Title: AI Arrival: From Technological Frenzy to Humanistic Awake...
URL Source: https://www.bestblogs.dev/podcast/fc3b1d6?amp%3Butm_medium=feed&%3Butm_campaign=resources&%3Bentry=rss_article_item
Published Time: 2026-07-05 08:00:00
Markdown Content: PODCAST
88
Zack Kass, former head of commercial operations at OpenAI, deeply analyzes the AI industry's "adoption gap" and "automation boundary," and explores the reshaping of human identity and the awakening of humanistic values in the midst of technological frenzy. 第 第一财经
Today 20:36 19 chapters View Source →
!Image 1: AI Arrival: From Technological Frenzy to Humanistic Awakening | Two Talks
00:00 00:00
00:02 1
I have made a case for a long time that there is no AI bubble.
00:06 2
but if I still to force you to choose， where do you see the risk as being more likely to be concentrated? How big is the gap? And is this shrinking or is wide new?
00:17 1
It's going to take generations to solve this.
00:20 2
If you could automate everything in your life， where would you .
00:24 1
stop and what you do with IT individually? What we do with IT collectively .
00:29 2
will define our outcomes. Are that thank you very much joining our program and .
00:34 1
welcome you to china.
00:36 2
So let's art with AA broader question， general observation of the AI industry today. Where are we? right?
00:44 1
We are building machines that possess human intellectual equivalence and now superiority. So we are building machines that are capable of the human brains capacity. And then some， we're also actively expLoring machines that can do physically I and machines that can understand words we call these world models. And this will extend the modality from language to robotics to speech to vision and and beyond.
01:17 2
And what would you say are the most of critical industrial characteristics and trans at this point?
01:23 1
By far， the most interesting trend continues to be the decline and inference costume. It's it's quite remarkable and especially being in china， it's it's it's worth talking about the deep seek moment. As we talk about in in the U.
01:37 1
S， we called the deep seek moment， was a wakeup call to the frontier labs that the frontier technology state of the art technology did not have to be prohibitively expensive， and people are discovering that the agents or the influence itself doesn't have to solve the problem. We can actually build declaration technology that is capable of doing software basically that's capable of doing really simple tasks that we previously been giving the agents. But my favor trend is even more impressed to me than how good the models are getting is how quickly they are compressing and open sourcing.
02:16 2
In the meantime， after the early excitement of this technology， the breakthrough， uh， people in the financial world is more and more seeking the sustainability of this， profitability of this. AI in the street.
02:29 1
Well， okay， I have made a case for a long time that there is no AI bubble. What I would say is this， there is certainly going to be companies that don't do well. There are certainly going to be investments that looks silly.
02:43 1
There is infrastructure that may not cover its own costs. But if you are in the arena， building valuable technology or deploying that technology effectively， you will do well on average， and the average person will benefit in the end because of the investments we're making an AI. And so I think the real question is， do we have enough infrastructure to to to manage the demand? That answer is probably no， and that introduces opportunity for other deep seek moments.
03:16 1
And if you're in an investor picking a specific companies probably dangerous. Picking an index of companies is safe. But I think more importantly， make sure that your beats are protected against breakthrough in things .
03:31 2
like architecture. Competition made me we're going to burn cash in the foreseen able future， so which to some extent capture the AI bubble. If you even you reluctant to call IT a life from a lot of investors thinking OK， maybe it's .
03:47 1
important thing about the vectors on which AI companies compete because that that helps explain where the money is going. So the first vector on which in a company， a few of them， is the frontier model. Now that is probably the most exciting and also most expensive of all of the competition and the value of the moonshot of winning the AG.
04:14 1
I race is so great that companies will just pour money into this if you remove that vector. Suddenly， these frontier lab companies， these model manufacturing companies， become really attractive on a investment basis because they actually spend very little relatives how much they make if they stop spending on training the frontier. The second thing that they compete on is distributing inference.
04:36 1
They buy the G， P， S. They build these data centers， and then they sell you the tokens. That model IT has its own risks， one of them obviously being that we're running out of compute everywhere and that you it's specular game at some point.
