Among my Silicon-Valley-and-adjacent friends, there is a general social norm against criticizing startups. This isn’t an aversion to criticizing tech companies in general — it’s fine to critique Big Tech companies. It’s criticizing a startup that’s considered poor form.

In this post, I will argue that there are no good reasons for this norm.

The startup equity market isn’t efficient

One argument is that the startup’s investors know better than you. This fails because the startup equity market is not efficient, as there is no way to short a startup. You can only go long one. So, a startup’s valuation is set by the VCs who are most optimistic about it, while the pessimists can only stay out. Edward Miller’s 1977 paper shows that a market can’t be efficient if it doesn’t allow short-selling. In Inadequate Equilibria, Eliezer Yudkowsky phrases it like this: the startup equity market can stay inefficient because it is also inexploitable. He gives this example:

Not everything that involves a financial price is efficient. There was recently a startup called Color Labs, aka Color.com, whose putative purpose was to let people share photos with their friends and see other photos that had been taken nearby. They closed $41 million in funding, including $20 million from the prestigious Sequoia Capital.

When the news of their funding broke, practically everyone on the online Hacker News forum was rolling their eyes and predicting failure. It seemed like a nitwit me-too idea to me too. And then, yes, Color Labs failed and the 20-person team sold themselves to Apple for $7 million and the venture capitalists didn’t make back their money. And yes, it sounds to me like the prestigious Sequoia Capital bought into the wrong startup.

The startup equity market is inexploitable like the housing market is. You should not expect an arbitrary house to be well-priced, because no short-seller can make a synthetic financial instrument that behaves just like that house and sell it into the same housing market frequented by the typical buyers of houses. Yudkowsky writes:

Some smarter agents might decline to buy [an overpriced house], and so somewhat reduce demand. But the smarter agents can’t actually visit Boomville and make hundreds of thousands of dollars off of the overpriced houses. The price is too high and will predictably decline, relative to public information, but there’s no way you can make a profit on knowing that. An individual who owns an existing house can exploit the inefficiency by selling that house, but rational market actors can’t crowd around the inefficiency and exploit it until it’s all gone.

Whereas a predictably underpriced house, put on the market for predictably much less than its future price, would be an asset that any of a hundred thousand rational investors could come in and snap up.

So: you can claim a startup is overpriced without offending an EMH-respecting economist.

Good startup ideas are rare

Precisely because going short a startup isn’t possible but going long one is, we should expect good startup ideas to be hard to find. An overpriced business idea is unsurprising, but an underpriced one should’ve been snapped up fast. The answer to “If this is such a good business opportunity, why has no one done it before?” can almost never be “Anybody else who could is too stupid/culturally stodgy/nontechnical.” (On the other hand, arguing that other players have a different preference function from you is probably fine. For example, in “There aren’t enough smart people in biology doing something boring”, Abhishaike Mahajan argues that the state of CROs is dismal because the best people don’t want to start one: the important innovations are “largely going to be logistical ones, not scientific”. If you value fun and prestige much less than similar-calibre peers, you can run a successful business.)

Fortunately, most startups in my experience argue only that a problem wasn’t solvable until recently because some technology or regulation just became available — e.g., the AI-agents-for-X startups founded after ChatGPT, companies like Amazon that were timed to the Internet, and drug discovery startups built on the Orphan Drug Act and neglected tropical disease priority review vouchers. But occasionally I hear a startup pitch arguing that incumbents just have a poor culture, or aren’t informed enough. This is usually naivete about how quickly free energy in a big world with incentives gets eaten. If the founders dig deeper, they’ll almost always find that the system is more complicated than they’d like.

Optimists only make money when they’re highly diversified

There is the Nat Friedman-style argument: “Pessimists sound smart. Optimists make money.” This may be true for VCs, but it’s probably not true for individuals looking to found or join companies. VCs diversify across multiple startups, and famously have power law outcomes. A few startups return the fund multiples over. This distribution of outcomes is so central to the business that the definitive history of the industry is named after it (and in it, Vinod Khosla — by any metric a successful VC — says that he took a sabbatical at the Santa Fe Institute to study systems with power-law outcomes). The downside on an investment is capped, but the upside can be enormous. This is especially true for seed and series A investors, who are also fittingly the most likely to dispense the gospel of optimism. For example, in “Early Work”, Paul Graham asks you to “switch polarity entirely, from listing the reasons an idea won’t work to trying to think of ways it could”.

But founders and startup employees don’t have this kind of highly parallel diversification. They commit to a single bet for years. In “Early Work”, Graham continues: “[I]n a field where the new ideas are risky, like startups, those who dismiss them are in fact more likely to be right. Just not when their predictions are weighted by outcome.” A career is serial enough and the average individual’s bankroll is small enough that blanket enthusiasm will lead to going bust before you see the fat tails of the distribution.

