This is an evergreen post. It’s linked from the header, and will be updated periodically. If I’ve made mistakes not listed here, let me know and I’ll issue a correction.

Vaccine Acceptance

Post: Was Vaccine Production Actually Delayed?
Summary: I wrote that historical ills, including the Tuskegee study may play a role in diminishing vaccine acceptance. Two links from Marginal Revolution (1, 2) indicate that there is no link.
Severity: Minor. This was an appendix to the post, and not central to the argument. Both links are to article published after my post, but one references this study published in 2007. So it only has a minor impact on my argument, but I am moderately embarrassed.

Techno-Pessimism

Post:: Isolated Demands for Rigour in New Optimism
Summary: Noah Smith is currently on part 3 of a 4 part rebuttal. I haven’t taken a good look yet, but will write a response if appropriate once the series is complete.

Substack

Post: How Substack Became Milquetoast

Type: Factual errors.
Summary: I wrote that Substack had 44 employees based on their LinkedIn profile. Nadia corrected me, they only had 17 employees at the time.
Severity: This weakens my argument that Substack has failed to use it’s immense resources effectively, but does not defeat the central thesis and is only mentioned in an epilogue to the main article.
Response: The article is corrected, with the original error preserved for posterity. An apology was issued over email.

Type: Punching down.
Summary: When I wrote this article, Substack was widely hyped, and had received little substantial criticism. Since then, they have been the subject of various critiques, some of them quite poor. As a result, I somewhat regret associating with this side of the argument. See for example, the misleading NYT headline The Site Trump Could Run To Next, and Here’s why Substack’s scam worked so well. My critique was focused on Substack as a replacement for eclectic blogs. I neglected to mention that it seems to be a much better replacement for op-ed pieces, and many of my criticisms do not apply in this capacity.

Effective Altruism Slowing Down

Post: Why Hasn’t Effective Altruism Grown Since 2015?
Type: Incomplete analysis.
Summary: Katja Grace wrote a reply, pointing out that I neglected several pieces of evidence pointing at continued growth. She was kind enough to email this to me directly.
Response: I wrote a reply here, agreeing with many of Katja’s points, but defending the bulk of my original piece.

Effective Altruism Shifting Priorities

Post: Responses and Testimonies on EA Growth
Type: Misattribution.
Summary: Scott Alexander wrote this comment, which I attempted to summarize in a follow up post. He then wrote that my summary of his position did not accurately capture his views.
Response: The summary was amended, and a retraction was added as a footnote.

S&P Annualized Returns

Post: The Byrne Hobart Portfolio
Type: Factual error.
Summary: I wrote that the S&P had returned “113.7% annualized since the May 2020 bottom”. The bottom was in March, not in May, giving annualized returns of 76.6%, not 113.7%.
Severity: Minor. The article is about how good Byrne’s portfolio is, and the mistake only strengthens this stance.
Response: This is corrected in the article. The original mistake is preserved and noted.

Forecasting Herd Immunity

Post: Contra StatNews: How Long to Herd Immunity?

Type: Misinterpreting evidence.
Summary: I wrote that antibodies only last 8 months. Alvaro noted that the papers actually indicate antibodies lasting at least 8 months.
Severity: Moderate. This alters projections by 1-2 weeks depending on other parameters. I was careful to list out model assumptions, but had said that antibody duration was among my more confident assumptions. There is an “Speculative armchair epidemiology” warning at the top of the article. This does not affect the central criticism of the original Stat News article. I repeatedly noted that the estimates should not be taken too seriously.
Response: This is noted in the article with a detailed correction.

Type: Mistaken assumptions.
Summary: Despite having previously summarized the survey data on vaccine acceptance, I neglected to realize just how low it would continue to be. I also forecasted 200,000 new confirmed cases per day, but they have since dropped drastically.
Severity: Moderate/Low. I didn’t account for low vaccine acceptance, or the lack of a vaccine for children. I did note that daily confirmed cases was my lowest confidence assumption.
Response: None.

