that someone relies on in a visit, which obviously a hard thing, then, yeah, like, you feel like you have the right to a lot of the other workflows that stem from the visit.
对我来说，这本质上是一场竞赛，看是电子病历系统不断改进，还是Heidi能做得更多、更好。
And intuitively to me, it's a race between, like, the medical records and Heidi doing more and doing it better.
所以我们必须快得多，就这么简单。
So we just have to be way faster, basically.
别无他法。
There's no other way.
没错。
No.
我完全同意。
Totally agree.
你对使用的文书助手的信任，远超过像Evidence这样的工具。
You put a lot of trust in the scribe you're using compared to, for example, Evidence.
你一分钟内就能在四个不同的平台上查一遍信息。
You can just query four different platforms in the span of a minute.
但真正做记录的时候，你得相信它能准确记录下这次问诊的内容。
But for actually scribing, you're putting your trust that it's gonna record this visit accurately.
它能正确获取所有信息。
It's gonna get the information correct.
它还能记住并识别出对话中真正重要的部分来记录下来。
It's going to remember identify the correct parts of conversation to sort of record.
所以你对记录员的信任程度很高，这一点我完全同意。
So putting a lot of trust in the scribe, which totally agree with.
是的。
Yeah.
对。
Yeah.
没错。
Yeah.
完全正确。
A 100%.
不错。
Cool.
我们来聊聊你新推出的一个功能吧，我记得上次谈话快结束时你简单提过，说是即将上线的功能。
To shift to one of your newer features, I think this is the one you briefly mentioned at the end of our conversation last time as an upcoming feature.
现在这些功能好像改名叫‘Comms’了，我们很想多了解一下这个功能。
The calls are now I think it's rebranded as Comms, but we'd love to hear more about this feature.
你们几个月前推出的这个‘Comms’功能具体是做什么的？
What is the Comms feature that you released a few months ago?
对。
Yeah.
它整合了前台服务，你可以通过它安排预约、核对保险信息，还能通过短信、电话和网页聊天来调整就诊时间。
So it does integrated front desk, so you can do scheduling, insurance verification, moving appointments by text, phone call, and over web chat as well.
你可以设置一个专属号码，患者随时都能联系到你的诊所，一天24小时、一周7天都行。
So, you know, you could set up a phone number that a patient can basically reach your clinic at any time 247.
另一个特别有意思、能让我们明显区别于其他产品的应用场景，是在美国那边常被称作‘AI护理’的那部分。
The the other interesting use case where I think we differentiate quite a lot is more in building in The US, they kinda call it like AI nursing.
这有点像我们一些竞争对手的做法，比如Hippocratic，他们设置智能代理来提供一些护理服务。
So sort of what some of our competitors do, like Hippocratic, where you set up agents to do some care delivery.
所以我们也有类似的想法：每次就诊结束后，可以自动安排跟进电话，并且这些电话会带上就诊时的完整背景信息。
So we have a similar idea where at the end of every visit, you can schedule calls that have the context of the visit.
比如我可以跟进你膝关节置换术后的恢复情况，了解你的症状反馈，同时还能记住你上次就诊时的具体情况——我觉得这才是这个工具真正强大的地方。
So I can check-in on you about your knee replacement and see how it's going and and get patient reported symptoms, but also, like, have the context of what happened in the visit, which which I do think it makes it much more powerful as a tool.
这样一来，效果就提升了很多，因为你不是在一次次重新开始，就像第一次约会一样。
So it just makes it more effective because you you're not it's it's like you're it's like it's fifty first dates.
我这么说有点老了，但你想想，你跟德鲁·巴里摩尔聊天，她连五分钟前说过什么都记不住。
I'm aging myself, and you're talking to Drew Barrymore, and she can't remember, like, what happened five minutes ago.
如果医生每次跟你见面都像第一次认识你一样，完全不记得之前的事，那这医生也太不靠谱了。
Like, a doctor wouldn't be particularly great if it forgot if they forgot everything about you every time they interacted with you.
再说回Comms，我知道市场上有成千上万的竞品，大多数公司都会推出自己的版本。
So, again, like with Comms, I know there's like a million competitors and like most of the records will launch their own things.
但归根结底，你真正想要的，是一个能真正了解你的AI对话系统。
But ultimately, I think you wanna be able to have a conversation with an AI system that knows you.
所以，它能访问你的病历，掌握你所有的用药历史，未来甚至可能帮你自动开药，还能根据你的病史，结合医学证据，先拟出一个初步方案，再推送给医生，说：嘿，我觉得咱们该这么处理。
So, like, can access the record, has all your medicines history, maybe one day can do repeats for you, perhaps even, like, take your history and, like, come up with an initial plan based on evidence and then surface that to a doctor and say, hey, here's what I think we should do.
你对这个安排觉得怎么样？
Are you okay with that?
这些系统最终都会彼此联动。
Like, basically, these things will interact.
现在很多功能都是孤立的，这其实是美国医疗市场销售模式造成的扭曲，但长远来看，这根本不是最终方向。
Any any, like, feature these, like, single threaded features is like an economic distortion because of how you have to sell in The US, but it's not actually where this ends up.
这跟Claude Code是一个道理。
So it's the same as like Claude Code.
归根结底，海迪已经把所有东西都整合了，你为什么还要用别的系统？你希望你的工具能掌握所有患者信息、最新医学证据，还能跟患者顺畅沟通。
Like, in the end, like, Heidi was through everything and why would you have, like, any other system because you want your tool to be able to have all the contacts, have your the evidence you want, and have it communicate with your patients.
未来一两年内，肯定会出现成千上万的竞争对手。
In the next one or two years, yeah, there'll be, like, a million competitors.
我就不一一列举名字了，但我觉得我们目前处于有利位置，因为我们有全球化的覆盖能力，能打造更通用、更长远的解决方案——我们公司不会因为拿下某一家医疗机构就成功，也不会因为没拿下就失败。
There'll be, you know, I don't have to name them all here, but there it's always gonna be I think I think we're in an advantage position because we have such a global reach, and we we can build, like, more horizontal things that make more sense in the long term because we're not we're not gonna live or die as a company based on us winning the next health system.
所以，对CMIO来说，直接说‘我需要个接待员’就更简单了。
So, like, it's easier for a CMIO to be like, I need a receptionist.
没错。
It's like, yep.
没问题。
No worries.
这个我们能做到。
Like, we can do that.
但五年后，你整个系统差不多一半都是靠Heidi在运转了。
But but in five years, it's like, well, basically, half your system is just run by Heidi.
这种价值主张，让你替换掉现在用的那些系统，根本不是难事。
And that value proposition, it's not gonna be hard to swap out a lot of the systems that you're seeing.
总之，我觉得这才是正确的方向，关键还是得落地执行。
Anyway, that this this is what I think the right strategy is, but, you know, gotta gotta execute, basically.
这说得通，本质上，有了完整的平台，每个功能都能更突出，更好地服务每个患者或医生，这就形成了一个完整的体系。
That makes sense, essentially, instead having the whole platform helps each individual feature stand out more and work better for each individual patient or provider so that that makes sense into a little form.
是的。
Yeah.
对。
Yeah.
这也不是什么新点子。
And it it's not in a new idea.
举个商业上的例子，你可以想想Salesforce或者HubSpot这样的系统。
So, like, to use a business example, you can think of a system like Salesforce or HubSpot.
没错，你当然可以单独用一个CRM系统、一个独立的网站、一个单独的表单工具、一个独立的预约链接，但到了某个阶段，如果你想优化整个系统，干脆就选一个统一的平台。
It's like, yeah, you could have a separate CRM, a separate website, a separate, like, forms product, a separate meeting link, a separate but at some point, it's like when when you wanna optimize the system, you just wanna pick one thing.
实际上，我觉得像Heidi这样的产品会越来越有吸引力，主要是因为医疗设备和安全方面的考量。
And actually, the reason I think there'll be a lot of gravity towards a product like Heidi is because of medical devices devices and safety.
我们都得做临床合规和管理，这些工作，任何单一功能的工具都做不到。
Like, we all have to do a level of clinical compliance work and governance that that any point solution won't do.
举个好例子，比如Evidence。
So, like, a good example is with Evidence.
