“接近”通用人工智能的商业：虚假宣传如何变得有利可图

This year, Mr. Alman said that AGI was imminent.

Now, he says that the term AGI itself is not useful.

Every few months, someone in big tech declares that we are on the edge of artificial general intelligence.

And every time that line gets blurred, a few more billions get raised.

I have never talked about AGI on my channel, but lately it keeps showing up in the comments.

So, I thought, wouldn't it be symbolic to share my thoughts and research on the theme of AGI?

Especially now that we're approaching the end of the year and I'm wrapping up my series on the business economics of AI.

I will now be dissecting the technical feasibility of AGI and when or whether it'll see the light of day.

This is completely outside of my line of work. I am not an AI researcher.

I would like to dig into the business of pretending that we're close to AGI and the economics of this narrative.

Let's dive in.

It is considered that there are three dimensions so to say in the realm of artificial intelligence AI, AGI and ASI.

AI stands for artificial narrow intelligence.

The purpose of the A and I is to perform a single specific task extremely well like the way Claude is generating text or Siri recognizing voice to a very certain extent or your phone recognizing your face.

This is the only type of AI that currently exists in 2025.

AGI stands for artificial general intelligence.

AGI is a theoretical term. The operative word here being theoretical.

AGI would supposedly match human level intelligence in the ability to learn, reason or transfer knowledge and ideally solve novel problems without having to be retrained.

The closest example of Agi that you might have seen is a human resembling robot in the movies.

And ASI which stands for artificial super intelligence is a purely hypothetical concept of AI that would surpass human intelligence across all domains.

ASI would hypothetically be able to solve problems and learn far faster than any human alive and have the ability to improve itself autonomously which is essentially technological singularity if I may.

Let's come back to the topic of this conversation which is AGI.

The problem of AGI is that very few people can define what it actually means.

Even prominent AI researchers and scientists cannot land on a single definition of this thing.

But ironically, there are quite a few people in the general public, so to say, who would gladly tell you the definition of AGI, thinking that they're well aware of what it is and what it's supposed to do.

Now, if you search for the definition of AGI, the word that you're going to come across is going to be consciousness.

Something like AI with consciousness or conscious AI.

So, some kind of magical AI entity with cognition.

But the question that isn't really being answered is what is consciousness?

The paradox is that the people who do PhD level research when talking about AGI do so with extreme caution without any specific timelines and they're doing so exactly because the definition of AGI is extremely vague and hardly quantifiable.

But those who do speak freely and confidently about a GI usually cannot properly explain what it means and what consciousness means in the context of artificial intelligence.

Nevertheless, the noise that they create is loud enough to induce anxiety into the part of the population who mistakes confidence for credibility.

Nevertheless, the noise that they create is loud enough to induce anxiety into the part of the population who mistakes confidence for credibility.

You guys know that I work in product and a regular discipline that I do is competitive research.

I have to look through competitor product pages, pricing tables, LinkedIn announcements, summarize it all, put it in an email, send it to myself and to my colleagues.

And I automate a lot of things for my business outside of work using N8N, which basically connects APIs, apps, and models into one system that runs automations end to end.

And if you automate as much as I do, your workflows get really expensive really fast because NA10 needs to be hosted and I need to pay for API credits for multiple apps that I have nodes for.

Every automation has a cost and it's fine when it's just one workflow, but what if you need 50?

The easiest and the most efficient way to do this is to self-host in the cloud via virtual private server.

And Hostinger offers one of the best self-hosted NA10 plans on the market.

And that plan can get you up and running in literally minutes.

With Hostinger, you can host unlimited workflows, unlimited parallel executions, and plenty of readytouse N8N templates.

And all of that at a fraction of a cost and with fantastic performance because you can choose what type of server you need depending on your needs.

All you got to do is go to hosting your website, choose your plan.

KVM2 is actually going to cover the vast majority of your needs.

It gives you everything you need to get started with a small number of workflows and the ability to scale them indefinitely.

