4 min read

Frequency Analysis #5: Crashes, Cuts and Hacks

In 19th and early 20th centuries New York City two waves of immigration rapidly changed the city. Every decade from 1850-1890, about half a million Irish immigrants arrived to a city of only three-five million. From 1900-1950 the city doubled in size from roughly seven million to 15 million residents. Most of the new residents in the 20th century were from Southern, Central and Eastern Europe. The first stop for many families, first the Irish and then Italians and Eastern European Jews were tenement buildings.

These group-family apartments were so high-density that you could hear your upstairs neighbors take off their shoes. After a long day working in the garment industry workers would come home and take off their shoes. First one... then the other. This is the origin of the phrase, "waiting for the other shoe to drop." While we're in the August doldrums of the news cycle with few truly novel developments, there were a lot of stories playing out their second act.


I don't write much about American politics here for three reasons. First and foremost, there's just so much commentary out there already, some of it good, much of it awful. Second, I'm disinterested in horse-race journalism or anything that smacks of a Keynsian beauty contest. Third, the US has the odd combination of being the most influential nation, and yet is politically solipsistic. When a domestic politics story does interest me, it is often because of how the world views our political circus. Here's a story from Qatari state-owned news company, Al Jazeera, about the Chinese reaction to Trump's indictment news.

The Trump indictment has been a long time coming and it's hard to tell just how many pairs of shoes there are, but one clearer pair of events is the attempted coup in Russia by Wagner Group chief Yevgeny Prigozhin. News this week of a Wagner Group plane crashing near Tver, reportedly with Prigozhin on it, is not exactly a surprise. We'll probably never know exactly what happened, but isn't that beside the point?

In other Russian aerospace crash news, the moon mission we previously discussed seems to have gone off track. The Lunar probe appears to have crashed into the Moon last weekend, setting back the beleaguered Roscosmos, which underwent large organizational changes in the past ten years. With the construction of a new headquarters and the Chinese partnership warming up, this setback is an embarrassment for the revitalization of Russian science, technology and national pride.


The tragic resolutions don't end in geopolitical news, however. Banks are now taking their turn shedding jobs after the software and internet sector's turn in 2022. Bank of Montreal, Wells Fargo, USAA and others are reporting layoffs in news that is not exactly unexpected, but does feel strange given the strong labor market and receding inflation. Two Canadian banks are also feeling the woes of a challenging sector. Many of these cuts are in the mortgage divisions of banks, already hard hit by high interest rates.

Housing finance is particularly sensitive to interest rates, although prices have remained stubbornly high as supply is weak and millennial demand for family formation isn't abating any time soon. Outside of housing the rate hikes have done little to slow economic growth, except, of course in software and high-growth startups. Companies making a bet that they can lose money for many years to establish a market leading position and reap sustainable profits in the future are also highly sensitive to interest rates. Since the price of a dollar today relative to ten years from now has gone up substantially, many business plans have to be recalibrated, that is, except for one area of software.


I am of course talking about the much hyped area of artificial intelligence. While the froth is starting to cool off, with analysts asking, is the AI bubble already over? We've entered the inevitable comedown from the euphoria of hype into the hangover of skepticism. But skepticism tempered with emerging use cases isn't a bad thing. After assuming that AI would be a general purpose technology that could do anything and take everyone's jobs, we're figuring out what it is actually good for.

One of those things it's good for is nonsense. Nonsense? Why would nonsense be useful? It turns out that throwing vast quantities of nonsense is quite useful for testing software. Google demonstrated that engineers could dramatically increase code coverage using LLMs. Computer security analyst Dan Geer noted in a 2015 speech titled "Driven by Data" that we have no idea how useful fuzzing even is. We can throw nonsense at an input variable all day, but without any idea how complete the nonsense we throw at the app is we can't make any assumptions about how likely the code is to perform as expected. Of course, we'll never generate all possible inputs to an app, but with LLMs we can get a lot closer!

Since we're verging on the philosophical anyway, here's a piece by AI researchers Yann LeCun and Jacob Browning on LLMS and language that claims we've met the limitations of LLMS and he is us. LeCun and Browning point out that the obvious and natural barrier to the power of LLMs is that they are trained on text and text is situated in context that the LLMs do not have. This is similar to linguist Noam Chomsky's point about LLMs inability to think. It is true that the current state of AI cannot reason, explain its output generation process or provide novel causal explanations, but that doesn't make it useless!

Hacker Arion Kurtaj (and an unamed 17 year old), member of the Lapsus$ hacking group, was found guilty of computer misuse to break into the networks of NVIDIA, Revolut, Rockstar Games, British Telecom and Uber. If teen hackers are still able to cause millions of dollars of damage to companies despite the efforts of the security industry and professionals, we have to wonder about how useful not just fuzzing is, but computer security as a whole.