Andy Wilcox (@andy_wilco) header
Andy Wilcox (@andy_wilco) profile picture

Computer Scientist, Dad, FinTwit Fan, formerly performance+SRE @twitter, now @roblox, wine snob by night.

Florida, USA
Joined April 2013
LLMs harness the teleodynamic nature of language, encoding the intent in human artifacts to simulate coherent paths across a river of affordances. Though powerful, they remain simulacra atop non-biological substrates, and diverge from the dynamics of teleodynamic biotic systems.
5
14
3
56
5,891
Andy Wilcox retweeted
Here's one Math breakthrough in data structures! In January 2025, undergrad Andrew Krapivin (with Farach-Colton and Kuszmaul) introduced Elastic Hashing - a non-greedy open-addressing scheme that disproves Andrew Yao’s 40-year-old conjecture from 1985. Key mathematical results: - O(1) amortized expected probe complexity: constant average work per operation even as the table approaches full capacity - O(log(1/δ)) worst-case expected probes (δ = empty fraction). For example, around 7 probes at 99% full By strategically probing beyond the insertion point to create firebreaks, it avoids clusters and coupon-collector bottlenecks. This delivers near-constant performance without reordering elements. Traditional greedy methods were believed to degrade linearly near full capacity. Elastic Hashing shatters that limit. Huge implications for high-load databases and efficient systems.
2
24
167
11,301
This @HedgieMarkets post illustrates where the infinity scalable asset-light technology model meets the physical realities of an asset-heavy business that faces an upward sloping supply curve. We have long argued that AI compute is just another bit-atom commodity (like crypto) that uses a lot natural resources to create a valuable (unlike crypto) virtual asset. On the bit side, Big Tech is a price-maker with fat margins. On the atom side, a price-taker. Big Tech grew up in bits — search, social, e-commerce, office software: asset-light, infinitely scalable, natural monopolies. Build once, serve billions, watch costs fall every year. So they assume AI is the same game and will spend whatever it takes to own the market. But inference is also atoms, i.e. land, critical minerals and electrons, which are mostly molecules. In the commodity world, competition drives price to marginal cost: P = MC, which is upward sloping as volume rises. The better the models get, the faster they compete their own margins down to the physical floor which rises with volume. You can already see it. Microsoft just cancelled Claude Code because the cost to run it exceeded the value it returned — demand retreating the moment price met real cost. The irony: the customer pulling back was itself a hyperscaler. In April, Uber confirmed once again that AI compute demand is price elastic. Bottom line: they assumed AI costs would keep falling like they always did on the bit side; however, on the atom side, there is a hard floor that likely rises in the short run. I am not denying that the margins are still fat. But it’s not the same model. These guys are running towards obsolescing their own pricing power. Why did Rockefeller stop at the gas station and not vertically integrate into cars?
🦔Microsoft canceled its internal Claude Code licenses this week after token-based billing made the cost untenable, even for a company with effectively infinite cloud resources. Uber's CTO sent an internal memo warning the company burned through its entire 2026 AI budget in just four months. American AI software prices have jumped 20% to 37%, and GitHub (owned by Microsoft) is dropping flat-rate plans for usage-based billing across its products. My Take The AI subsidy era is ending in real time. The same company that put $13 billion into OpenAI and built the Azure infrastructure powering most of Anthropic's compute just looked at the bill from a competitor's coding tool and decided it was not worth paying. That is not a productivity failure on Anthropic's end. Token-based pricing is forcing every enterprise customer to confront the actual cost of running these models at scale, and the number turns out to be far higher than the flat-rate experiments suggested. This ties directly to my Gemini Flash post yesterday. Anthropic, OpenAI, and Google all raised effective prices in the last six months. Enterprises that built workflows assuming AI costs would keep falling are now watching annual budgets evaporate in months. Two outcomes look likely from here. Either enterprises scale back AI usage to fit budgets, which slows the revenue ramp the labs need to justify their valuations ahead of IPOs, or the labs cut prices and absorb the losses, which makes the unit economics worse at exactly the wrong moment. Both paths land in the same place, the numbers stop working, and somebody has to take the writedown. Hedgie🤗
36
178
27
902
243,162
Fun read from Eric if you ever used a terminal. Honestly I feel lucky to have learned (waves hand) all this stuff in 24x80.