They're also having a Black Friday sale right now which gives you 60% off.

But even without the discount, Hostinger's plan allows you to use N8N itself for free and pay only for VPS hosting.

This first of all automatically saves up to four times your automation expenses.

And secondly, you get your own server.

Go to the check out page, pick the server location that is closest to you and the application you want to install.

Plug in your coupon code to get a discount and you're good to go.

And then once you're in the app, simply click on manage app and log into your NA10 account directly.

If you're a content creator, a founder, a business owner, large or small, and you're serious about scaling automations or just want to experiment without burning your budget, this setup gives you the best quality for your money.

You don't have to worry about security and privacy because everything you do stays private.

And don't forget self-hosted NAD is often required for businesses that must follow privacy rules like GDPR or SOCK 2.

You can check it out at hostinger.com/tech10 and use the code tech10 at checkout to get extra discount on any yearly VPS plans.

It's a limited time deal, so don't miss it.

And huge thanks to Hostinger for sponsoring this part of the video.

Here's a chart of predictions from the most prominent names in AI research who are way less optimistic or certain about the timeline for the AGI and the predictions he offer are very different from the media narrative that makes you believe that AGI is imminent.

the podcast episode that I referenced in the previous videos in which Dwaresh Patel interviewed Andre Karpathy and Cararpathy said that we are at least 10 years away from AGI and he clearly states that even that prediction is based on pure extrapolation and speculation really.

I covered extrapolation bias at length in one of the previous episodes and in my opinion the extrapolation bias is one of the biggest contributors to the current bubble that is present even among professional investors.

Sam Alman's now famous statement that he made when giving a talk at Stanford summarizes the delusion of chasing AGI whether it makes sense or not.

He said that he doesn't care if they burn $500 million or $5 billion or $50 billion a year.

They're making AGI and it's totally worth it.

So the way AGI is being described today is LLM's multimodality, some kind of magic equals AGI.

So we're definitely making AGI. We're burning billions and it's totally worth it.

Except we don't know what it is and what it really takes.

This framing makes an idea that would typically be labeled as unsustainable delusion into a vision.

OpenAI projects burning 115 billion through 2029 despite posting a 13.5 billion loss against a 4.3 billion in revenue during 2025 first half.

And the AGI narrative justifies this.

This wouldn't fly for anybody else. But when Alman says it, it's fine.

And in my view, there is another layer to this.

When we say that we must control AI for safety, this argument becomes justification for market concentration.

Because how do you apply anti- monopoly rules when critics argue that safety is used for monopoly and proponents argue that concentration is necessary for safety and the argument follows a structure where AGI is inevitable.

Therefore, whoever gets there first will capture extraordinary value. Therefore, current losses don't matter.

Now map this to the problem that I will once again remind you of.

We still cannot define what AGI is.

I found the following snippet in a fortunes article.

Among the biggest factors in AGI's sudden fall from grace seems to have been the roll out of OpenAI's GPT5 model in early August that landed with a thud.

Yes, the release of GBT5 had mixed reviews to say the least and I actually was among those who did not like it at all at first.

And then there is research from Yale Law Policy Review on how this anti- monopoly approach to AI governance shows concentrated control across the AI stack.

The semiconductor market is dominated by **Nvidia**（一家领先的图形处理器和AI芯片制造商） with 92% market share.

cloud computing, **AWS**（Amazon Web Services: 亚马逊的云计算平台）, **Azure**（Microsoft Azure: 微软的云计算平台）, **Google Cloud**（Google Cloud Platform: 谷歌的云计算平台） controlling 63% globally.

What the AGI safety narrative is doing is that it helps to legitimize this concentration by framing it as protective rather than extractive.

This purely hypothetical conversation about the imminence of AGI creates superficial urgency and the urgency benefits fundraising.

And that is why it is being lobbied so much.

For example, **Ilia Sutskever**（OpenAI联合创始人兼首席科学家）'s safe super intelligence raised $2 billion at a 12 billion valuation without a working product or revenue.