Age of Top YC Founders

Post: Replying to Robert Wiblin on Young Rationalists
Type: Factual error.
Summary: I got a couple ages wrong in the original spreadsheet. Mean/median age at founding was previously listed as 26, but is actually 27. Mean/median age in 2020 was listed as 36 and 35, but are actually both 37. Thanks to Gytis Daujotas for catching these.
Severity: Minor. Ages are off by 1 or 2 years, which slightly weakens my argument.
Response: The post and data has been corrected, with the original mistake noted.

This is an evergreen post. It’s linked from the header, and will be updated periodically.

Stat News

Post: Contra StatNews: How Long to Herd Immunity?
Type: Factual errors.
Summary: A Stat article forecasting time to herd immunity misrepresents basic research findings and contains non-trivial arithmetic errors.
Epistemic Status of Source: High – Stat is widely acclaimed and often featured on Marginal Revolution as an authoritative source.
Attempts to Contact: They have a form for correction requests.
Response: N/A.

Substack

Posts: How Substack Became Milquetoast, The Best Writing Against, For, and On Substack
Type: Abstract theoretical issues.
Summary: Substack’s incentive structure enforces homogeneity.
Response: Nadia Eghbal (Head of Writer Experience) replied here.
Notes: Substack did later change their leaderboard, though probably not in response to my writing.

Lambda School

Posts: Austen Allred is Consistently Deceptive, Lambda School’s Incredibly Naive Incentive Alignment, Correlated Returns are Insufficient for True Alignment
Type: Factual errors, abstract theoretical issues and misleading statements.
Summary: Lambda School’s CEO has repeatedly lied about student outcomes and misrepresented legal action against the company. Their incentive alignment is not theoretically sound.
Epistemic Status of Source: Low – I cite podcasts, tweets and forum comments from Allred. Startup founders are already notoriously dishonest.
Attempts to Contact: Email to Lambda School Chief of Staff.
Response: N/A.
Notes: I later clarified that despite its flaws, Lambda School is not necessarily inferior to a 4-year college degree.

Walter Issacson on Leonardo da Vinci’s Mathematical Ability

Post: Beware the Casual Polymath
Type: Factual error.
Summary: Issacson claims that da Vinci found a solution to the Delian problem. He details an incorrect solution. Further investigation reveals several other incorrect solutions by da Vinci, and no correct ones. Correct solutions discovered by mathematicians do not match Issacson’s description.
Epistemic Status of Source: Medium – It’s a pop-history book, albeit a very popular one.
Response: N/A.

NFTs are stupid. I don’t feel compelled to make this case. Enough people have already flaunted their opinions (as well as their ignorance).

But how stupid is the conventional fine art world? Fairly stupid:

The drab free port zone near the Geneva city center, a compound of blocky gray and vanilla warehouses surrounded by train tracks, roads and a barbed-wire fence, looks like the kind of place where beauty goes to die. But within its walls, crated or sealed cheek by jowl in cramped storage vaults, are more than a million of some of the most exquisite artworks ever made.

Treasures from the glory days of ancient Rome. Museum-quality paintings by old masters. An estimated 1,000 works by Picasso.

…Swiss officials initiated an audit in 2012, the results of which were published two years ago. The results revealed a huge increase in the value of goods stored in some warehouses since 2007, led by an increase in high-value goods like art. Though the audit did not specifically measure the increase in stored artworks, it estimated that there were more than 1.2 million pieces of art in the Geneva Free Port alone, some of which had not left the buildings in decades.

Or more succinctly from MSCHF [1]:

We bought a Damien Hirst and the first thing the seller asked us was “know anybody in Connecticut to ship it to?” for who in their right mind thinks of anything, upon acquiring work by one of the 21st century’s most name-recognizable artists, than to flinch from the waiting blow of New York sales tax?

Bitcoin is stupid too. Thanks to the University of Cambridge, you can see that it’s energy costs are somewhere between Argentina and Ukraine, and over 0.5% of all energy consumption globally.

That’s just the negative externality. Bitcoin is stupid for its users too, and can only handle a maximum 7 transactions per second.