Ideally, founders and employees should have no systematic bias in evaluating ideas at all. But if there must be a bias, it’s not clear it should be optimistic. The historical record doesn’t show that successful founders are broadly optimistic about the probability any given idea will work; instead, they seem like cautious skeptics who bet on an idea that survived a high bar. As Luke Muehlhauser writes, much of what’s described as entrepreneurial “optimism” is probably “ability to understand expected value”. For example, Elon Musk estimated that Tesla and SpaceX only had a 10 per cent chance of success, but considered the upside high enough to found them anyway. Jeff Bezos told investors that there was a 70 per cent chance they would lose their investment. Demis Hassabis considered relative optimism between technologies important enough that he declined to let Facebook acquire DeepMind: Mark Zuckerberg sounded as enthusiastic about AI as he did about virtual reality, augmented reality, and 3D printing, and Hassabis wanted someone who understood that AI would be bigger than all those other things. Andy Grove named his life story “Only the Paranoid Survive” and led Intel’s painful retreat from memory chips. Jensen Huang told employees for years that Nvidia was “30 days from going out of business”.

Optimism isn’t a virtue

There is the idea that only building counts, and criticizing is much easier than building, so criticism has no value. The most literal example of this is Theodore Roosevelt’s quote that “[i]t is not the critic who counts. […] The credit belongs to the man who is actually in the arena, whose face is marred by dust and sweat and blood.” This seems obviously wrong. Really, credit doesn’t belong to any of these critics?

  • Theo Baker, the Stanford freshman who uncovered the university president Marc Tessier-Lavigne’s academic misconduct in the school’s student newspaper.
  • Data Colada, which found that Dan Ariely and Francesca Gino’s behavioral psychology papers had fabricated data.
  • John Ioannidis, the academic who wrote the first public criticism of Theranos.
  • Thomas Herndon, the grad student who found that a landmark paper used to justify austerity had coding errors and questionable data exclusions.
  • Frances Kelsey, the FDA reviewer who refused to approve thalidomide despite pressure from the manufacturer; the drug later turned out to cause serious birth defects.
  • Dan McCrum, the Financial Times journalist who uncovered billion-dollar fraud at the fintech Wirecard and was prosecuted by the state regulator BaFin for it.
  • Hindenburg Research, the short-seller that authored a report on the $13 billion EV startup Nikola which led to the founder’s fraud conviction.
  • Recently and much more down to earth: the anonymous Twitter poster @fynnso, who found that Cursor’s Composer 2 was an unlicensed finetune of Kimi K2.5. And just this morning, @Clashluke, who found that Underdog’s Ternary is a modified PrismML Bonsai 2, with attribution removed.

Obviously, the critic counts: preventing a misallocation of scarce capital and time is valuable.

Perhaps the concern is that being wrong when you predict failure isn’t adequately punished. There are two replies to this. First, I don’t see any evidence that being wrong when you predict success is punished especially gravely, despite the harms from it being no less real: wasted dollars and time. Second, I don’t think it’s even true that incorrect dooming goes unpunished. Frequently, it becomes cultural legend:

  • Gary Marcus, Margaret Mitchell, and Francois Chollet aren’t generally well-regarded for having underrated LLM capabilities.
  • The “Dropbox is technically trivial” HN comment is still a widely-shared screenshot.
  • Alex Karp is fond of raining on Michael Burry and others who shorted Palantir.
  • Ernest Rutherford’s “nuclear energy is ‘moonshine’” comment is infamous.
  • So is the New York Times editorial which declared manned flight would be impossible for “one million to ten million years” just 69 days before the Wright brothers’ demonstration at Kitty Hawk.
  • So is Paul Krugman’s prediction that the Internet would be no bigger than the fax machine.
  • So is Paul Ehrlich’s losing end of the Simon-Ehrlich wager.
  • So is Robert Malthus.
  • So is Thomas Edison’s loss in the War of the Currents.

These stories have no less real estate in popular memory than stories of incorrect championing: Theranos, WeWork, FTX, Segway, Friend.

Two things make it even more likely now for people to be measured about their pessimism: the recent norm of asking people to bet on their beliefs, popularized by prediction markets taking off, and the Internet ensuring you leave permanent records of your beliefs.

Sharing obvious knowledge

An omerta against criticism leads to decision-relevant information getting siloed. The effects are felt most strongly by beginners who lack connections or expertise. For example, see this tweet by the ML researcher Keller Jordan:

When I didn’t have connections, I didn’t realize that everyone who was connected already knew that overclaiming was rampant, and even knew that specific highly cited papers didn’t reproduce. So I just naively took the highly-cited claims that I found in the literature at face value, which kinda drove me nuts for a bit and wasted a good amount of my energy

The guard against piling on startups is a norm to criticize with calibrated severity — not a norm against criticism altogether. Fortunately, a norm for calibrated criticism is a stable one, as fake-wolf-criers lose credibility and timely-alarm-sounders gain credibility for saving people time and money.