04:50 1
But the other risk is， is pretty obvious， which is， well， if demand for inference does not keep climbing that a bunch of these company to be holding a bag， they're going an enormous a computer. They can't. They can't live back.
05:03 1
The third is the application. This to me is is probably the most interesting of the competition layers. Because you in with opening eye， you have chat BT anthropic， you have clod google as german I etta.
05:18 1
And this has long been seen as this other game， but IT might actually be the dog game like IT might be doubted. The companies are ultimately measured because it's how you can sell the add， it's how you can capture the data， is how you can retain the customer. IT might also be the most effective way for most companies to actually use this technology. And I think what we forget is that companies don't have to win on all the basis. They can win on one of the basis.
05:50 2
But if I still need， I would have to， to force you to choose， where do you see the risk of as being more likely to be concentrated?
05:59 1
The frontier and IT turns out of a theory that proposed late last year called diminishing model returns， theory that says， at some point， you and I don't care about the next model， and at some point the enterprise doesn't care about the next model or on average. And that's the point at which the fast follow the open source version， the compressed version becomes economically viable and then IT becomes economically interesting because it's much less expensive. And if if that proves valuable， if that proves viable， then the company is investing in the frontier will look like they have wasted an enormous amount of money training the models for everyone else to actually capitalize on.
06:39 2
I also know that in your work， the next renaissance， you mentioned the adoption gap， right? So the .
06:45 1
adoption gap basically proposes the space between what technology is capable of and what we what we use IT for. And there are a few reasons for any given adoption gap. One of is um a lot of people are just not give are not aware of what the models are capable.
07:05 1
There is just a poor understanding of the actual technological capability. The second reason that that plagues a lot of companies is there is no vision. There is no actual the goal post have moved or disappeared so far.
07:21 1
That is actually quite hard for companies to figure out what they are aspiring to. The third reason is the saddest， which is it's actually so hard to build this stuff. And if you need people to actually deploy this technology and we don't have a lot of those people.
07:37 2
how big is the gap? And is this shrinking or is widening?
07:40 1
It's widening. It's widening in the U. S. Well， first， first， what technology can do is expanding. But the bigger issue is we have policy that， that restricts what we are capable of doing with some technologies， and we don't have companies capable of actually introducing the technology in their business. So there's just this bottle neck at the point of adoption that um yes， that that is really limiting in a company's ability to take advantage of the latest.
08:13 2
greatest how company can solve this question or company alone can solve this.
08:17 1
Well， I spent most of my time on this like I spent most of my time trying to figure how to solve the adoption gap specifically for mid market companies and family businesses.
08:27 1
Family on businesses have interest in legacy and min market companies too， and there is a profound importance in making sure that the average company owned by， owned by a family， owned by a husband and wife， owned by someone oil， by a group of people for a long period time， that these people do well with this technology. I have comes to believe that AI is most beautifully adopted by non economic actors. At what I mean by this is， if everyone who who to make more money principally adopts AI we will build a more economically efficient world. But that may not be a more spiritually rich world. That may not be a more culturally rich world.
09:10 2
So for companies， how can they just try to help the a lot of the employees to um walk at this gap?
09:19 1
Well， okay. Or a lot of the .
09:21 2
times is psychological pressure or against that， I have one piece of .
09:25 1
advice if you are a business leader and maybe I have two piece of advice. My first use of advice is pick a castle on the hill， pick a place in the distance that you can inspire your people to work towards. But the second problem， I would tell ceos， if you can tell your employees a story about the ways that this technology is going to improve the business， to build a Better world， then you shouldn't expect anyone to care much.
09:52 1
And so many companies have failed to actually chAllenge their employees to think bigger. IT is very hard for the average employee to be inspired or to know why they should materially change the behavior. If the tool doesn't call to you， IT means that you're probably fine or IT means you haven't chAllen yourself to think like i'm not convinced that AI needs to be used by everyone all the time.