Altcoins are better, and Ethereum 2.0 will get up to 100,000 tps, but as a whole the ecosystem is still stupid. Here’s Alameda Research CEO and crypto billionaire Sam Bankman-Fried:

The problem is we didn’t actually have any [Tether] sitting around, so we had to wait for the blockchain, and that took a while… we did not happen to have any Tether on Binance right now, and so you know, it took us basically 20 minutes just to get anything there.

But how stupid are conventional stores of value? Fairly stupid:

The largest gold repository in the country is the Federal Reserve Bank of New York, on Liberty Street, which holds two hundred and sixteen million ounces of gold—worth more than two hundred and ninety-three billion dollars at today’s prices—for thirty-six foreign governments and organizations… And the city’s basements are pocketed with old gold-vault spaces, left over from the days, in the nineteenth and twentieth centuries, when the dollar, like most major currencies, was tied to the gold standard. Banks were required to keep gold in reserve, since anyone with an account could request his savings in gold coins to, say, settle a debt in a foreign country. “People really did this,” the financial historian Richard Sylla said last week. “They’d take gold out of the vaults, and take it a few blocks and put it on a ship.”

But that was the past, surely today’s system is more sophisticated:

Gold bars are moved between the compartments whenever one account holder pays another. Staff wear steel-toe footwear to protect their feet in case they drop one of the gold bars weighing 28 pounds (13 kg). Every time the compartments are opened or gold is moved, three Fed staff members are required to oversee the transaction. Each compartment is further locked behind a padlock, two combination locks, and the seal of the Fed’s auditor.

Or in Europe:

How does one move 14,000 gold bars and 1,000 boxes of gold coins across the city without losing anything along the way? This might sound like the set-up of a joke, but it was the very serious challenge facing the Dutch central bank in Amsterdam as it shifted some 200 tonnes of gold worth around €10bn (£9.1bn) to a new vault outside the city.

The result was a meticulously and closely guarded operation that included an army and police escort with motorcycle outriders and helicopters following overhead. The slow overnight convoy carrying the precious cargo took some 22 hours to travel 20km (12 miles) away at the weekend.

What about Decentralized Finance? The 2014 Mt Gox. hack resulted in a loss of 850,000 bitcoins, equivalent to $450 million at the time, or $47 billion at today’s value.

But that was just one of many exchanges to get hacked. Binance lost $40 million in 2019. Most recently, $275 million was stolen from Ku Coin in September last year.

It would be hard to imagine a worse system, but centralized finance is trying. The Wikipedia page for AIG reads like a bad anti-capitalist parody. In 2004 they had to pay the SEC $126 million for selling a “non-traditional insurance product” which “would appear to be insurance and accounted for as insurance, but did not involve any actual risk transfer”. In 2006 they paid a $1.64 billion settlement for “bid-rigging schemes”, “securities fraud” and “sham reinsurance transactions”. And then 2008 hit and AIG collapsed, propped up only by a $180 billion bailout from the federal government.

Though work on Bitcoin had begun before the financial collapse, the genesis block does include the line “Chancellor on brink of second bailout for banks”. DeFi isn’t the solution to all problems, but it was intended as the solution to at least one.

During the weed legalization debates, the rhetoric technique of inverting the Schelling point served to remedy failures of the imagination. In essence: “if weed were legal and alcohol were not, how could anyone justify legalizing it?” In contrast to weed’s relative safety both short and long term, alcohol results in an estimated 95,000 deaths per year in the US alone and accounts for 28% of all traffic-related deaths. Globally, the WHO estimates 3 million annual deaths. Not to mention being “involved in a majority of intimate partner assaults, and over a third of murders and sexual assaults”.

None of that necessarily means weed should have been legalized, but it does force you to question your intuitions, and wonder how much of your rationalizations are just the result of status quo bias.

You can use this same technique to improve your thinking about crypto. If Bitcoin was already the defacto standard, moving to gold would seem ridiculous. If DeFi already worked, no one could credibly suggest greater centralization.

None of that is to say that crypto will win. The existing Schelling point does matter. It matters because it hides the stupidity of the existing system, and blinds us from clear headed comparison.

Crypto is stupid. Everything else is also stupid. It’s all okay.

See Also
Gwern - Bitcoin is Worse is Better

Disclaimer
I own some crypto assets.