我们绝不会靠制药公司的广告赚钱，因为我们知道，最终我们是要开处方的。
Like, we would never monetize with pharma ads because we know that we're ultimately gonna prescribe one day.
这意味着，你不能让一个受礼来 Lilly 资助的医生去开 GLP 类药物。
And that means you can't have a doctor that's funded by Eli Lilly prescribing, like, GLP ones.
这简直太荒唐了。
That's, like, crazy.
你懂的。
You know?
对吧。
Right.
这明显是利益冲突，太离谱了。
Like, it's crazy, like, conflict of interest.
但眼下，这确实是个利润极高的模式，Open Evidence 团队会赚得盆满钵满。
But right now, of course, that's like an incredibly lucrative model, and the Open Evidence team will get, like, crazy rich.
但长远来看，这真的对吗？从道德和伦理上讲，这真的合适吗？
But, like, is it the right long term or, like, right moral ethical thing to do?
但如果你只专注在 pie 的某一个部分，其实效果会非常好。
But but, like, it's really good if you're just focused on, like, one area of the pie.
而且说实话，我希望自己能一直干下去，这行我打算干一辈子。
And, yeah, I think I think I'm hoping like, I'm in this like, I'm a lifer.
我一直都想做这一行。
Like, I've always wanted to do this.
我相信，未来十年，医疗健康领域将是 AI 变革最彻底的行业之一。
This is like I I believe health health care will will be, like, one of the most transformed industries by AI in the next ten years.
如果你的目标就是这个，那我前面说的这些——想提升服务能力，就得这么干。
And if that's your aspiration, then then, like, all the things I'm saying, like, if you wanna increase capacity, this is how you do it.
你得把这些模块一个个建起来。
Like, you you build these components.
对。
Yeah.
我们在每一个垂直领域都打得非常激烈，必须赢过那些只专注单一领域的竞争对手。
We compete viciously on each vertical, and we have to win against competitors who are just focused on that that area.
但归根结底，最终只会有一种方式。
But in the end, it'll be one thing.
所以，像那些其他产品，甚至电子病历，都只是暂时的、片面的视角，看不到未来真正的样子。
And so, like, all these other products, even the medical records, like, they're all just, like, short term short term views of, like, what this will ultimately be.
不是每个人都能明白。
And not everyone will get it.
有些人根本理解不了。
Some people won't understand.
还有些人没耐心去关心。
Some people won't have their attention span to care.
但到最后，如果你有个工具能帮着医生做检查、开药、诊断，全程协作，功能还重叠，那你还干嘛非得用别的东西来处理行政事务呢？
But in the end, like, if you have a tool that can investigate, prescribe, diagnose, do the whole thing in partnership with a doctor and have some overlapping, then why would you why would you use, like, something else for, like, the administrative part?
你懂我的意思吧？
Because, you know what mean?
不管怎样，我觉得关键还是回归本质：到底什么东西是真正能改变医疗行业的核心要素。
Like anyway, that's that's I think it's just like trying to be first principles, like, what what has to exist to really, like, change health care.
当然，短期内我们还是要开发医生能用、能卖给机构的产品。
And then and then, of course, in the short term building product that doctors can use and we can sell to organizations.
但说实话，在这个行业里工作确实挺让人沮丧的，你总听到有人说：‘这不就是个聊天机器人吗？’
But, yeah, it's it's always because it can be frustrating working in the industry, like, you just, you know, you're you're always getting like, oh, isn't this just chatty bitty?
或者：‘为什么不能直接让某家公司来搞定这个？’
Or like, isn't this just like, can't can't just, like, insert company do this?
可问题是，他们确实都能做出某个功能。
And it's like, yeah, they they all can do, like, a feature.
但他们能像我们这样完全打通、无缝协作吗？
But can they do it as well as we can all connected?
绝对不行。
Definitely not.
没错。
Yeah.
我就这么认为。
That's what I think.
不。
No.
我觉得你在某种程度上，是在复制谷歌的成长路径——从一个搜索引擎，发展成一套完整的系列产品。
I think you're, in a way, following the model that Google used to grow from, you know, just being a search engine to having this entire suite of products.
虽然你们所处的领域和谷歌不一样，不是医疗健康，但你们在医疗领域采用的模式非常相似，我觉得这个思路特别有意思，潜力很大。
You know, they're in very different space, not health care, but follow you a very similar model in the health care space, which I think is a really interesting idea, and there's a lot of potential there.
我觉得Comms和你们主打的文书助手功能很不一样。
I think one thing about Comms is that it's very different from your primary scribe feature.
对吧？
Right?
它是面向患者的。
It's patient facing.
这背后会带来一整套全新的问题和障碍，风险也比做文书助手或者证据收集要高得多。
There is a whole new like, whole suite of potential other problems or obstacles to overcome, and there's a lot more risk there as well, I think, compared to doing a scribe visit or even Evidence.
你是怎么看待这些风险的？风险和回报之间你是怎么权衡的？还有，推出这个面向患者的新功能时，最难解决的问题是什么？
How did you think about sort of the risk, and how did you think about what the risk reward there, I guess, and what was the hardest problem in launching that new patient facing feature?
是的。
Yeah.
这是个好问题。
That's a good question.
实际上，很多都是技术上的挑战，比如让海蒂足够有同理心，能在复杂的环境中妥善处理对话。
Actually, a lot of it was technical, like, making Heidi empathetic enough, manage, like, conversation in difficult environments.
从临床风险的角度来说，这确实是我们在反复思考的问题。
From the clinical risk side, it's definitely something we think a lot about.
我们其实一直都有保险，为海蒂在不同产品中的表现提供保障，尤其是在医生使用时。
So so, like, we have actually always had insurance that underwrites, like, Heidi's performance across different products when when clinicians are using it.
这可不是我们随便说说而已。
So it's not something we can we just say.
我们真的是用实际行动来证明自己的信心。
Like, we we actually put our money where our mouth is.
但目前它主要还是集中在安全的行政流程这类保守的场景上。
But it's still it's still mainly fairly guardrail towards, like, safe administrative workflows today.
说实话，我平时爱谈世界会往哪儿走，但今天，我们主要还是在收集病史和行政事务这一块。
You know, I I do pontificate about where the world goes, but but, like, today, we're still mostly in that history gathering administrative side of things.
当我们提供信息时，对能做什么、不能做什么，我们都有非常严格的限制。
And when we're if we're giving information, we we have, like, very rigorous constraints about what we can and can't do.
是的。
Yeah.
我觉得这个风险模型，跟医生们其实有不少共通之处。
I think it's interesting, like, the risk model generally shares some of it with the clinician.
这取决于他们的诊疗习惯、医院的运作方式，还有他们通常会接触到哪些类型的病人。
So it depends on their practice, depends on how their their hospital works, what kind of patients they're likely to see.
我们一般会让医生们参与进来，一起做这些决定。
And we generally let let the doctors kinda share in that that decision making.
毕竟，最终要对 Heidi 的使用负责的是医生，我们只是在背后支持，确保它尽可能负责任地运行。
Like, ultimately, they're responsible for the Heidi they set loose in the world, and we're here to support it and make sure it's as responsible as it can be.
但我觉得，从病历记录助手这个应用来看，医生们会自己选择那些合适的使用场景。
But if I think what we're seeing with scribe is that clinicians will self select for sensible use cases.
你知道，让海迪现在去管理自杀热线，这显然不太合适。
You know, like, probably is not responsible to have Heidi managing a suicide hotline right now.
我们还没达到那个证明标准。
Like, we'd haven't met that burden of proof.
也许有一天可以，但就目前来看，确实不行。
Maybe one day, but right now, like, sure.
也许它能当个初步筛选工具，但不太适合做决定，或者给出关键性建议。
Maybe it's like the first screener, but probably not as something that's deciding, like, or giving, like, really critical advice.
证据系统也是类似的情况。
With Evidence, it's similar.
我们有一套非常严谨的引用机制。
Like, we we have, like, a very thorough citation logic.
整个过程完全透明。
It's completely transparent.
你可以控制信息来源，这一点在海迪身上是独一无二的，据我所知。
You can control sources, which is unique to Heidi as far as I'm aware.
你可以自行选择偏好的信息来源，还能精确控制答案具体来自哪些地方。
Like, you can pick which sources you prefer and, like like, very fine grained control exactly where your answers are gonna come from.