It can be unsettling when you notice that a technical assumption you've been making for 40 years has quietly expired. This happened to me a few minutes ago. I maintain a game called "greed". It's an old-style game from the days of character-cell terminals. Not quite a classic deserving of museum status like Colossal Cave Adventure or nethack, but worth keeping alive because it's still solidly playable. And people still are playing it, because yesterday I got a minor bug report about it. Nothing user-visible, just a silly C build problem. I fixed it. Then, because I'm generally trying to get my old C projects out of C into more modern and safer languages, I tried asking my robot friend to port it to Rust. Which it promptly did. But then I noticed something that irritated me. The Rust code had a bunch of unsafe blocks in it, which went directly against my reasons for moving it to Rust. On further examination, I discovered that it was calling the C curses library to do its screen painting. This is where I have to explain about curses. It's an ancient C library for writing TUIs. It looks in your environment for a variable named TERM, uses its value to dredge a bunch of magic strings out of a system-wide database called "terminfo" that tells it how to manipulate your terminal, and then uses those magic strings for screen painting. On modern systems, TERM is always some variant of a color ANSI terminal. In times past, when people attached a wild variety of character cell terminals to Unix systems rather than just sitting at the console, it could have been lots of other things. Those days are gone, but the habit of always going through terminfo so you can support a couple of hundred terminal types has persisted. I prod robot friend to find me a pure Rust equivalent of curses so I don't have to do unsafe and call C code. It says, yes, there is such a thing and it's called crossterm. I tell it: change this code to use crossterm. Robot friend grinds for a bit, and then tells me it can't do that because I don't have cargo (the Rust package manager) installed. This is because I never write Rust by hand. When I ship programs written in Rust, it's because I ported them from some other language and don't expect to ever touch them again without having a robot to do the code-grinding for me. This is when things get slightly strange. It tells me that instead of porting to crossterm, it has written into the greed Rust source its own little screen-painting backend the implements a subset of curses calls and (this is the important part) assumes it's talking to a color ANSI terminal. Robot friend is not an old Unix hand. It doesn't know the unwritten law of the deep magic that you always go through terminfo because...because you might have to support hundreds of terminal types that no longer exist in this century? I blink. I look at the Rust code for the back end. It is small and elegant. No more unsafe. No more dragging around a bunch of C library code. This is ... the right thing? I push it to the public repository. What sealed the deal is that code, even code in a language as rebarbative as Rust, is wet clay now. If, against all odds I get a bug report that says somebody wants to play greed on something that isn't an ANSI terminal emulator, reinstating full curses support will take a one-sentence prompt to my robot friend and mere minutes. I hadn't had to directly confront before the fact that the entire set of assumptions that made TERM and terminfo a thing are as obsolete as dial-up acoustic modems. Still, the moment when I tossed away one of the ancient laws of Unix coding felt a bit like the universe lurching sideways. Indeed do many things come to pass...
2
130
Andy Wilcox retweeted
THE PEOPLE'S CHAMPION
2
1
8
962
アメリカでこの『ピザの上にピザをトッピングしたピザ』を見たとき「こんなヤツらに勝てるわけない…」と思った
425
4,374
414
45,236
6,031,486
Andy Wilcox retweeted
Replying to @debuggerdev
The next 5 years are going to be wild for people who can live 2–3 layers above the code.
8
8
288
46,535
Andy Wilcox retweeted
Crude scenario analysis, courtesy of shrubstack.com
36
139
17
1,553
225,039
Thanks Tony for a great conference and experience! Was so kind of @TgMacro, @DoombergT, and Sir JJ to spend a few moments with my son. Awesome meeting @SantiagoAuFund and @hkuppy! Fun to share a laugh with @ttmygh. We discussed macro all the way home, be back next year!
