Dan Luu 写系统问题时，习惯先测量、再对照、最后才下结论。把「Ed Zitron's AI prediction track record」放到智能体、评测和线上系统里，真正要问的是：默认做法会不会系统性失败。下面用中文整理成可执行的工程笔记，去掉原站导航和无关链接。

2024: Meta, Google, and Microsoft are dying

I was curious how well the predictions of the most widely cited AI skeptic I've seen (Ed Zitron) have done, so I looked at how his predictions panned out. To disclose my own biases, I've never had a particularly strong pro or anti AI progress position. For example, in 2022, I did a comprehensive look at predictions Futurists made, including well-respected folks like Kurzweil and found them to be generally wrong on both the prediction results as well as the

One comment I've seen from a lot of AI skeptics when someone responds to an AI skeptic is that all of the people who are saying that AI isn't fake are self-interested liars. Personally (to my obvious detriment), I have no particular financial interest in AI companies. I own whatever the standard share of them is via boring index funds. I have some seed stage investments, but just due to the timing and what's gotten big, that part of my portfolio is underwe

How do people cite Zitron?

Because there are quite a few prediction results, let's look at one in detail before the complete list to get an idea of the kind of reasoning Zitron uses. We'll arbitrarily look at this November 2024 talk where Zitron says, among other things, the major tech companies (like Meta and Google) are dying and they're thrashing around on AI because they don't know how to grow .

Zitron specifically named Meta as a company that's dying ("it's a dying product, and it's kind of a dying company"). Meta's revenue and profit (GAAP operating income) have been

People with attention to detail on Zitron

When he talked about companies not knowing how to grow ("none of these companies anymore really know how to grow … in the desperation to try to reignite growth in a dying ecosystem the tech industry is going to shove this [AI] shit into everything"), he named Google and then Microsoft. Alphabet (Google's parent company) has had the following revenue and profit numbers:

And Microsoft's numbers have been (note that, for consistency, all numbers are calendar year numbers and not fiscal year numbers):

Some Zitron predictions

Although this wouldn't be in the spirit of Zitron's statement, one could argue that Meta is actually dying, it just hasn't died yet. However, the reasoning in Zitron's argument is incorrect here—the Meta, Google, and Microsoft ecosystems are not dying. Given how fast these companies are growing (in terms of revenue and profit), it doesn't seem that AI is, as Zitron implied, some kind of desperation move they're reaching for because "they don't know how to

To make the case that these things are dying, he pulls on minor issues that are not positioned to cause the very large changes he suggests are about to occur. For Meta, he cited some kind of alleged MAU drop for Facebook. Rather than use Meta's own MAU figures or any kind of revenue or profit numbers, he seems to have used numbers from Similarweb. My experience with 3rd party tracking numbers like this is that they're quite inaccurate and generally useless

Comparing to respected Futurists

For Google, he cites Prabhakar Raghavan, who he calls truly evil and "a computer scientist class traitor that sided with the management consultancy sect", as having done some kind of grievous damage to Google search. In his rants about Raghavan, he never credibly establishes that Raghavan is doing severe harm to Google search, and the Google search engineers who've commented on his rant don't seem to agree with the Raghavan as sole or even major reason for

But even if we posit that Zitron is right and the villain Prabhakar Raghavan defeated the hero Ben Gomes, causing some kind of issue for Google search, this still doesn't make the case that Google revenue growth is in trouble at large because they have a number of other major products (such as YouTube and Google Cloud) that could drive growth even if search wasn't growing.

How long can you maintain an incorrect position for?

Every significant part of the chain of reasoning here is not only incorrect, it's not plausible if you know anything about Google or big companies in general. I'll be the first person to say that Google search quality has some serious problems and that Google has been increasing the relative priority of revenue over the user experience over time. This was a source of consternation for a number of user-focused engineers at Google when I was there in 2013.

For one of the issues Zitron cites, ads being confusing to users, in 2013, I asked a search engineer about Google changing the background color of ads to look more like search results because there was a previous study that showed that more an ad looked like a search result, the more users got confused over whether a result was an ad or a real search result, and I'd heard that Google deliberately made the ads not look like search results to avoid user conf

值得单独记下的观察

• Feb 2024 : "I believe we're reaching the upper limits about what generative AI can do and how accurate its outputs can be." Wrong 4
• March 2024 : "Have We Reached Peak AI?"; another prediction that hallucinations mean that AI progress is limited to then-current levels Wrong
• April 2024 : "As I previously warned, artificial intelligence companies are running out of data …"; another prediction that models can't improve because there's no more data Wrong 5
• July 2024 : "Generative AI, as I said back in March, is peaking, if it hasn't already peaked. It cannot do much more than it is currently doing, other than doing more of it faster with some new inputs" Wrong
• July 2024 : "Generative AI models aren’t getting more energy-efficient, nor are they getting more “powerful” in a way that would increase their functionality" Wrong (models continued to get more powerful) 6
• August 2024 : "generative AI is a dead-end technology that has peaked” Wrong
• September 2024 : "o1 shows that OpenAI is both desperate and out of ideas", with a re-iteration of the idea that models can't improve due to lack of data Wrong
• Oct 2024 : "[OpenAI revenue] growth is already slowing, and will slow dramatically as we enter the new year" Wrong (OpenAI exceeded the forecasts and contiued to grow quickly)

落地时建议先做的 5 件事

1. 用自己的真实负载测，而不是只用公开榜或厂商数字。
2. 把评测设计成能抓到失败模式：平均分好看但尾部崩溃，仍然算失败。
3. 智能体默认不会好好用测试；要写进流程，而不是写在口头规范里。
4. 性能和正确性都要有基线，改模型或改语言前后必须能对比。
5. 结论写成可回滚的决策：哪一版配置、哪一版评测集、谁签字。

和智能体产品怎么接

龙虾PRO做 OpenClaw 落地时，同样吃「先测量再扩面」这条纪律：技能、数字员工和网关都要有可复现评测，而不是只看一次演示通过。

本文侧重全链路风控方法论。落地时请用自身业务单据做回放验证，不要把示例阈值直接当生产策略。 相关：风控体检 · 方案资源