我们之所以这么做，是因为在全球范围内，尤其是在英国和欧盟，这种系统很可能被认定为医疗设备——因为他们认为医生会直接依赖这些回答，而不会去核对来源。
So the reason we had to do that is because in a global setting, that is a likely to be a medical device in The UK and EU because they take the perspective that the doctors will rely on the answers and they won't review the sources.
所以我们必须确保对信息来源的呈现完全准确。
And so we have to bear the responsibility that we accurately represent the sources.
我们已经测试过所有类似的工具。
And, like, we've tested all the tools.
它们都会以不同程度产生虚假信息。
Like, they all hallucinate to varying degrees.
每个工具都有问题。
They all have problems.
所以我们收购了Automedica，因为他们采用了一种确定性的引证方式，我们最终把这套机制整合进了产品里。
That's why we acquired Automedica because they had a deterministic approach to citations that we end up bringing into the product.
所以我觉得，相对而言，我们的风险还是更低的。
And So I think I still think, like, relative speaking, it's lower risk.
我记得刚当住院医生的时候，还经常上网查东西，然后拿去给病人用。
Like, I remember googling stuff that I went on to inflict on patients, you know, as a junior doctor.
所以我觉得，归根结底，这还是给临床医生用的工具。
So I think, ultimately, it's still a tool for the clinicians.
是他们在操作使用。
They're wielding it.
他们可以自己做判断。
They can decide.
但一旦涉及到让系统独立为病人做决策，那就是另一个层面了，我们必须对自己有更高的要求。
When when you get to, like, autonomous decision making for patients, that's like a different level of scrutiny and ballgame, I think, that we have to hold ourselves to.
没错。
No.
这说得通。
That makes sense.
对。
Yeah.
归根结底，还是医生自己决定要不要用这个产品，也该由他们判断一次就诊是否太复杂，或者某个话题适不适合交给Heidi这样的模型处理。
At the end of the day, the clinician is the one choosing to use this product and should have the clinical judgment about whether a visit's too complicated or if, you know, if topic is appropriate for a model like Heidi.
这么说就合理了。
So that makes sense.
没错。
Exactly.
不过，我这还是第一次听说保险的事。
This is the first I've heard about insurance, though.
我压根没想到还有这一层。
I didn't realize that was a thing, actually.
是的。
Yeah.
对。
Yeah.
没错。
Yeah.
而且很明显，很多平台都会这么做，万一平台表现不如预期，或者造成伤害的话。
And, obviously, like, I think a lot of platforms have it, you know, in case, like, the the for whatever reason, the the platform, you know, doesn't perform as expected or causes harm.
但话说回来，既然我们是为医疗行业打造的，这么做才是负责任的，因为我们是正经的医疗企业，不是那种靠医疗套利赚钱的初创公司，像某些YC公司那样。
But, yeah, I think, again, like, back to being built for health care, like, that's a responsible thing to do because we are a real health care business, not not just, I don't know, some YC company or, like, found some, like, health care arbitrage where you can make money out of.
如果你真想进入医疗领域，就必须负起责任。
Like, if you're trying to be in health care, you have to be responsible.
没错。
Yep.
完全同意。
100%.
我完全赞同。
Totally agree.
你有没有考虑过让Heidi更多地面向患者？还是说目前的主要目标仍然是辅助医生和医疗提供者？
Do you foresee Heidi expanding more into the patient facing space, or is your goal primary right now to sort of assist the clinician, the provider?
是的。
Yeah.
这有点复杂。
It's a bit tricky.
说实话，这挺棘手的，因为我们主要服务的是医生。
It's a little tricky, I think, because ultimately, we serve clinicians.
根据我们的经验，我自己也直觉地觉得，医生们并不太喜欢患者用Heidi产生的那些对话和数据，还觉得这些是患者该自己负责的事。
And from our experience, and I even know, like, intuitively myself, we doctors don't love the idea of having these conversations or this, like, data that patients are creating with Heidi that they are responsible for.
所以这其实是个微妙的平衡——毕竟，作为患者，谁都希望有这样的功能。
So so it's it's this, like, tricky balance of, like, it's obviously something everyone as a patient wants.
我们都希望能获得建议、提问题，或者帮我们理清那些复杂得让人头疼的医疗体系。
Like, we all want the ability to get advice or, like, ask questions or or help us navigate the complicated health care systems that we all we all live in.
所以我个人非常有兴趣去做这件事。
And so I definitely am very interested to do it.
但我们一定要非常谨慎，确保它真正为我们的诊所和医院带来价值，而不是干扰医生和患者之间的关系，而是起到补充作用。
It's just something we wanna do really carefully in a way that's that adds value for our practices, our hospitals, and doesn't doesn't get between the relationship they have with their patients and is additive.
但我确实能看到它的潜力。
But I definitely see it.
你可以想象，患者端的证据应用，通过通讯工具帮他们预约就诊，或者整理整个诊疗过程中的各种事项。
Like, you know, you can imagine patient patient facing versions of evidence using Comms to help schedule appointments for on behalf of patients or organize different things in their care journeys.
所以我们确实很感兴趣，也在积极关注，但目前还没有任何正式的公告。
So it's it's definitely something we're we're we're actively interested in, but don't have any, like, strict announcements.
不过我觉得，在就诊过程中向患者介绍这个概念是个不错的时机。
But I do think being in a visit is a good place to introduce them to the idea.
这背后确实有潜力。
So there's definitely something there.
关键是要把执行细节做到位。
I would just have to nail the execution.
这说得通。
That makes sense.
你一再提到的一件事，也是我们过去几个月里最重磅的发布之一，就是Heidi Evidence。
And one thing you brought up over and over again, and obviously, is one of the bigger press release we had in the past few months is Heidi Evidence.
这确实是个全新的产品。
It's a really you know, it's a new product.
这和传统的记录员，甚至和Comms功能都完全不同。
It's very different than the typical scribe or even the Comms feature.
但就像你之前说的，从直觉上看，这确实说得通。
But like you said already, intuitively, it makes sense.
你可以打造一个完整的平台，从头到尾支持临床医生的工作。
You can build a whole platform to sort of support the clinician from end to end.
你已经提到过一点，让Heidi与众不同的是，你可以选择从哪些来源获取信息，以及优先使用哪些。
What you've already mentioned one thing that makes Heidi unique in that you can select which sources to you want to get information from, which ones to prioritize.
在这样一个竞争激烈、越来越难做出区别的市场里，你还会从哪些方面来突出Heidi Evidence的优势呢？
How else are you thinking about sort of differentiating the the Heidi Evidence tool in a very crowded market that is sort of getting harder and harder to differentiate yourself in?
是的。
Yeah.
我觉得比起具体的来源或方法，更重要的是产品的运作方式，从我的角度看是这样。
I think more than any particular source or approach, I think it's the from my perspective, it's the product the way the product works.
如果有人用过，就会发现它更像一个能调用多种工具的智能对话系统。
So if if anyone's used it, it's much more like like an agentic chat that has access to a load of different tools.
你可以导入患者联系信息，创建文档和PDF，发送邮件，还能生成表格。
So you can load in patient contacts, you can create documents, PDFs, you can email people, you can create sheets.
你还可以让它总结你这一周的情况，它会自动分析所有会话记录，智能整理出摘要，如果你把所有内容都转录了，它还能帮你做精准的周度复盘。
And, like, you can do things like summarize my week, and it will look through all the sessions, load it in agentically, summarize it for you, do a surgical audit on your week if you transcribed everything.
我觉得这更像一个为临床工作者提供支持的核心平台。
So it's I think it's I see this more as like a backbone for doing work for clinicians.
精准回答循证问题非常有价值，而我们实现这一点的方式和其他人不一样。
Answering evidence questions incredibly well is very valuable, and our approach to doing that isn't isn't the same as others.
其他人试图让模型专门化，训练多个代理去内化各种问题的答案。
I think others are trying to almost get the model to, like and specialize different agents to internalize the answer to questions.
而我们的方法更偏向搜索型，因为我们服务的不仅是个人用户，还有大量机构，他们都有自己的数据和信息来源，希望上传进来。
Whereas our approach is more slightly more search oriented because we know that our because we serve a lot of organizations as well as individual users, they have their own data, their own sources that they wanna upload.