3
64
Andy Wilcox retweeted
Microsoft $MSFT disclosed for the first time that 45% of its $625bn book of future cloud contracts was from OpenAI.
150
359
150
4,907
1,205,121
Probably the most important graph for health, ever. Insulin resistance predicts age-related diseases. >200 healthy people tested for insulin resistance. After 6 years, no one in lowest third got heart disease, stroke, cancer, hypertension, or diabetes.
21
76
5
510
99,207
Andy Wilcox retweeted
(Me): so, we create this "virtual memory" to trick processes into thinking they have a contiguous address layout (Medieval peasant): Would not thy abstraction swell the tally of mem'ry lookups? (Me): well, yes. to deal with that we created the "translation lookaside buffer"
27
1,010
25
15,869
387,665
Andy Wilcox retweeted
“Gonna be home late tonight babe, I’m out monitoring the situation with a couple of guys. We might even stay and monitor a second situation if something comes up”
trillion dollar idea: sports bar but just for situation monitoring with live X feeds, flight radar, a bloomberg terminal, and Polymarket screens
365
3,037
180
42,943
2,484,804
Andy Wilcox retweeted
A good way to test if people actually want to live in a space colony is to build a city in Antartica. It's a thousand times more pleasant and would cost a fraction of the price, so if no one is interested in that then there will never be a space colony.
368
569
135
18,453
620,237
Andy Wilcox retweeted
#NRCNews: We've completed our final safety evaluation for the application submitted by TerraPower to build Kemmerer Power Station. nrc.gov/sites/default/files/…
3
23
7
118
17,574
Luckily since the Louvre made NFTs of their jewelry, even though the crowns physically were stolen, they still own the same assets. Because the tokens still exist and are in limited supply just as before. Nothing has changed. few understand blockchain technology.
309
1,031
211
15,150
619,431
My bad guys, we turned on the new tick frequency volatility surface fitter in us-east-1 yesterday and I got a call from Jeff he was a little mad
9
4
148
14,845
Satellites Are Leaking the World’s Secrets: Calls, Texts, Military and Corporate Data | Andy Greenberg & Matt Burgess, WIRED With just $800 in basic equipment, researchers found a stunning variety of data—including thousands of T-Mobile users’ calls and texts and even US military communications—sent by satellites unencrypted. Satellites beam data down to the Earth all around us, all the time. So you might expect that those space-based radio communications would be encrypted to prevent any snoop with a satellite dish from accessing the torrent of secret information constantly raining from the sky. You would, to a surprising and troubling degree, be wrong. Roughly half of geostationary satellite signals, many carrying sensitive consumer, corporate, and government communications, have been left entirely vulnerable to eavesdropping, a team of researchers at UC San Diego and the University of Maryland revealed today in a study that will likely resonate across the cybersecurity industry, telecom firms, and inside military and intelligence agencies worldwide. For three years, the UCSD and UMD researchers developed and used an off-the-shelf, $800 satellite receiver system on the roof of a university building in the La Jolla seaside neighborhood of San Diego to pick up the communications of geosynchronous satellites in the small band of space visible from their Southern California vantage point. By simply pointing their dish at different satellites and spending months interpreting the obscure—but unprotected—signals they received from them, the researchers assembled an alarming collection of private data: They obtained samples of the contents of Americans’ calls and text messages on T-Mobile’s cellular network, data from airline passengers’ in-flight Wi-Fi browsing, communications to and from critical infrastructure such as electric utilities and offshore oil and gas platforms, and even US and Mexican military and law enforcement communications that revealed the locations of personnel, equipment, and facilities. “It just completely shocked us. There are some really critical pieces of our infrastructure relying on this satellite ecosystem, and our suspicion was that it would all be encrypted,” says Aaron Schulman, a UCSD professor who co-led the research. “And just time and time again, every time we found something new, it wasn't.” The group’s paper, which they’re presenting this week at an Association for Computing Machinery conference in Taiwan, is titled “Don’t Look Up”—a reference to the 2021 film of that title but also a phrase the researchers say describes the apparent cybersecurity strategy of the global satellite communications system. “They assumed that no one was ever going to check and scan all these satellites and see what was out there. That was their method of security,” Schulman says. “They just really didn't think anyone would look up.” The researchers say that they’ve spent nearly the past year warning companies