Patrick McKenzie Profile picture
I work for the Internet and am an advisor to @stripe. These are my personal opinions unless otherwise noted.
168 subscribers
Nov 5 8 tweets 2 min read
If I clipped every good Byrne Hobart or Matt Levine line I’d never get around to writing my own stuff but this from Byrne is too good to not share: Image An extraordinary fact about finance is that there are some firms which are financial service providers specifically for scams which sometimes, almost as an industrial accident, bafflingly end up in a contractual relationship with a legitimate, successful company.
Oct 27 11 tweets 2 min read
How much could would you write if you could one-shot 10-100 line shell scripts or similar almost all of the time, in 10 seconds? You would write a stupid amount of code. Who cares if it is disposable? Dispose of it; it's basically free. "But what if the code is wrong?"

Skill issue, code is free to you. Write a test suite too, designed to be thrown away in under a minute. Write three independent implementations and vote on the answer. etc, etc
Oct 22 10 tweets 2 min read
Me to financial firm: *address change form*
Financial firm: Is this five digit number a post code?
Me to financial firm: Oh you have asked exactly the right person for geeking out about post codes. Did you know...
Second thoughts: That was not the efficient way to answer. "Why didn't they know what a post code looks like?"

Because a post code can look like so many things, like 100-0001, 20500, or SW1A 1AA, to use three codes from three nations that all correspond to a particular famous building/complex within them.
Oct 16 6 tweets 2 min read
Still working on a few essays about what I learned on using LLMs for coding but if you want a sneak peak, Complex Systems this week discusses the game I made in some detail.

I’m probably adding one essay to the series on LLMs for taxes. It feels a bit weird to need to continue saying this, but yes, LLMs are obviously capable of doing material work in production, including in domains where answers are right or wrong, including where there is a penalty for being wrong. Of course they are.
Oct 9 33 tweets 6 min read
So October 15th, the extended US tax deadline, is just around the corner, and I have some observations which are more about LLM progress than taxes.

Background: many people professionally involved with LLMs estimate 2026-2028 as the year where one can get an LLM to "do taxes." I have a fairly complicated situation and have put more of my points into tax procedure than many AI researchers, and I did not previously expect to actually have this capability available in 2028.

On basis of experience with review, but not full execution, rethinking that.
Oct 5 5 tweets 2 min read
I think the economic logic of this is inevitable, in that you’ll ~need video to get your N+1 marketing post through the noise on social media even as a Taylor Swift, and your choices are either a $100k shoot *per post* or a relatively junior employee with some taste. Also expect the friendly neighborhood barbecue joint to get the same-y Madison Avenue look on their ads versus the Comic Sans and pink aesthetic that they might otherwise have ended up with.
Sep 4 10 tweets 2 min read
Listening to a podcast (Trillions) a guest made an interesting claim:

Guest: You know when you swipe a card at [coffee shop] part of the fee pays for your ability to reverse the transaction with the coffee shop. But come on, no one reverses transactions with [coffee shops]. I have removed the name of the coffee shop so that no one believes I am commenting on private financial details when I say: oh you sweet summer child.
Aug 3 5 tweets 2 min read
The existence of YouTube does not make reading and writing less valuable. It gives children a constant companion who is responsive, preternaturally so, to their desires and curiosity. (I devote a bit of brain space—not too much but I pray not to little—to making sure that constant companion does not make the entire world look like a pale imitation of itself, which would be wrong but could easily look accurate to the subjective experience of a child.)
Jul 29 15 tweets 3 min read
This letter is what happens when a credit card PM misunderestimates how much of an adverse selection effect they have in a new portfolio. (In particular note the cap on cash back and the carveout for particular transaction types which some users are able to generate arbitrarily high amounts of or would naturally have arbitrarily high amounts relative to “normal” CC use.)
Jul 17 5 tweets 1 min read
If I can give a slightly more optimistic take on this: much of how commercial software development is done trades some resources for others, in ways that might not be rational for people with very different strengths than e.g. AppAmaGooFaceSoft or BigCo customers. A lot of AWS services exist so that two teams don't need to have a meeting.