而且，我觉得在美国，这种模式相对独特。
And I as again, in The US, I think it's relatively unique.
Heidi里有个资料库功能。
Like, we there's a library feature in Heidi.
你可以上传任何你喜欢的资料来源。
You can upload any sources that you like.
你可以创建属于你自己的资料库。
You can create all your own collections.
你从海蒂那里得到的所有答案，都会基于你所在机构的偏好和数据来控制。
And then all the answers that you get from Heidi are are controlled within your organization's preferences and data.
所以我觉得，一方面，我们确实得确保答案足够精准。
So I think of this as on one hand, yes, we have to make sure our answers are great.
但我觉得人们最终选择海蒂的原因，是它能更好地融入他们现有的工具体系，还能帮他们完成临床工作。
But I think why someone would ultimately choose Heidi is it works better with their existing tool set and that it can do a genetic work for them.
其实，懂医学、能准确回答循证问题，我觉得将来会变得越来越普遍。
And actually, like, knowing medicine and having the right answer to an evidence based question, I think it will become increasingly commoditized.
所以，真正的价值并不在这里。
So so, like, that's not where the value accrues.
真正的价值在于用这些信息去做真正有用的事，而在这方面，我觉得我们做得更好。
The value accrues to, like, doing useful work with that information, which I think we're a better place to do.
展开剩余字幕（还有 311 条）
但我们得证明这一点。
But we just have to prove it.
我们必须做出更好的产品，赢得医生们的信任，让他们愿意转向我们。
Like, we have to build something better and win doctors' hearts and minds and get them to come across to us.
没错。
Right.
不对。
No.
这话说得通。
That that makes sense.
我觉得你说得对，这东西迟早会变成标准化服务。
I think to your point, the it's gonna become commoditized.
所以你得抢在前面，一开始就要比对手更出色，让使用者觉得你的平台有价值，而且和其他工具整合得特别好。
So you just have to be sort of first, you have to be better in the competition initially to sort of get people using your platform or ones they're using, and they see the value and how integrated the other tools.
这么说确实有道理。
It sort of makes sense.
你最好固定用一个平台，别老换来换去。
You wanna stick to one platform and not switch around or jump around from platform to platform over time.
对。
Yep.
没错。
Yeah.
我同意。
I agree.
你提到Heidi Evidence和其他竞品相比更偏向搜索功能。
You mentioned that you Heidi Evidence is more of a search based compared to some of the competition.
能再详细说说吗？
Could you go into more detail about that?
比如说，我输入一个情况：一位48岁男性，有胸痛病史，来就诊，可能是某种症状。
So let's say I input something asking about, you know, 48 year old male with a history of chest pain presenting for, I don't know, whatever condition.
系统是怎么理解这个输入，分析之后，给出一个能直接帮到我临床决策的回复的？
How does the model take that, analyze it, and come up with the output that is relevant to my visit and I can act on right now.
是的
Yep.
对
Yes.
根据你的提问，系统会先进行协调，决定使用哪些工具以及搜索哪些内容。
So so based on your inquiry, there's an orchestration step that's deciding what what tools to use and then what queries to search.
比如，可能会搜索人口统计信息，或者根据你设定的条件，比如只查北卡罗来纳州的数据，或者杜克大学发表的研究之类的，来筛选不同来源的信息。
So, you know, maybe that searching for demographic data or we run a search for different sources filtered by, you know, whatever constraints you put in, like, only in North Carolina or, you know, more like studies that were published at Duke or you know?
所以，本质上这是一个协调过程，这个过程能调用海蒂拥有的所有工具，未来我们还会让海蒂接入更多任意的功能。
So, yeah, I think, basically, it's an orchestration step, and that orchestration step has access to all the different tools that Heidi and in the future, we will have access to basically any arbitrary ability that Heidi has.
从查阅病历、给患者发短信，到打电话，你完全可以把它理解为海蒂在替你做事。
So everything from, like, reading things in the record to texting patients to calling, like, you can think of this as just, like, Heidi doing work for you in the end.
你可以给它一个复杂的任务，比如确保普拉纳瓦做好手术准备。
So you could give it a complex objective, like, make sure Pranava is prepared for surgery.
然后海蒂会制定一个计划，持续四周跟进，帮你查找证据，确认麻醉师对你麻醉的安全性，确保你做好了准备，心里有底，还能随时向它提问。
And then Heidi comes up with a plan and executes that over four weeks to check-in with you, look up evidence, make sure that, like, you know, you're likely to get your anesthetist is likely to be happy putting you to sleep and make sure that you're ready and you feel confident and you you can ask questions from it.
这跟现在直接回答一个问题其实是不一样的。
And so so that is different to, like, just answering a question right now.
做这件事的时候，得查不少证据，比如，感冒后接受麻醉到底有多危险。
Like like, there are evidence pieces in doing that task, you know, looking up, like like, how dangerous is it for someone to go under an anesthetic after they've had a cold.
因为在整个过程中，你可能会说：‘我最近有点流鼻涕，好像要感冒了。’
Because somewhere in that work, you you say, oh, I feel like a bit sniffly.
我觉得自己快感冒了。
I think I'm getting a cold.
所以我觉得，这种方式要实用得多。
So so it it's I think it's just more useful.
如果让我总结一下目前见过的那些工具，很多都特别依赖原始研究。
Like, it's less I I think generally, like, if I'm summarizing the tools that I've seen out there, some are very anchored to primary research.
它们主要做的是系统综述，回答一些临床研究类的问题，这当然有用，但漏掉了临床实践中的实际处理。
So they're trying to do, like, meta reviews and give you, like, clinical research oriented questions, which is useful, but, like, but that is skipping medical editorialization.
研究结果和实际医疗操作，根本不是一回事。
Like, the research does not equal, like, practice.
研究当然是关键依据，但实际做法更多取决于医学界的共识，也就是医生们普遍认同的处理方式。
So research is obviously a key input to that, but actual practice is dictated by, like, edit medical editorialization, what doctors' consensus agree is, like, the thing to do.
而且不同国家、不同城市之间的做法差异很大。
And that varies a lot by country and city.
比如抗生素耐药性的问题就各不相同。
Like, there's antibiotic resistance.
各种疾病的流行模式也不一样。
There's different patterns of disease.
我刚认识一位日本骨科医生，他手头有明确的数据，显示全髋置换其实不如只换股骨头。
Like, I just met an orthopedic surgeon in Japan who has clear data that that total hip replacement is probably inferior to just the ball replacement.
我一时想不起那个问答环节的具体内容了。
I'm forgetting the the Q and A front.
因为日本人的平均体重指数普遍偏低，而且做髋关节置换时年龄通常更大。
Because in Japan, the they generally have lower BMIs, and they're getting the hip replacements at a more elderly age.
所以就不必太担心髋臼表面的使用寿命了。
So you don't need to worry about the the life of the acetabular surface.
所以手术更简单，麻醉时间更短，并发症也更少。
So so it's a less complicated surgery, less time under anesthetic, less complications.
在这种情况下，这种做法大概是可以接受的。
And in that setting, like, that's probably an okay practice.
但在美国，全髋置换术还是更占主流。
But in The US, total hip replacement trumps the board board replacement.
对的。
Yep.
但你在原始研究里根本找不到这种说法。
So the but you won't you won't find that in the primary research.
可能有一篇日本的论文提到过，但这远不是共识性的做法。
Like, there's probably one Japanese paper that says that, but that's not like a consensus practice.
所以研究本身并不能给你明确的答案。
So then the research doesn't tell you the answer.
于是你就遇到相反的问题：人们只盯着指南，用指南来给出答案。
Then you have the opposite problem where people just focus on guidelines only, and they give you answers based on guidelines.
但这样就忽略了另一个关键问题：最新研究到底怎么说？
But that also misses some of the nuance of like, what's the latest research say?
作为医生，我一直这么想：我这么做是有理由的，但有没有什么证据能证明这做法是对的呢？
Well, like, I've always felt as a doctor that like this, I do this, but I don't like, is there anything that's would would say that this is, like, correct?
这更多是凭我的直觉。
Like, it's my intuition right here.