and agencies whose sensitive data they found exposed in satellite communications. Most of them, including T-Mobile, moved quickly to encrypt those communications and protect the data. Others, including some owners of vulnerable US critical infrastructure whom the researchers alerted more recently—and declined to name to WIRED—have yet to add encryption to their satellite-based systems. Researchers have pointed to the surveillance dangers of unencrypted satellite connections before, but the scale and scope of the new disclosures appear unrivaled. The researchers’ work looked at only a small fraction of geostationary satellites whose signals they could pick up from San Diego—roughly 15 percent of those in operation, by the researchers’ estimate. This suggests vast amounts of data are likely still being exposed over satellite communications, says Matt Green, a computer science professor at Johns Hopkins University who focuses on cybersecurity and reviewed the study. Large swaths of satellite data will likely be vulnerable for years to come, too, as companies and governments grapple with whether and how to secure outdated systems, Green says. “It's crazy. The fact that this much data is going over satellites that anyone can pick up with an antenna is just incredible,” Green says. “This paper will fix a very small part of the problem, but I think a lot of it is not going to change.” “I would be shocked,” Green adds, “if this is something that intelligence agencies of any size are not already exploiting.” Half Conversations, Broadcast From Space The phone calls and text messages the researchers obtained, in particular, were exposed due to telecoms’ often overlooked use of satellite communications for offering cellular coverage to normal phone users who connect to cell towers in remote locations. Some towers in desert or mountainous regions of the US, for instance, connect to a satellite that relays their signals to and from the rest of a telecom’s core cellular network, the internal communications of the network known as “backhaul” traffic. Anyone who sets up their own satellite receiver in the same broad region as one of those remote cell towers—often as far as thousands of miles away—can pick up the same signals meant for that tower. Doing so allowed the research team to obtain at least some amount of unencrypted backhaul data from the carriers T-Mobile, AT&T Mexico, and Telmex. The T-Mobile data was particularly significant: In just nine hours of recording T-Mobile backhaul satellite communications from their single dish, the researchers collected the phone numbers of more than 2,700 users as well as all the phone calls and text messages the researchers received during that time. They could, however, only read or hear one side of those conversations: the content of the messages and calls sent to T-Mobile’s remote towers, not sent from them to the core cell network, which would have required another satellite dish near the one T-Mobile intended to receive the signal on the other end. “When we saw all this, my first question was, did we just commit a felony? Did we just wiretap?” says Dave Levin, a University of Maryland computer science professor who co-led the study. In fact, he says, the team didn’t actively intercept any communications, only passively listened to what was being sent to their receiver dish. “These signals are just being broadcast to over 40 percent of the Earth at any point in time,” Levin says. Mexican telecom Telmex also transmitted unencrypted voice calls, the researchers found. The researchers further discovered that AT&T Mexico transmitted raw data over satellites that included users’ internet traffic—most of which was encrypted with HTTPS by the apps or browsers they used—but also some calling and texting metadata. They also found decryption keys that the researchers believe could likely have been used to decipher other sensitive information the AT&T Mexico network transmitted—though they didn’t attempt this. Starting in December 2024, the researchers began contacting the affected telecoms. T-Mobile responded by encrypting its satellite transmissions within weeks, but responses from other cell carriers were mixed. “Last year, this research helped surface a vendor's encryption issue found in a limited number of satellite backhaul transmissions from a very small number of cell sites, which was quickly fixed,” a T-Mobile spokesperson says, adding the issue was “not network-wide” and that the company has taken steps to “make sure this doesn't happen again.” In another statement after this story was published, T-Mobile noted that it has also added Session Initiation Protocol (SIP) encryption for all customers across the US “to further protect signaling traffic as it travels between mobile handsets and the network core, including call set up, numbers dialed and text message content.” A spokesperson for AT&T says the company “promptly” fixed the issue. "A satellite vendor misconfigured a small number of cell towers in a remote region of Mexico,” they