10:16 1
And in fact， most people should probably use technology less to find more happiness in this world. We watched social media hollow out an entire generation. We watched the effects of screen addiction. We watched the effects of abundant technology without regard for what would happen to children. And we now are watching a lot of parents ask a lot more questions than that they wish they had asked.
10:43 2
Parents are worried um but a lot of the college graduates are actually fearful because AAI is taking over so many entry level of jobs.
10:52 1
I think the kids are terrified because they have been over exposed to negative information and other exposed to positive information. But the current trends don't suggest that AI is going to destroy the labor economy. IT suggested that is gona change work pretty materially.
11:11 1
But unemployment has is actually down over the last six months um and entire level work is experiencing a growth in new jobs， jobs that we know didn't describe six months ago. And so， so much of this is just baked into a culture. You a lot of people are pretty attached right out of their professional identity. And I think the great threat that AI presents people is not an economic one， is an emotional one. It's the idea that sure， i'll have work， sure， i'll be able to put food on the table， but will I be able to confidently say what I am and who I am?
11:47 2
I you tell me about your professional.
11:50 1
that is your identity and the as AI fractures s career letters， even if IT produces more working. Even if the work is safer， even if the work pays more， and all these things are probably going to be true， people will really struggle， I think， to figure out who they are when they can no longer attach their purpose to a career letter. I don't have an answer for this. I think we should acknowledged that we are on earth for much more than just work， that there is so much more purpose and dignity and family and friends. But it's going to take generations to solve this.
12:23 2
Which specific sectors are， do you think face the toughest st of time raining this adoption gap?
12:28 1
Education number one， imagine your teacher， you in the classroom， a kid can pull out their phone and check everything you're saying or .
12:35 2
actually have something smarter than their teacher .
12:38 1
or just going to learn on their own. And the purpose of the educator becomes reduced. library.
12:42 1
And here's a book， here's a phrase， instead of the coach of the inspire that we want， the teachers want to be in， the children need. That's not a technological problem. Maybe it's a policy problem.
12:53 1
It's a leadership problem. It's a vision problem. IT has to be someone or group people taking up the mental and saying， this is what I can look like. The other industry that I think is， uh， traditional health care.
13:05 1
I think health care is going to do really well， but I think owning a hospital is actually only the box itself is pretty risky because the box， meaning the physical hospital， the place where where health care is， is administered. The real way that you build Better health care is not by investing more technology into the box. It's by pulling technology out of IT.
13:28 1
Driving down the cost of health care means removing the ancelles elements， the most critical elements of primary care of trios of imaging， of pathology and administering them in a much lower cost environment in on your phone， yes， in a clinic， all the spaces that um that are less expensive to run in the hospital with the actual risk to the hospital itself is pretty great. As things become this immediately， it's a result of the design flaw. We need to figure out a way to actually deliver their their service.
14:02 2
What do you see as the core chAllenge right now? At this point?
14:06 1
One of the major constraints to adoption is the number of smart people that can deploy this technology. So II tell people we're so early in in， in these models， every individual sees the tool as a thing they need to use， but actually slowly and then suddenly， IT will become embedded infrastructure. We're going to see the technology moved to the background.
14:29 1
And most of IT comes down to training and enabling engineers to deploy into these businesses and actually help them build infrastructure. And one of the reasons why I think there will be a lot more jobs than people think， I think I actually create a job boom because I think we're going to discover how much work needs to needs to go into actually delivering this technology. And most companies think what they need to do is get their employees using the tools.
14:56 1
They talk about adoption， but adoption is a red hiring. What you should that is business outcome and value. By making technology invisible， the best companies are going to figure how to make AII feel like infrastructure so that their employees don't have to constantly click a button.
15:12 2
IT happens. IT happens around them .
15:14 1
more naturally， invisibly. You do not think about all the ways in which electricity powers your life. You do not think about all the ways in which the internet powers your life.
15:24 2
So what brings you to the students of pick university this time around?