而海蒂的优势就在于，我们两种方式都兼顾了。
And and basically, where Heidi will excel is, like, we have both.
所以它完全按你的需求来运作。
So so it's an agent at your whim.
我们并不是为了某个特定目标去过度优化。
So it's we're not, like, we're not over optimizing for any end task.
我们只是相信模型的发展趋势。
We're just we're believing in the trajectory of models.
我们相信，模型会变得足够强大，到时候，针对临床研究做微调和针对指南做优化，其实没什么实质区别了。
Like, we believe the models will get good enough that there won't actually be a meaningful difference between having micro hyper optimized for clinical research or hyper optimized for guidelines.
你只要把工具和数据给它，用上加速硬件，再配上一个近乎无限的上下文窗口，它自然就能搞清楚该怎么满足你的需求。
You just, like, give the tools the source, use accelerated hardware, and use, like, an effectively infinite context window over time, and it will figure out, how to answer what you want, basically.
不好意思。
Sorry.
我知道咱们现在说得有点太快了。
I I know we're going on, like, turbo speed.
但你明白我的意思吗？
But does that make sense?
因为任何你现在为了某个特定方向而设下的局部最优解，迟早都会被模型的持续进化超越。
Because, like, any any, like, local maxima you create today by, like, focusing on one or another path will get trumped as models continue their trajectory.
是的。
Yeah.
我觉得是这样。
I think.
要说哪里可能不太成立，坦白讲，大概就是涉及到成本问题的时候。
The the I guess the the where this maybe doesn't work, like, be intellectually honest, would be if if it's like a cost conversation.
我觉得，如果你能稍微优化一下，查询成本其实会更低。
Like, I do think the query cost will be cheaper if you might kind of optimize.
没错。
Right.
对。
Yeah.
但就连这一点，我们也可以应用到不同的工具上。
But, like, even that, we can we can do that for the different tools.
我们可以打造一个高度优化的科研工具，再做一个高度优化的临床指南工具，它们各自都能媲美独立的完整产品。
So we could have, like, a highly optimized, like, primary research tool and a highly optimized clinical guideline tool that almost, like, compete with those whole products in their own right.
但它们本质上都是同一个通用智能工具的一部分，能完成大量任务。
But they're part of just, like, a general agentic tool that can do a lot of work.
是的。
Yeah.
说到底，我实在想象不了，未来你会跟一个提供证据的系统聊天，同时却在别的地方记录信息，而那个跟你互动、接触病人的系统却完全没有这些信息。
So basically, because I just I just, in the future, can't imagine you're chatting to, like, an evidence product, you're scribing in something else, the thing that talks, engages your patient, doesn't have that information.
然后你还会有一堆AI工具，各自为政，有的挺好，有的简直不堪用。
You then also have a record that has AI tools that are, like, disparate, and, like, some are good, some are crap.
那种情况，根本不是我想象中的未来。
And then, like, that's just, like, not the future I imagine.
我想象的是，你直接用Heidi，它能搞定95%的事情。
I imagine you just use Heidi, and it does, like, everything 95% as well.
而且这些功能组合起来，效果比任何单一工具强十倍。
And in combination, it's, like, 10 x better than anything.
但每个单独的工具，有的方面更出色，有的方面又稍微差一点。
But each one is, like, is actually, like, is better in some ways, but maybe slightly worse in some ways.
但整体加起来，比任何单独的产品都强。
But the sum of the parts is better than than anything on its own.
对。
Right.
不对。
No.
这确实很有道理。
That definitely makes sense.
想自己持续训练或微调模型几乎是不可能的，因为新模型出来得太快了。
It's next to impossible to train keep training your own models or fine tuning to a level because new models are coming out way too fast.
等你把模型训练完，整个行业的基准都已经提升了十倍。
By the time you finish training your own model, creating a new model, the benchmark across the industry has risen 10 times.
所以我觉得你说得对，通用模型现在已经越来越成熟，未来会越来越没必要自己训练模型，除非你真的想优化成本，或者有特别的需求。
So it makes sense, and I agree with your point in that the general models are getting to a point, and they will continue to get to a point where it won't make as much sense to sort of train your own models unless you really want to optimize for cost or have a specific reason to.
但就像你刚才说的，你可以用一个通用模型，按需调用一些专门针对特定任务的小模型。
But, again, like you said, you could have a general model that pulls really niche specific models for specific use cases or tasks.
这一点我完全同意。
So I totally agree with you there.
我觉得没错。
I think Yeah.
你之前还提到过，Evidence公司收购了Automedica这件事。
One thing you already mentioned also is the Automedica acquisition that you did for Evidence.
我对英国和欧洲的医疗政策了解得不是很多。
I think I don't know as much about The UK UK health policy, European health policy.
你能说说为什么这次收购这么重要吗？
Can you tell us why was that acquisition so important?
你们在英国市场要想成功，到底需要AirLock提供的那个‘沙盒’环境做些什么？
What is this sandbox that you or, I mean, with AirLock that you have to have access to in order to successfully play in the or win in The UK space?
是的。
Yeah.
我之前好像提过一点，他们做这些工具的方式，我其实挺认同的。
So I think I kinda mentioned it earlier, like, how they do these tools, which which I kind of intuitively agree with.
我觉得医生们，说实话，我也说不太准。
Like, I think I think doctors I don't know.
这一点在人机交互领域早就被广泛认可了。
This is, like, well established in computer human interaction.
在新技术刚出现的时候，大家通常都会持高度怀疑态度，对每一点都严格审视。
Like, there's a period of time where where you have, like, a high skepticism and you scrutinize everything at a really high level.
就像你第一次开启特斯拉的全自动驾驶功能时，会特别紧张，注意力高度集中。
It's, like, the first time you turn on FSD in a Tesla, you're, like, very anxious and, like, paying attention.
但很快，你的信任感会超过工具实际的能力。
And then, like, your your your trust goes faster than the the abilities of the tool in reality.
于是，在还没完全摸清特斯拉可能出现错误的所有边界情况之前，你就已经放松了警惕。
So so, like, you're, like, turning off, like, any sense of danger way before, like, you found all the edge cases of where, like, Tesla like, the Tesla's gonna make a mistake.
没错。
Right.
同样地，我们必须向监管机构证明，Heidi的行为是稳定可靠的。
And in the same way, what we have to prove to regulators is that we have consistent behavior of Heidi.
归根结底，我们需要证明的是，Heidi能准确呈现它所检索到的内容。
So so Heidi and what we have to prove ultimately, I think, is that we accurately represent the content that we retrieve.
第一，Heidi不会凭空编造答案，它的回答是有现实依据的。
So one, that Heidi is not like generating novel answers, like it's grounded in reality.
第二，当我们找到相关信息时，能准确无误地呈现出来。
Two, that when we find the the content, we surface it accurately.
所以在Heidi里，每个回答都会明确标注出原始出处的完整引文。
That's why in Heidi, you have like the exact cited quote of where something came from.
你可以通过一个小预览轻松核对，直接看到内容在原文中的位置，从而双重确认答案确实源自那段文本。
And you can easily scrutinize it with like a little preview and you see like where in that text that content is to sort of like double validate that the answer is represented in that content.
而且即使它存在错误率或失误率，也不要求它完美无缺，但必须保持稳定一致。
And then also that if it does have an error rate or mistake rate, not that it it's perfect, but that it it's consistent.
这就是为什么我们自己掌握大量数据和模型、真正作为一个AI公司至关重要——因为每次大厂推出新模型，我们都会紧张，知道之前所有历史模型都会被彻底弃用。
So so that's another area where owning some of like, owning a lot of the stock and models and actually being an ML company is really important because what we've experienced from the hyperscale is is every time there's a new model launch, we at Heidi get get stressed because we know that all of the historical models are gonna get nuked.
它们会被扔到劣质硬件上，做量化压缩，然后被边缘化，因为所有人都在测试新模型，想用最好的表现赢得第一波好评。
Like, they're gonna get put on some shit hardware, quantized, and, like, deprioritized because everyone is testing the new model, and they wanna make it as good as possible to get the best first reaction.
是的。
Right.
我们甚至能明显看到，同样的模型卡片上，内部评估的智能水平都在下滑。
And, literally, like, we see, like, the intelligence, all our internal evals get worse on the same same model card.