say. Telmex did not respond to WIRED’s request for comment. Whether other cellular carriers around the US and world—outside the visibility of the researchers’ satellite dish—have encrypted their satellite-based network backhaul data remains an open question. The researchers say they didn’t see any unencrypted Verizon or AT&T US traffic from their dish. The AT&T spokesperson says that its US and Mexico networks are separate, and it is “rare” to use satellites for cellular backhaul. "We typically route traffic on our closed, secure backhaul network,” the spokesperson says. “On those rare instances where data must be transmitted outside our closed network, it is our policy to encrypt it." Verizon did not respond to WIRED’s request for comment. Beyond just cell towers in remote locations, it’s possible that a lack of encryption for cellular backhaul data could make anyone on the same network vulnerable, points out Johns Hopkins’ Green. Hackers might be able to perform a so-called relay attack with a spoofed cell tower—using the surveillance hardware sometimes called a stingray or IMSI catcher—and route any victim’s data to a cell tower that connects to a satellite uplink. “The implications of this aren't just that some poor guy in the desert is using his cell phone tower with an unencrypted backhaul,” says Green. “You could potentially turn this into an attack on anybody, anywhere in the country.” Military Helicopters and Power Grids, Exposed The researchers’ satellite dish also pulled down a significant collection of unprotected military and law enforcement communications. They obtained, for instance, unencrypted internet communications from US military sea vessels, as well as the vessels’ names. (A spokesperson for the US Defense Information Systems Agency acknowledged WIRED’s request for comment but had not provided a response at the time of writing). For Mexican military and law enforcement, the exposures were far worse: The researchers say they found what appeared to be unencrypted communications with remote command centers, surveillance facilities, and units of the Mexican military and law enforcement. In some cases, they saw the unprotected transmission of sensitive intelligence information on activities like narcotics trafficking. In others, they found military asset tracking and maintenance records for aircraft like Mil Mi-17 and UH-60 Black Hawk helicopters, sea vessels, and armored vehicles, as well as their locations and mission details. “When we started seeing military helicopters, it wasn’t necessarily the sheer volume of data, but the extreme sensitivity of that data that concerned us,” says Schulman. The Mexican military did not immediately respond to WIRED’s requests for comment. Just as sensitive, perhaps, were industrial systems communications from critical infrastructure like power grids and offshore oil and gas platforms. In one case, they found that the Comisión Federal de Electricidad (CFE), Mexico’s state-owned electric utility with nearly 50 million customers, was transmitting its internal communications in the clear—everything from work orders that included customers’ names and addresses to communications about equipment failures and safety hazards. (A CFE spokesperson acknowledged WIRED’s request for comment but didn't provide a response before publication.) In other cases they have yet to publicly detail, the researchers say they also warned US infrastructure owners about unencrypted satellite communications for industrial control system software. In their phone calls with those infrastructure owners, some owners even expressed concerns that a malicious actor might have the ability to not only surveil the control systems of their facilities, but also, with enough sophistication, potentially disable or spoof them to tamper with the facility’s operation. The researchers obtained a vast grab bag of other miscellaneous corporate and consumer data: They pulled down in-flight Wi-Fi data for Intelsat and Panasonic systems used by 10 different airlines. Within that data, they found unencrypted metadata about users’ browsing activities and even the unencrypted audio of the news programs and sports games being broadcast to them. They also obtained corporate emails and inventory records of Walmart’s Mexican subsidiary, satellite communications to ATMs managed by Santander Mexico, as well as the Mexican banks Banjercito and Banorte. A spokesperson for Panasonic Avionics Corporation said they “welcome the findings” from the researchers, but claim it “has found that several statements attributed to us are either inaccurate or misrepresent our position.” When asked, the spokesperson did not specify what the company considered was inaccurate. “Our satellite communications systems are designed so that every user data session follows established security protocols,” the spokesperson says. “Generally, our users choose the encryption that they apply to their communications to suit their specific application or need,” says a spokesperson for SES, the parent