Anthropic CEO 万字长文：AI 将取代整个白领岗位

There has been one AI lab CEO who has been pretty consistent about his belief that utterly transformative AI will arrive in the next year or two or at least before 2030. Even that is less than 50 months away, which is weird to think about. But anyway, that lab that he leads, Anthropic, also happened to be the makers of Claude Code and Claude Co-work more recently and of course the Claude 4.5 Opus and Sonic models that power them.

I feel that the CEO Dario Amday does have a unique insight about the next one to two years. Which brings me to the last 48 hours during which Amade published an almost 20,000word essay on where he sees things going for better or worse. Yes, I've read it in full the old school way as well as several of the papers referenced in it. His last essay, Machines of Loving Grace, became the preoccupation of Silicon Valley for quite a while.

So, I just want to give you a giant head start on the conversations that will happen in 2026. For this video, I've broken it down to the four big claims that he makes about our near future. These predictions come under the umbrella of navigating the difficult teenage years of LMS, the adolescence of this technology. First, he predicts that tools like Claude Code will go from automating individual tasks like codew writing to automating entire job categories like software engineering.

He also references law and finance where the recent integration into Excel might help you at the moment do an individual task but he foresees it doing the entire job that you're doing. The engine behind that prediction is the smooth and simple extrapolation of the scaling laws. The fact that in his words AI systems get predictably better at essentially every cognitive skill that we are able to measure. Feed in more data and compute and you get a smooth unyielding increase in AI's cognitive capabilities.

This is him telling you to keep your eye on the ball and ignore those headlines about AI hitting a wall or being a bubble. Yes, certain tools are overhyped in the short run and certain companies may go bust, but the underlying curve is strong, consistent, and predictable. He argues the key claim again though is that he's saying that will take us from automating individual tasks within a job to the entire job. Let's hear out a bit more of his side of the argument before I add in some other context.

First, he says he's already predicted something like this in Machines of Love and Grace. That's powerful transformative AI that could be as little as 1 to two years away. Well, going back to that October 2024 essay, technically he predicted that it could come as early as 2026. So, it should be more like he predicted 0 to one years away. Back to this new essay, he does that a few times, not quite acknowledging how his predictions have shifted back a little bit.

Of course, as always with every single prediction, he caveats it heavily. He cites the incredible evidence that some of the strongest engineers and presumably some of the highest paid who work at Anthropic are now handing over almost all of their coding to AI. Notice that's their coding though, not their entire job. What's the difference? Well, I use Claude Code almost every day, and its best suggestions are genius that I wouldn't have come up with.

Its worst ones would destroy almost any app you create. But remember, we have a second extrapolation to contend with. Not only will we go, in his mind, from all coding to all software engineering being done, but from software engineering to all other white collar jobs. In his words, it cannot possibly be more than a few years before AI is better than humans at essentially everything as long as that basic exponential continues.

If anything, he thinks the exponential could speed up as AI starts to automate the job of doing AI research. This would create, he says, a feedback loop which is gathering steam month by month and may only be one to two years away from a point where the current generation of AI autonomously builds the next. In some he says there's a good chance all of this is coming in 1 to two years and if not that a very strong chance it comes in the next few before 2030.

Remember that this is coming from a lab leader who has overseen a 10x revenue growth year on year. Even in Silicon Valley, that is unprecedented growth for a company of his size. For this first of four predictions though, I'm going to add two caveats and then let DemsSarbis add in a third. And my first caveat is that I think he's slightly exaggerating the pace of progress in coding. He said, for example, in the last 2 years, AI models went from barely being able to complete a single line of code to writing all or almost all of the code for some people, including engineers and anthropic.

Well, one of the first experiments I did on chatbt in November of 2022 was get it to write some code and it created this miniature fitness app which I felt was amazing and really cool. So, it could write a single line of code. In fact, I remember viral videos of coders going, "Oh my god, we're all going to be automated based on the original chatbt November 2022. That's what 3 and a4 years ago." And this whole writing all or almost all of the code thing.

Well, I heard an estimate from an OpenAI engineer recently that their model Codeex, which is not a million miles away from Claude Code, was automating about 20% of their code. For Carpathy, it's about 80%. So, I think even if you focused on Clawude Code and Anthropic, you'd be probably talking more in the 80% 90% of code automation rather than 100%. My second caveat is on that extrapolation from software engineering to jobs in finance, consulting, and law.

I'm not at all saying those jobs are harder, but I think the feedback loops are longer. You overlook something in a law contract and that might come back to bite you in, say, 3 years rather than 3 seconds or 3 minutes with unit tests in software engineering. If an AI model skips out on a bit of nuance while analyzing the headcount while doing a consulting report for say McKenzie or Bane, the negative ramifications of that might not play out until the medium-term.

Then back to that engine, those scaling laws of more compute, more training tasks that for him have yielded a smooth increase in AI's cognitive capabilities. I would say he is now one of the only AI lab CEOs who thinks that this increase has continued to be smooth. I don't know whether that's because Anthropic focuses so much more on coding, but here's Google DeepMind CEO Deis on those same scaling laws.

Scaling laws um are going very well. So we're definitely seeing increased capabilities by putting in more compute, more data, uh, and making these models generally larger. So that trend is continuing. Um, may not be not as fast as it was a couple of years ago. So um, there's some talk of diminishing returns. Uh, and and but but there's a big difference between sort of no returns and exponential. And I think we're somewhere in the middle where there's very good returns and that's worth doing.

Um on top of that if I to you know in terms of like getting all the way to AGI artificial general intelligence um you know it may be that there's one or two uh big innovation still needed as well and maybe missing in addition to the scaling up of um kind of the existing ideas. Second mega prediction is that he foresees an unemployed or very low wage underclass of up to 50% of the population. You may or may not have seen plenty of viral posts on Twitter or X about you only have a few months to escape the permanent underclass.

Somewhat strangely, he thinks that this will affect those of lower intellectual ability, which he says is harder to change more than others. Honestly, I think this is a potentially quite toxic message to send to 18year-olds or 20somes because you're implying that they have to scramble to do everything to make their wages in the next year or two. Forget the long-term, drop everything, maybe inves

原文超出正文长度上限，此处截断——上游还有内容，完整版见上方「原文 ↗」。