Black Swan's Taleb Says Nvidia Rout Is Hint of What's Coming
bloomberg.com
bloomberg.com
https://techcrunch.com/2025/01/27/ai-startup-deepseek-pauses...
I don't believe it was a cyber attack; I think it's just usage overload. If I'm right, the next steps will be for them to throttle usage by price increase or some other method. Or they will have to find another order of magnitude of efficiency, and then that will be gobbled up and maxed out days after they introduce it.
The theme here is the demand for AI is nearly limitless and the supply (for now) is constrained by Nvidia.
Can you explain what information you're basing this on? Is there a particular reason not to believe them?
Typically, from a public relations standpoint, you'd be better off saying "we're super popular and haven't scaled enough yet" over "we got hacked".
But I don't really know much about DeepSeek. Do they have a pattern of lying?
So it does come as a plausible explanation.
Maybe there's a history of the company lying, or some other evidence that it's not an attack that I haven't seen, etc.
Note they offer a free reasoning model while OpenAI charges $200. And the reasoning is a lot more transparent.
My opinion is probably both an attack and an unmanageable surge right on Spring Festival (Chinese NY).
It's been red all day:
But I had no issues today with web. Yesterday night it was spotty.
They will recover and I am sure this is more than user interest. Look at how much is being claimed by competitors already. They are absolutely hammering them right now.
> too many investors have been bidding up prices of firms related to AI without properly knowing the details of how it functions or is able to succeed
There is no investment without risk. The question is why are there so many investors that appear to be grossly insensitive to gargantuan levels of risk. And the answer is probably: its all funny money to them. We live in the era of 100K bitcoin after all.
"Serious" money simply reflects a serious society: transparent books, reliable social contracts, diversified undertakings, resilient people and nature, truthful and lawful character, stable and boring. Money that reflects real economic exchanges rather than speculation.
The inverse attributes produce funny money: obfuscation, debasement, double books, smoke and mirrors, exploitation, captured institutions, subversion of the law and rampant speculation.
There is no easy way to good money. No secret sauce. The entire society must be behind its value.
Maybe we should treat predictions as messages ('do something if you don't like it') not certainties.
Careful there. Implying that some cultural norms are superior to others are going to get you into trouble in the wrong crowd.
> Why would it be surprising to see China with its capital and human resources catch up with the West and surpass it?
Because the message that diversity === strength is a powerful one. Plenty of people simply cannot wrap their heads around how positive an effect homogeneity can have on results; their built-in PC filters cut in before they even get to the thought.
I've observed decades of how, in the west, not only can you simply not say something, but any insinuation that you are thinking it is liable to result in consequences for wrong-think.
I think the east is going to catch up and surpass the west in science and technology. The best way to prevent this is to seed their society with the same ideological tarpit that the west finds itself in and hope like hell it catches on so that they too can have enforced inclusivity training on company time.
I do not actually agree with assertion being made either - the west has problems, but its biggest problems are complacency, economic policy, and an unwillingness to face difficult problems, not lack of homogeneity.
Who said anything about free speech? I said, and still stand by, that some cultures norms are superior to others, and that that sentiment is not acceptable to many in the west.
Therefore people have worked extremely hard over the decades to stop those sort of things.
I see from your profile you're from South Africa, so I'd assume you are more capable than many others of recognising that.
Perhaps things swung a bit too far one way (I genuinely am undecided), but we should realise the intentions in this started out well. And we'll swing too far back the other way if we're not careful.
You know, I didn't say anything about race.
Please don't misrepresent my point in order to argue against it because ... well, there's a reason that there's a common name for that sort of thing.
> I see from your profile you're from South Africa, so I'd assume you are more capable than many others of recognising that.
Not just "from", I grew up in Apartheid South Africa.
I saw the rubicon speech live on TV, not read it in a wikipedia article. I had to watch where I go because of the "whites only" signs in certain parts of the country.
And yet, I still get lectured on race by people who have never lived under legislated racism, which is a good deal harsher than "systemic racism".
> Perhaps things swung a bit too far one way (I genuinely am undecided), but we should realise the intentions in this started out well. And we'll swing too far back the other way if we're not careful.
Some cultural norms are superior, and have superior results, producing better and more productive members of society on average.
And some cultural norms are primitive and regressive, and produce less productive or net-negative productive members of society.
Race has nothing to do with it.
So for the benefit of doubt, the world has over many decades tried to fight against the sentiment that some cultures are superior because (and pay attention here) this line of thought (whether it is right or wrong), has led to horrific outcomes in the past, and still does.
I never made the point that "some cultures are superior".
You responded as if I said "some cultures or races are superior", and your second response refined my point to "some cultures are superior", both of which are incorrect characterisations of my position.
There's a big difference in the meanings between what I wrote and what you read, twice.
Do you imagine that people who think "some cultural norms superior to others" make it part of their identity?
its VERY like the dotcom boom.
This is kind of feeling like that to me. When this is over, I'm wondering if NVidia doesn't come out even stronger than it was when we started? There's just something that smells way too familiar about what's happening.