而当你做的产品是医疗设备时，这种情况是绝对不能接受的。
And so in the in when you're a medical device, like, that's unacceptable.
你没法想象，做个X光检查，结果每天都不一样。
Like, you can't you can't just imagine you had an x-ray, and, like, depending on the day of the week, it was different.
这种报告你该怎么看？
Like, how would you read report it?
你根本没法信任它，因为你根本不知道它会怎么表现。
Like, you can't trust it because you don't know how it behaves.
所以这就是我们整体的看法。
So that that's the the general perspective.
实际上，全球范围内，我对FDA对某些工具的态度还挺惊讶的。
And actually around the world, I think for a lot of like, I I'm sort of surprised with the FDA's posture on some of these tools.
他们确实非常开放，这也没问题。
They're just like super progressive, which which is fine.
我们也能从中受益。
Like, we get to enjoy enjoy that as well.
但我认为，一旦涉及到临床决策，你就得从硬件开始，把每一步都牢牢掌握。
But I suspect as you go closer to clinical decision making, you wanna earn everything down to the hardware.
我之前说模型在进步，指的是开源模型，你最终拿过来微调，让它适配你的具体需求，然后你得对它的运行结果负责。
And when I say said before about models getting better, I'm talking about open source, and then you ultimately, like, take it, fine tune, make it perform for your use case, and you take responsibility for how it runs.
别想耍花招。
Like, no no cheating.
你不能光说‘哦，是啊’就完事了。
You can't just be like, oh, yeah.
你不能直接调用Anthropic的接口，让它直接开处方。
We just hit Anthropic and it prescribes.
我猜Doctronic就是这么干的。
Like, I'm pretty sure that's what Doctronic does.
他们用的是公开API，这在我眼里简直太离谱了，居然还能被接受。
Like, they're using a public API, which to me is like insane that that is acceptable.
因为我知道，等Opus五代一出来，Opus 4.6立马就显得很蠢了。
Because because I know, like, when Opus five comes out, Opus 4.6 is suddenly dumb.
然后他们就懵了：这怎么回事？
And they're like, what?
我们能接受这样吗？
We okay with that?
所以归根结底，这只是标准不一样罢了。
So anyway, so it's just it's just this a different standard.
你就像一台X光机一样被对待。
Like, you're treated like you're the X-ray machine.
你得每次都输出同样的X光片。
You have to spit out the same X-ray.
你的X光片不用完美，但得一直保持同样的缺陷。
Your X rays don't have to be perfect, but they have to be consistently imperfect.
如果我们有错误率，就会记录下来。
So if we have an error rate, we document it.
如果有安全问题，我们也会上报。
If we have safety incidents, we we report them.
如果出现幻觉或者别的问题，我们会提前告诉医生，让他们心里有数。
We have, like, a hallucination rate or a problem, we share that with doctors and they know ahead of time so they know what they're dealing with.
我觉得，任何做医疗相关产品的人，都得有这个担当，这是我个人的看法。
And I think, like, anyone building in medicine has to have the wherewithal to do that is my, like, hot take.
我不认为那些不愿意做这种艰苦工作的人，有资格去参与医生和患者相关的事情。很多时候，他们都是非临床团队，根本不懂自己推出的产品背后有多大的风险。
I I don't think I don't think people who aren't willing to do that hard work deserve to be doing things with doctors and patients because very often, they're nonclinical teams, and they don't understand the stakes of the tools that they're launching.
如果团队的负责人是医生，而且他们真的懂，那我就放心了。
Like, if if the leader's the doctor and they they get it, then that's fine.
我信任他们。
I trust them.
我知道他们曾经亲眼见过，有人因为一个错误而遭遇了惨痛的后果。
Like, I I know that they've, like, been across from someone who's had some, like, terrible outcome from a mistake.
但如果他们没经历过，我总是很怀疑——他们根本没真正理解这有多重要。
But if they haven't, then I'm always, like, net skeptical of, like, they just don't get it.
他们的视角是患者视角，但医学不是这么运作的。
Like, their perspective is a patient perspective, which is not how medicine is done.
这种视角很重要，但它不等于医学本身。
It's important, but it's not it's not medicine.
是的。
So yeah.
我们之所以要拉上Automedica的人，是因为他们真的花了整整两年时间。
It was important for us to get the Automedica guys on board because they had spent literally, like, two years.
其实，血管外科医生跟我很像。
Actually, vascular surgeons are very similar to me, actually.
哦。
Oh.
纯属巧合。
Complete coincidence.
这一切都是为了确保引用有据可查，尽可能准确。
That all on this, like, deterministic backup to to citations to make sure that it's as accurate as possible.
对。
Right.
不对。
No.
这完全讲得通。
That totally makes sense.
顺着刚才的话题，你提到你们在英国和欧盟都能拿到医疗器械的批准。
I think just to bridge off that, you mentioned your this is to sort of where you get medical device approval in UK and and EU as well.
不过你们现在的业务已经覆盖了一百多个国家。
But you're operating across, I over a 100 countries now.
如果你们网站信息是最新的，应该是116个国家。
I think it's a 116 countries if your website's up to date.
你们现在面临的最大挑战是什么？是拿到某些国家的医疗器械审批和合法许可，还是在这么多不同的市场快速扩张时遇到的其他难题？
How do you is is the hardest challenge you're facing right now sort of getting that med that legal approval, the medical device approval in these certain countries, or what's the biggest challenge you're facing when you're expanding at such a rapid scale across all these different environments?
我觉得最大的挑战其实跟其他企业差不多，主要是团队规模的扩大，以及确保团队高效运作。
I think the biggest challenge is actually mostly mostly just the same thing as any other business, you know, like scaling up team, making sure they're effective.
另外在产品方面，需要在不同国家之间找到平衡点，好在临床方面的互动模式各国都比较相似，这点挺幸运的。
Also, I'd say on the product side, threading the needle well across different countries and making sure I mean, lucky for us, clinical interactions are fairly similar.
不过各国的收入体系和编码规则差异很大，全球范围内还有很多细节需要我们逐一克服。
There's very different revenue and coding schemes, and there's, like, you know, there's totally different a lot of differences we have to navigate around the world.
但说到底，你在意大利、北卡罗来纳，或者我现在所在的墨尔本摔伤了腿，实际处理方式差别其实不大。
But but practically, what happens when, you know, you hurt your leg in Italy or in North Carolina or in Melbourne where I am today, it doesn't really change a hell of a lot.
你会见到医生，可能先在急诊做个初步评估，然后被转给另一位医生，最终解决你的问题，比如做射频治疗之类的。
You know, you've got a clinician, maybe in an emergency department doing an initial assessment, and then referring you to another clinician who ultimately, like, fixes fixes the impasse or takes you for an RF or something.
所以啊，说实话，我也说不准。
So see, I I I don't know.
这真不是监管方面的问题。
I it's not like, it's definitely not regulatory.
也没啥特别头疼的痛点。
There's not, like, any specific pain.
就是做生意本身不容易——找到顶尖的人才，给他们搭建起能让他们发挥的体系、环境和文化，还要对得起客户的期待，真正 obsess 于改善他们的生活。
It's just it's just, like, building business is hard, finding finding incredible people, creating structure and environment and culture for them to succeed, and then living up to expectations with our customers and making sure that we we're just, like, obsessed with making their lives better.
这些天我最关心的，其实就是这些事。
I think that's the thing that I'm most mostly thinking about these days.
哦，这我懂。
Oh, that makes sense.
组建一个优秀的团队真的很难，要找到那些在文化、思维和使命上都跟你契合的人，太不容易了。
Like, in building a good team is very difficult to do, and it's hard to find the right people that are cultural fit, intellectual fit, sort of have the same mission orientation as you.
这我懂。
So that makes sense.
去年十月我们聊的时候，你提到在各个运营国家都设立了类似Revolut那样的半自主团队，能根据当地情况灵活做决策。
And last October when we talked, you mentioned that you were having sort of semi autonomous Revolut style teams in all these different countries you're operating in where they can act semi autonomously or sort of make their own decisions depending on what's the best, you know, decision for that specific environment they're in.
现在还是坚持这个理念吗？
Is that still the philosophy you're taking?