15:29 1
Uh， well， I was very grateful to be invited to talk about research of them doing for my new book， which is called the automation bounty. And the automation boundary asks a very simple question， which is， if you could automate everything in your life， where would you stop? And it's an interesting question asked， because machines are getting really good and really cheap. And increasingly， we have to figure out not just what we automate， but what we don't and where we should draw the line.
15:58 2
You talk about the a boundaries and a very interesting question. I would like to throw back this to you， uh， if you could automate everything in your life.
16:09 1
where would you stop? I would stop much further than I think that I would love， love to make sure that I never automate. I love exercise.
16:18 1
I love playing sports with friends. I love time with friends. I like the long drive， and I love， love being in a car and looking out on the ocean.
16:31 1
And so I was that I would do way more if I automated everything else. So I think the automation boundary is that we are going to toe IT on both sides， good and bad. We're going to get IT wrong multiple times like we've already got in wrong with tone.
16:44 1
That's an automation boundary that we should have crossed a long time ago. On the other hand， we are probably gonna to ate things that we shouldn't have. We automated daycare for kids.
16:55 1
We gave every kid and ipad or a screen. We shall have done that. So we are going to constantly figure out the things that we shouldn't， shouldn't automate.
17:04 2
and we're going to walk them back and forth. But you mentioned an autonomists driving and also， uh， industrial AI and AI in government services. We human are actually almost like near zero tolerance of any area in a lot of areas I mentioned before.
17:19 2
So there is the natural anxiety of losing control here. So how company if we are approaching that automation， barry， how company um should a built this trust with users will cross the gap. The reason that .
17:33 1
automation boundary is going to be emotionally chAllenging for people as societal thresholds. This idea of what do we want machines to actually do， and that is， you point IT out.
17:45 2
we hate watching .
17:47 1
machines make mistakes. But IT presents this funny world where everyone goes. Well， how accurate the machine.
17:54 1
I say， well， okay， look， you're going to go to， you're going to get the cornea surgery. And the machine is ninety nine point nine percent reliable. IT doesn't mistakes.
18:03 1
Sometimes you， I don't know， I don't know， I trust that I know the human is eighty two percent reliable. Most people still pick the human. We see this all the time. But a lot of IT is actually just going to cause this very strange a sort of stop and start as we explore what IT is we want to automate.
18:22 2
So what role should government play in this scenario? In this process?
18:28 1
I think one job is to pass policy to manage the model manufacturers. And anyone building machines on three basis alignment is the machine does the machine care about， is the consequences of its actions to explain inability， can the machine explain itself? And three model behavior， what is the behavior of the model? Is IT wired to be sick.
18:53 1
Ohana is IT wired to create addiction. The other thing government should do is past very punitive policy for people who use technology to do bad things， because they are far more capable now than ever before. But the last thing governments need to do， and I， I just cannot stress this enough， governments have to pass policy that defuses the value of the technology to the benefit of the average person.
19:18 2
You mentioned A， A term that a metered intelligence that is going to rain about with this. AI， so what kind of the promising future IT can give us? I wrote a paper .
19:28 1
in twenty twenty one titled unmetered intelligence in which I argued that， uh， intelligence was a resource and that we were beginning a trend where all， like all resources IT would decline in cost and and expand in abundance or or availability and that at some point in the future we would observe a world where everyone had access to cognition， unlimited cognition. I am not arguing that everyone would be smart.
19:56 1
Unfortunately， unmade an intelligence doesn't mean that you are brilliant to have access to brilliance. And what you do with IT individually， what we do with IT collectively， will define our outcomes. And I frame that now as I talk about what AI ultimately means， it's not a tool， is cognition at scale.
20:14 1
And how we leverage IT for good， how we protect our humanity， how we discover what IT is we desperately want to keep， are real. What IT is that we wanted give up. This is how we will rediscover the human experience.
20:29 2
Thank you that really enjoying this .
20:32 1
one conversation.
20:33 2
Thank you.