And **Thinking Machines Labs**（一家AI研究公司） secured billions purely on AGI promises.

I want to make a quick pause here because I've been noticing lately that people in the comments choose to hear what they would like to hear and I want to clarify the lens through which I approach all of my content and all of my research.

I analyze every single topic from product and business perspective.

My interest is purely academic and I'm studying how narratives translate into business.

The point I'm trying to make is that this would never fly in normal circumstances.

This is not about safe super intelligence.

Whether I like it or not, I have immense respect for Mr. Saskca and his work.

What I'm highlighting here is that we're talking about billiondoll investments justified by an idea that lacks a basic definition.

Another popular fear is how AGI will affect the labor market.

In fact, I got inspired for this episode after seeing the comments under my own videos where people were saying, "You're so cynical now, but wait till AGI comes for your job."

So, I was like, "You know what? Maybe I'm the one delusional here. Let me do the research."

So, I did some digging, found a bunch of AGI related future of tech, future of job reports, and after analyzing them for a week, I stand by what I said.

Let me show you.

Goldman Sachs and IMF 300 million jobs globally could be affected which is 9.1% of global workforce.

My next logical question is is 300 million a lot comparing to major historical crisis from the past.

Yes, it is a lot.

For example, great depression, jobs lost 15 million and that was only in the US.

The unemployment rate was 25% at peak.

The duration was 4 years to peak and then 10 plus years afterwards.

It really required World War II mobilization to recover from the Great Depression.

2008 financial crisis. Jobs lost 27 million globally.

The peak unemployment was 15 million in the US.

The recovery took 5 to seven years with wage stagnation.

And lastly, COVID 33 million increase in global unemployment in 1 year.

Unemployment peak 13%. 6.5 globally.

The recovery pattern was different. It was K-shaped. High earners gain jobs but low earners lost.

If we step away from the crisis analogy and consider normal annual churn in the US, for example, we're looking at 50 million job separations every single year in the US alone.

But these are immediately replaced by the new hires.

But the prediction around AGI and its impact on employment is net loss, not replacement.

And the more I was reading, the more I kept questioning what is the likelihood of this prediction fulfilling itself without AGI because the entire research is based on the assumption that AGI exists.

But it doesn't. It doesn't exist.

The report analyzes how generative AI could impact 300 million jobs.

But the displacement it describes requires AGI level functionality performing general cognitive work across entire occupations replacing fully replacing human workers at scale operating without any sort of supervision.

That is not A&I.

Their research objective and the methodology is solid.

They analyzed 800 plus occupations for automation potential.

But the predictions are based on the technological assumptions that AGI exists, which again it doesn't.

I'm not implying that this research is not substantive or that it isn't worth your attention.

All I'm trying to say is that it assumes that AGI exists.

And remember how I told you in the last video that a lot of people hear almost works and extrapolate that to works.

And as a result of this extrapolation, there is a lot of general anxiety about something that doesn't exist.

Compared to the whole AGI narrative, the AI bubble with its inflated expectations and deceptive marketing pales in utter insignificance because at least the technology exists.

You can touch it, you can smell it, you can use it, you can see it.

People dramatically overestimate current AI job displacement.

The claims about AI job displacements are highly questionable, and I made a series of videos dissecting layoff data.

But the problem is that the AGI narrative keeps fueling that anxiety.

There is a recent analysis from **Brookings**（布鲁金斯学会: 美国一家公共政策研究机构） and **耶鲁预算实验室**（Yale Budget Lab: 耶鲁大学的一个研究机构），顺便说一句，这是一项很棒的工作，我强烈建议你们阅读。

AGI叙事：允许巨额亏损的“通行证”

我在这里想剖析的最后一个角度是财务角度，因为在研究过程中我观察到，围绕AGI的投资动态与AI SaaS（软件即服务）和膨胀的风险投资资本有所不同。