我想知道，这种自主权到底有多大？
And I'm curious how much sort of how far does that autonomy go?
每个团队能跟主流做法偏离多少？
How much can each individual team differ from the mainstream or from your typical path?
是的。
Yeah.
没错。
Yeah.
我们目前还是坚持这种方式。
Definitely still the way we're working.
比如证据传播团队和今年我们要发布的所有内容，基本都来自这些独立团队。
Like, all of Evidence Comms and all the things we'll announce over this year, mostly coming out of these individual teams.
没错。
Yeah.
我觉得是这样，每个团队都对自己的某些指标或成果负责。
I think so so the it's like they have ownership over some metrics or outcome that they're shooting towards.
同时，我们也需要一些统一的支持和必要的约束。
And then you want, like, central support and constraints on the things that are required.
举个例子，合规和医疗设备相关的监管事务。
So a good example is compliance and medical device, like, the regulatory side.
你肯定不希望每个团队都自己搞一套监管对接机制。
You know, that's you don't want every team setting up their own regulatory engagement framework.
所以这类事务就由总部统一支持，比如我们在推进证据收集时，每周都会对齐一次。
So so that's, like, an example of a central function that supports you know, as we build evidence, we catch up every week.
我们会跟他们保持沟通，提前打个招呼，让他们提前开始规划，尽管他们严格来说并不属于我们的团队。
We check-in with them, give them a heads up, and they start planning for us even though we're like, they're not in the team, so to speak.
他们更像是在做内部咨询的工作。
They're, like, acting as almost like internal consulting.
营销团队也是类似，比如全年有十场活动安排，但他们事先并不知道我们会推出哪些产品。
Similar with marketing, you know, there may be, like, a event schedule and you have 10 events that you're planning for the year, and they don't know ahead of time what products we're gonna launch.
所以这种张力——其实所有业务机制本质上都涉及某种对立力量的平衡。
And so the tension, like, all business or, like basically, like, all mechanics deal with, like, forces in opposition at in some way.
因此，构建文化的关键就在于，你把这种对立放在哪里。
So when you're constructing culture, it's about, like, where do you put the opposition?
在我们这里，希望这种建设性的张力是这样的：区域团队会……抱歉，让我重说一下。
So so in our case, the the, like, constructive hopefully, constructive tension is the the regional team will have like, sorry.
中央营销团队会制定活动日程，并据此安排自己的工作节奏。
The central marketing team will have an event schedule, and they'll be planning their their lives.
而各个垂直团队则清楚，他们需要在自己的领域达成某种收入增长或用户采纳的指标。
The individual vertical teams will know that they have, like, a revenue adoption, adoption, some metric that they're gunning for in their their area.
所以就得靠这个团队去推动，激励中央团队把他们的产品放进所有活动里。
And so it's up to that team to to galvanize and and basically motivate the central team to put their product in all the events.
那为什么活动团队要花那么多钱支持他们呢？
And, like, why should the events team, like, spend all the money on them?
他们知道Scribe肯定能行。
They know scribe's gonna work.
他们有个共同目标，但人总是更愿意维持现状。
They have some central goal that they're shooting towards, and they it's always easier to stick with the status quo.
所以你需要这些垂直团队里有像创始人一样的人，愿意碰钉子，不达目的绝不罢休。
And that's why you need, like, those individual teams to be led by almost founders within the business that are willing to step on toes and just, like, achieve their goals come hell or high water.
他们要么就为自己的目标拼到最后一刻。
Like, they're they're either they're just gonna, like, die on their own sword.
他们会不计一切代价，确保自己负责的领域取得成功。
Like, they will do anything to make sure that their their their area succeeds.
合规部门则是个硬性关口，有权否决他们。
With compliance is like a hard break that they get that can veto them.
安全始终是我们优先考虑的事情，远超过任何业务目标。
And safety is always something that we prioritize over, like, any of our business objectives.
对于中央团队，我们还要设立激励机制，让他们也愿意推出新产品。
And then with the central team, creating incentives for them to also want to launch new products.
总的来说，我们的文化非常注重创新，大家总是想不断推出新东西。
And I think generally, the culture is, like, super excite innovative, and, like, we always wanna launch new things.
虽然我看起来像是在描述一场内部斗争，但其实并不是这样。
So although I'm presenting it as, like, this, like, internal fight, it's not really.
不过，也有可能变成这样。
But it can be.
如果你有三个产品都争着要曝光，该怎么决定哪个优先、哪个往后排呢？
Like, if you have three products and they all want airtime, like, how do you prioritize which gets in and which doesn't?
这就得靠领导者来证明进展，就像在公司内部做资源分配一样。
So it's kind of up to the leaders to proving traction, almost like internal capital allocation.
我觉得自己还在边做边学，但这个模式确实挺有效的。
And I think and I'm still learning by doing it, but I think it's an effective model.
要在现有团队里营造出推动新产品所需的紧迫感和创新氛围，真的很难。
It's very hard to create the level of urgency and innovation you wanna see for new products in the existing team.
因为现有团队要维持数百万次会话、数十万用户的稳定运行。
Because the existing team is supporting millions of sessions, hundreds of thousands of users.
这完全是两码事。
Like, it's like a different thing.
他们的任务是确保系统永不宕机，而不是从零到一重新起步。
Their job is to never go down, which is different to, like, get zero to one again.
是的。
Yeah.
我觉得我们做得还不错。
So I I think we're doing a pretty good job.
你看，我们已经推出了两款新产品。
Like, you know, we got two new products.
才过了六个月而已。
It's been six months.
还不错。
Not too bad.
我们可以年底再回顾一下，看看进展如何。
We can you can check-in at the end of the year, see how we've gone.
目标是一致的，但每个团队成员都会根据自己的团队设定具体的工作目标。
That's the same North Star, but everyone has their own individual goals they're working towards depending on the team, essentially.
这说得通。
That makes sense.
没错。
Exactly.
而且，我们确实都在朝着同样的目标努力。
And, like, yeah, we all we all share the same goals.
我们公司文化很透明。
Like, we're pretty transparent as a culture.
我们会主动和团队分享所有正在发生的事情。
We share everything that's going on, like, very actively with our team.
所以今年我们的目标，说白了就是实现这个AI护理伙伴的愿景。
And And so, like, our, you know, our goals this year is all about, like, basically fulfilling that AI care partner vision.
我们已经赢得了和合作伙伴深入合作的权利，而且临床医生们都很喜欢Heidi，这很棒。
You know, we've we've earned the right to to do more with our partners, and and clinicians love Heidi, and that's great.
现在我们已经把它们上线了。
And and so now it's about like, we've launched them.
很好。
Great.
接下来我们要努力让这些新产品也获得同样的喜爱和认可。
Now we have to try to get to the same level of love and adoration as described with these new products.
这说得通。
That makes sense.
你今天已经提过好几次了。
You mentioned this a few times today.
上次我们聊的时候，你也说过，你长期的愿景是让Heidi不再只是一个工具，而更像一位可以信赖的同事。
You mentioned it last time we spoke as well that your long term vision is for Heidi to not really be a tool, but feel more like a colleague or sort of someone you can rely on.
过去六个月里，你觉得这个目标是变得更近了，还是因为新学到的东西反而觉得更远了？
In the past six months, has that goal seemed has does it feel like it's more reachable now, or does it feel like it's farther away based on what you've learned in the past six months?
是的。
Yeah.
完全没错。
Totally.
我觉得，看到Evidence现在能做到的这些，我们离那个愿景越来越近了。
Like, I think when I when I saw seeing what Evidence can do now, I think it's like we're getting closer and closer to that vision coming to fruition.
是的。
Yeah.
我不能透露我们即将发布的所有内容，但简单说，这些功能并不独特。
I I can't I can't spill the beans on everything we're gonna launch, but but, like, basically, it's it's not unique.
我之前跟人说过，我在澳大利亚的夏天经历了一段特别忙碌的时期。
So, you know, anyone who's I I was telling people I had, like, a hot cursor summer over the Australian summer.
所以，相当于在美国这边，我们经历了一个特别繁忙的冬天。
So, like, you know, a hot cursor winter in in The US.
所以，当Opus 4.5发布的时候，简直是个转折点。
So, basically, like, when Opus 4.5 came out, it was just like a watershed moment.