company of Intelsat. “For SES’s inflight customers, for example, SES provides a public Wi-Fi hot spot connection similar to the public internet available at a coffee shop or hotel. On such public networks, user traffic would be encrypted when accessing a website via HTTPS/TLS or communicating using a virtual private network.” The researchers reported the swaths of unencrypted satellite communications from the Mexican government and Mexican organizations to CERT-MX, the country’s incident response team, which is part of the government’s National Guard, in April this year, before separately contacting companies. CERT-MX did not respond to WIRED’s repeated requests for comment. A spokesperson for Santander Mexico says that no customer information or transactions were compromised, but confirmed that the exposed traffic was linked to a “small group” of ATMs used in remote areas of Mexico where using satellite connections is the only option available. “Although this traffic does not pose a risk to our customers, we took the report as an opportunity for improvement, implementing measures that reinforce the confidentiality of technical traffic circulating through these links,” the spokesperson says. “While we cannot share specifics, we can confirm that our communications lines have been evaluated and confirmed secure,” a spokesperson for Walmart says. (The researchers confirm that they observed Walmart had encrypted its satellite communications in response to their warning.) “The information of our customers and infrastructure is not exposed to any vulnerability,” a spokesperson for Grupo Financiero Banorte says. Banjercito could not be reached for comment. “SIA and its members remain diligent in monitoring the threat landscape and continue to participate in various security efforts with government agencies, industry working groups, and international standards bodies,” says Tom Stroup, the president of the Satellite Industry Association, adding that it does not comment on specific company issues. Time to Look Up The amount of Mexico-related data in the researchers’ findings is, of course, no coincidence. Although their satellite dish was technically able to pick up transmissions from around a quarter of the sky, much of that swath included the Pacific Ocean, which has relatively few satellites above it, and only a small fraction of the transponders on the satellites it did see were transmitting data in the direction of its dish. The result, the researchers estimate, was that they examined only 15 percent of global satellite transponder communications, mostly in the western US and Mexico. That suggests anyone could set up similar hardware somewhere else in the world and likely obtain their own collection of sensitive information. After all, the researchers restricted their experiment to only off-the-shelf satellite hardware: a $185 satellite dish, a $140 roof mount with a $195 motor, and a $230 tuner card, totaling less than $800. “This was not NSA-level resources. This was DirecTV-user-level resources. The barrier to entry for this sort of attack is extremely low,” says Matt Blaze, a computer scientist and cryptographer at Georgetown University and law professor at Georgetown Law. “By the week after next, we will have hundreds or perhaps thousands of people, many of whom won’t tell us what they’re doing, replicating this work and seeing what they can find up there in the sky.” One of the only barriers to replicating their work, the researchers say, would likely be the hundreds of hours they spent on the roof adjusting their satellite. As for the in-depth, highly technical analysis of obscure data protocols they obtained, that may now be easier to replicate, too: The researchers are releasing their own open-source software tool for interpreting satellite data, also titled “Don’t Look Up,” on Github. The researchers’ work may, they acknowledge, enable others with less benevolent intentions to pull the same highly sensitive data from space. But they argue it will also push more of the owners of that satellite communications data to encrypt that data, to protect themselves and their customers. “As long as we’re on the side of finding things that are insecure and securing them, we feel very good about it,” says Schulman. There’s little doubt, they say, that intelligence agencies with vastly superior satellite receiver hardware have been analyzing the same unencrypted data for years. In fact, they point out that the US National Security Agency warned in a 2022 security advisory about the lack of encryption for satellite communications. At the same time, they assume that the NSA—and every other intelligence agency from Russia to China—has set up satellite dishes around the world to exploit that same lack of protection. (The NSA did not respond to WIRED’s request for comment). “If they aren't already doing this,” jokes UCSD cryptography professor Nadia Heninger, who co-led the study, “then where are my tax dollars going?” Read more: wired.com/story/satellites-a…
22
52
10
271
38,229