I'm thinking one could wait for NVidia to become a victim of negativity hype, and I bet you could buy and go long at that point. If the market gets irrational and cuts 80% of NVidia's value, man, it's pretty hard to think up a better opportunity than that laying around? I guess we'll see what happens.
Racism
What is probably more important is their quality. Though higher quantity tends to help with getting more of the world's best, too.
YCombinator makes money betting in startups, it sure has a vested interest in disruption. Would we be up in arms at YC regularly pointing out potential disruption paths ?
Being prudent is appropriate every hour, not twice a day, even if someone who is reckless can be lucky 11 hours of the day.
Another way to think about this is that returns are linear, and drawdowns are convex. 90% of your returns are going to be a normal distribution, but the tails become something like a pareto. The simplest thing to do is to hold convex payoff assets - ie buy puts. The devil is in the detail of knowing enough about options to determine what to buy to avoid bleeding all your profits away. This is the business model of Universa - they know what to buy, how much of it, and to a lesser degree when to buy it.
It’s a neat little PR headspace he’s gotten himself into. Basically any time a reporter is surprised they can call him and get a quote along the lines of “people should not be surprised when they are occasionally surprised, and they should anticipate being surprised some more in the future.” Really helps fill out the word count for the day.
It may not have been obvious to a certain type of investor, but many here on HN have said that Nvidia is overvalued. Also given the speed of which many others have court up to OpenAI, it wasn't unreasonable to expect someone to overtake them or find an alternative path.
*"Google 'We Have No Moat, And Neither Does OpenAI'" - https://semianalysis.com/2023/05/04/google-we-have-no-moat-a...
This may just be Taleb retrofitting terms to give him talking time. So while yes, it was entirely predictable, if you actually research what you're buying into and understand it... which most don't.
It’s mildly hilarious that you’re deriding Taleb for thinking (and investing) at a level that all of bubble-chasing investment houses are incapable of achieving. He’s made so much money precisely because this thing you’re describing as so stupidly obvious, is actually, not, stupidly obvious to people.
Aren't they high now? Or do you mean 1980's high?
The gist of his advice is generally that for the classic equity investments (ie buy the Nasdaq), your return distribution is heavily negatively skewed. You make money most of the time, and occasionally you suffer a fairly significant drawdown - ie a market crash. Nobody knows when the drawdowns will happen - that's exactly his thesis. Don't try to time the market, that's a fool's game. Structure your portfolio such that when the crash happens, you cut this convex downside.
Another way to think about this is that 90% of your returns are going to be a normal distribution, but the tails become something like a pareto. The simplest thing to do is to hold convex payoff assets - ie buy puts. The devil is in the detail of knowing enough about options to determine what to buy to avoid bleeding all your profits away. This is the business model of Universa - they know what to buy, how much of it, and to a lesser degree when to buy it.
managing options is hard. when do you close the hedge after it gives you a profit? when do you put it back on.
This is probably the most intuitive way to reason about it, but there's a subtlety there that relates to the magnitude of the crashes and their probability of happening. Intuitively, if the normal distribution was correct in estimating market behavior, you'd not expect to see 10 sigma events ever (we are looking at something like 1/10^20 of this happening). And yet, these events happen with some regularity when you try to use normal distributions for market returns.
So there are two aspects to this: one is that the normal distribution and people underestimate how often crashes can happen, and they also underestimate how big those crashes can be. It's the latter that is arguably more dangerous, because if you lose money more frequently than expected, you might end up earning less, but if you underestimate how bad a drawdown can be, you can be easily wiped out before you can react.
So why do we use the normal distribution so often in finance? Because it is convenient. It works fine for 99.99% of the cases, and it is easier to deal with the tails as a special beast rather than always having a complicated model to look at. There's also an element of lottery. Significant crashes happen once every 10 years roughly, so you don't need a great deal of luck to make a lot of money without seeing one :D
> managing options is hard. when do you close the hedge after it gives you a profit? when do you put it back on. It is definitely quite hard. Options are nonlinear instruments with a significant complexity to their behavior. This is why people pay Universa and other, lesser known tail risk funds to handle this complexity. You also have instruments like Variance Swaps, which allow one to easily lock in convexity pnl - considerably removing the timing aspect. You can construct these instruments synthetically using options, but this is not for the faint of heart.
It is doable on your own, but not without understanding options in depth and having tools that allow you to manage an option portfolio. These days, a person who is decent at freshman math and knows how to use python+pandas can easily manage this entirely on their own with a handful of scripts, but they'll likely need to spend a few months learning the theory and then it can take a year to build the neccessary intuition.
At this point I'm thinking AI is here to stay. But reliance on CUDA or a small set of companies probably not.
Sounds very un-Taleb like to me.
> Guess what Nassim is trying to say is that given the absurd amount of concentration on the US market, over a third of SPX is Mag, it will take a long painful ride to shake off the moronic crowding in positioning