就像当智能代理终于能正常工作了，大家才意识到，这事儿真能做成。
It's like when when, like, when agents, like, finally worked, you know, the the idea was, like, possible.
你真的可以做出东西来。
You could build things.
它能自己把事情做完。
It could follow through and complete work.
你甚至能让它帮你填Excel表格。
You could do Excel sheets.
但我还是觉得，在医疗行业，普通人家根本不会天天盯着这些事，他们根本不在乎。
And I still think in health care, like, the average person who isn't, like, obsessing over this every day is just not they don't care.
他们只管看病人。
Like, they're not they're seeing their patients.
看完就回家了。
They're going home.
他们可能试过ChatGPT，或者用Gemini生成过图片，用Grok做过视频。
They maybe have tried chat GPT, maybe made an image on Gemini or video on Grok.
就是随便玩了玩，然后觉得，哦，这还行。
Like, they've they've, like, dabbled and they're like, oh, yeah.
挺有意思的。
This is cool.
我之前真没觉得这有什么大不了的。
Like, I didn't get what the fuss was about.
对我来说，Evidence是第一次让我忍不住说：天哪。
I think Evidence for me is the first time when I was like, holy shit.
这简直能省下我一整个周六。
Like, this could this could have saved me a whole Saturday.
做外科审计的时候，我每周至少得花一天周末，和别的住院医生一起泡在那儿。
Like, with surgical audits, I would literally spend sometimes both days in the weekend, but at least one day on the weekend every week with all the other residents.
还得一个个整理幻灯片来做外科审计，真是累人。
Like, man arming these these slides to do our surgical audits.
每次看一个病人、接收一个新入院或转诊，我们都要问：他们的病史是什么？发生了什么？我们做了什么？有没有并发症？有没有再次入院？能从中学到什么？
And it would be like, every patient we saw, every admission, every referral we received, what was their history, what happened to them, what did we do, Were there any, like, complications, any readmissions, anything we can learn from it?
完全是手工操作。
And it was manual.
百分之百靠人工。
Like, 100% manual.
根本没法自动化处理。
They couldn't couldn't do it in an automated way.
用了Evidence之后，只要系统和病历连通，你一句话就能搞定。
With Evidence, you basically one shot it with a single prompt as long as you're integrated with the record.
你只需要说：好吧。
So you can just be like, okay.
比如普通外科，GSU这一周的手术记录，给我整理一份。
For, like, general surgery, like, GSU one, do a surgical order of our week.
再把所有转诊记录的截图、笔记里的内容，直接复制粘贴进去，按回车就行。
Also, like like, upload screenshots of, like, all of the referrals you got, like, notes app, copy paste into context, hit enter.
这得花点时间。
It will take a while.
大概要五分钟，还会消耗不少令牌。
Like, it would take five minutes, consume a bunch of tokens.
但最后它会生成一个表格，真的比我亲手做的还要好。
But then at the end, it will spit out a table that is, like, legitimately better than anything I made.
而且说实话，这事儿挺让人震撼的。
And and and the thing is, like, it's just it's just kind of profound.
要是不用在周六干这活儿，我本可以多做几台手术。
Like, if I didn't have to spend my Saturday doing that, I could have been operating more.
我本可以成为一名更出色的外科医生。
Like, I could have been a better surgeon.
我本可以那天去帮别人做助手。
I could have been assisting someone else on that day.
但我得找一个有外科住院医思维的人，来制作和呈现这些内容。
But I was like, you needed someone with, like, this surgical trainee's mind to create this content to present.
这次审计其实非常有价值，也是一次很好的练习。
And the audit was very valuable and useful practice.
所以也不能说这工作完全没意义。
So so it's not like it wasn't useless work.
正因如此，我有时候不太愿意承认，哎，这个AI助手就只是在搞行政事务。
That's why sometimes I hate to, like, admit, like, this, like, oh, AI scribe is just administration.
但这根本不是行政工作。
Like, it's not administration.
我得真正理解前胫动脉的生理机制，才能搞清楚病人出了什么问题，才能写出有深度的讨论内容，读得懂论文——这些活儿我非得亲自做不可。
Like like, I needed to know that like, I need to know about, like, anterior tibial artery, like, physiology to know, like, why something went wrong for this patient and to be able to create a content for a reasonable discussion and, like, read papers and like, I needed to do the work.
但就是因为这种所谓的行政事务，我反而没法专心做我本来该训练的事。
But but that took me out of of doing what I was training to do just because this is, like, an administrative thing that we do.
说实话，我也说不准。
See, I don't know.
我觉得这跟其他领域的发展轨迹其实是一样的，我可能有点跑题了。
I think I think it's the same trajectory as every other area, so I'm rambling a bit.
但说到底，我觉得这些系统最终会帮我们承担大量工作。
But, basically, I just think that in the end, it's like, these systems are gonna do a lot of work for us.
每个人跟系统的互动方式都会不一样，比如制作审计用的幻灯片这类事。
Every person will have a different version of their interaction, you know, and maybe maybe, like, creating the slides for the audit.
让海蒂去做我很乐意，但我更想专注于学习。
Happy for Heidi to do, but I wanna do the learning.
我希望自己能提炼出临床洞察，让海蒂把病例和典型个案推给我。
So I wanna come up with, like, the the clinical insights, so Heidi can surface their cases to me, the vignettes.
我来负责反思和总结。
I can do the reflective practice.
所以我觉得，其实就是这么回事。
So I think it's just yeah.
但要是只把AI当速记员用，那它就真的只是个事务性工具了。
It's not it's whereas with scribing, it is, like, very transactional if you use just as a scribe.
比如，它只是在生成临床病历。
Like, it's generating clinical notes.
有时候，人们甚至会把海蒂上面的所有内容都删掉。
Sometimes people even, like, just delete everything off Heidi.
它就像个单纯的工具，跟录音笔差不多，我觉得这用起来是挺方便，但并不是最终能长久下去的核心价值。
Like, it's, like, purely, like, a tool, you know, like a voice recorder, basically, which I think is use but it's not the ultimate, like, thing here that's gonna persist.
不会。
No.
完全能理解。
Totally makes sense.
我已经能看出来，你正在一步步打造一个真正的伙伴。
And I can already see how you're sort of working towards building that companion, that partner.
就像你说的，你手头还有很多计划，我很期待看到你接下来推出什么，以及海蒂会怎么继续发展。
And like you said, you have a lot in the pipeline, so I'm looking forward to seeing what you release and how Heidi continues to grow.
快结束之前，你对还没试过海蒂的临床医生有什么想说的吗？
I guess as we wrap up, do you have any last comments or any message for clinicians who have not yet tried Heidi?
是的。
Yeah.
直接用 Evidence 就行。
Just use Evidence.
很简单。
It's easy.
还是免费的。
It's free.
没错。
Yeah.
这根本就是道德上的选择。
It's the it's the moral choice.
就算 Open Evidence 和 Doximity 功能一模一样，我们也不会把你的数据卖给药企。
Like, even if open evidence and Doximity are exactly identical, we're not selling your data to pharma.
这就是你应该用它的原因。
So that's the reason to use it.
而且说实话，我觉得可能是我所知道的唯一一家真正有志于推动医疗服务变革、真正想重塑医疗交付方式的公司。
And and, yeah, I think I think, like, genuinely, like, probably the only company that I know of that's actually, like, has the ambition to go towards care delivery and, like, really meaningfully transform how health care is delivered.
所以，我希望所有临床医生都能给我们反馈，尽管挑刺，告诉我们哪些好、哪些不好。
So, yeah, I want I want all the clinicians giving us feedback, roasting us, you know, telling us what's good, what's bad.
我们对反馈反应非常快。
We're super responsive to it.
所以，别犹豫，直接告诉我们你的想法，希望我们能配得上大家的期待。
So, yeah, just give us your feedback, and and hopefully, we live up to it.
听你这么说真好。
It's great to hear.
非常感谢你，汤姆。
Thank you so much, Tom.
很高兴你能来。
It's been great having you.
谢谢你，马纳夫。
Thanks, Manav.
关于 Bayt 播客
Bayt 提供中文+原文双语音频和字幕，帮助你打破语言障碍，轻松听懂全球优质播客。