Click any line to jump · Hover to ask AI
Content Overview
演讲者概述了 AI 行业的发展现状与未来趋势，强调技术广泛应用、成本下降及模型开源化等积极变化，同时指出潜在风险，如部分公司盈利难题、基础设施不足等。教育、医疗领域的应用挑战及社会对自动化技术的接受度和信任建立成为焦点。AI 将重塑就业市场，创造新机会而非导致大规模失业。提出‘自动化边界’概念，主张在享受 AI 便利的同时，避免过度自动化，维护人类生活质量与价值。演讲者乐观展望 AI 前景，同时呼吁关注并解决伴随的社会经济问题，包括技术潜力与实际应用间的‘采用差距’，以及如何通过政策和领导力创新，确保技术公正分配和伦理使用，最终实现 AI 提升人类生活的同时保护人性和发现真正珍视的价值。
#### Speaker Summaries
发言人2
深刻剖析了 AI 行业的现状与未来，聚焦技术与社会间的“采用鸿沟”，指出风险集中点，深入探讨自动化对就业的影响。他强调，政府和企业需在推动 AI 发展的同时，构建公众信任，应对教育、就业市场对 AI 技术的恐惧。他提出，企业应助力员工适应技术变革，政府在转型中扮演关键角色，展望 AI 带来的积极影响，倡导社会共克时艰，实现技术与社会的和谐共生。
发言人1
长期主张人工智能领域不存在泡沫，认为构建具有人类智力甚至超越人类的机器是一项需几代人努力的事业。他指出，AI 技术的进步将从语言扩展到机器人、语音和视觉等多个领域，随着技术成本的降低，AI 将更广泛地融入社会。尽管存在投资失误和基础设施不足的风险，但有效利用 AI 技术的公司将从中受益，社会整体将因 AI 投资而获益。他强调，AI 的应用不仅限于经济领域，还应关注其对文化、精神和人类身份的影响。他提出，随着 AI 技术的普及，教育和医疗等行业将发生深刻变革，政府应制定政策引导技术的合理使用，确保技术进步惠及大众。最终，他呼吁社会共同探索如何在自动化与保持人性之间找到平衡，以期通过 AI 技术提升人类生活质量，同时保护和弘扬人类的独特价值。
AI frontier model adoption gap automation boundary cognition at scale diminishing model returns human intellectual equivalence infrastructure model manufacturing open sourcing technological capability unmetered intelligence AI bubble investment psychological pressure
Content Overview
中文指路→AI降临：从技术狂飙到人文觉醒丨两说
当前，人工智能行业的发展来到了一个关键路口：一边，模型能力持续突破、推理成本大幅下降、多模态与世界模型加速探索；另一边，公众对于智能时代的就业、身份和人机关系的担忧也在同步升温。
从模型竞争到落地应用，我们在何地？从“技术能做什么”到“用技术实际做什么”，我们如何走？当人工智能成为基础设施的“无限量智能”时代正在到来，我们去何处？在技术与社会的交汇处，如何定义AI的边界？如何守护“人之为人”价值？
本期节目，第一财经记者葛唯尔对话OpenAI前全球商业化负责人、人工智能与商业战略专家扎克・卡斯（Zack Kass）。结合他深耕AI商业化一线实战经验与行业洞察，共同拆解人工智能行业的机遇、难点与未来走向。
本期嘉宾：
扎克・卡斯 | OpenAI前全球商业化负责人、人工智能与商业战略专家
本期主持：
葛唯尔 | 第一财经记者
你将听到以下内容： 00:46 人工智能行业正处于什么阶段？ 02:53 人工智能存在全局性泡沫吗？ 06:47 什么是“采用鸿沟” 11:01 人工智能并不会摧毁就业市场 12:32 哪些行业在跨越AI采用鸿沟时最吃力？ 17:35 自动化的边界到底在哪里？
播客编辑：阿贵
监制：yoiyo
“一财，两说，三思笃行”，全新财经洞察节目《两说》，通过第一财经与顶级专业人士的共创，以专业、深度、独到见解为标尺，通过巅峰对话的表达样态，发现、探讨与深入思考经济领域的前沿热点问题。
Make your daily reading actually fit you.A daily brief built from the sources you follow. Get started free
00:00 / 00:00