That *is not a criticism of either AWS or those two teams.* That is a preference one can have about time allocation and corporate structure, and capitalism will help you satisfy it.
Jun 17 7 tweets 2 min read
So strange that card program managers make such a show of doing this careful balancing act when everyone who reads the Atlantic knows that the real source of rewards is cross-subsidization of elite cardholders by poor people. </sarcasm> Less sarcastically: it’s a math problem conducted by people who are pretty good at math, and the marching orders they get are “In general and in steady state, all of our card programs should be margin accretive. Make it happen. If you can’t you’ll need a senior signoff.”
Jun 14 18 tweets 3 min read
I think the so-called Bitcoin treasury companies have just reinvented exchange tokens: there is an asset with X real world utility but not naturally leverageable. It should flow to place in world where most leverage is bolted onto it; immediately incentive compatible. Repeat 100x And then “Holy %}*]^ how did so much of it end up in a place with grossly deficient risk management?!”
Jun 4 11 tweets 2 min read
(n.b. This is extremely well-known among companies which have a business process where you sign things. Most of them use a signature to demonstrate solemnization rather than authorization or authentication.) As I've mentioned previously, solemnization is a sociolegal tripwire to say "There are many situations in society and in business where you're Just Talking and up until this exact moment we have been Just Talking *and after this point* We Were Not Just Talking. Do you get it?"
May 27 6 tweets 1 min read
Apparently Japan Post is debuting the most obvious improvement in addressing for last two decades: address virtualization.

You sign up with them and get a short alphanumeric code. Their DB holds a pointer to physical address. If you move, you tell them, pointer changes. And then when dealing with an e-commerce merchant instead of doing the traditional laborious address entry (which in Japanese usually requires redundantly providing the pronunciation of the address as well) you just give them the code.
May 23 21 tweets 3 min read
Listening to @_rossry ‘s new podcast about drug development and the first episode about operational competence issues in clinical trials is giving me flashbacks. Link, inciting comments, and story time:

developmentandresearch.bio/meri-beckwith/
May 6 9 tweets 2 min read
Interesting article about falling backwards into founding a non-profit and then doing policy advocacy work, which had a number of points which resonated with me: Image huffpost.com/entry/utah-sch…
Apr 23 10 tweets 3 min read
In today's very surprising example of things an LLM could be good at:

I had a print failure while running a resin print in the wee hours of the morning.

Debugging these is a bit maddening. They arise from a combination of software, math, chemistry, and unpredictable chaos. They're also very underdocumented. (In what is surely a first in the history of manufacturing.)

The community is spread between various Facebook groups and Discords, and writes little down formally. Most recorded lore is in YouTube videos, and aimed at low-skill enthusiasts.
Apr 12 6 tweets 1 min read
This is 50% of my cycles this year compressed into a tweet. There are Sorts within the Sort, all the way down.
Apr 3 5 tweets 1 min read
I don’t have anything novel to contribute on the substance of but have to again comment, pace Situational Awareness that I think kicked this trend off, that single-essay microdomains with a bit of design, a bit of JS, and perhaps a downloadable PDF are…ai-2027.com … a really interesting form factor for policy arguments (or other ideas) designed to spread.
Mar 27 10 tweets 3 min read
This week on Complex Systems, a continued discussion of credit card rewards, interchange, and what I believe is a persistent misconception about how society should want justice done via payments systems.

It ends with the following, which the team took the liberty of putting into a short clip. (Sound on if you like hearing my voice, but video is subtitled.) Last week the Atlantic published an opinion piece which argues that the poor are subsidizing the rich's receipt of credit card rewards. This view has wide currency among certain advocates and among opinion writers.

It is not true.
Mar 20 7 tweets 2 min read
An argument I have had with some credit card enthusiasts for a very long time, paraphrased.

Enthusiasts: I’m robbing the bank blind!
Me: Doubtful? They are probably pretty happy to have a portfolio of you.
E: Oh by carefully layering promotions and making a spreadsheet and… Me: So checking my understanding: you spend a lot of money on credit cards.
E: Yes, that’s the whole point.
Me: And in a nation which makes it illegal to underwrite using an IQ test, you have self-constructed an IQ test.
E: Yes and I pass it obviously.
Me: Right. Tracking.