Q McCallum

@qethanm.bsky.social

652 Followers 172 Following

research: (ML+gen)AI, risk, complexity, finance history LATEST BOOK: https://twinwolvesai.com/book/ website: https://qethanm.cc newsletter: https://complex-machinery.com (some posts in 🇫🇷 🇩🇪 🇷🇺)

The cover of Paul Scharre's book "Army of None"
Excerpt from article: Greg Walton, a cybersecurity researcher who has studied similar Chinese systems, said the exposure of this one was not an anomaly. It was a consequence of China's "surveillance sprawl," which is fueled by an expanding ecosystem of vendors, contractors and public security agencies, he said. Mr. Walton, a senior investigator at the SecDev Group, a Canadian research firm, said that each new platform "increases the number of places where sensitive personal data can be misconfigured, copied or left externally discoverable."
Top: excerpt from Complex Machinery 064, 2026/06/25: > There's a running (half-)joke that corporate genAl use cases start with the premise that you're an idiot. Which seems harsh, but only somewhat. Many of them sound like they're based on a caricature of someone who exists only in the mind of a desperate product manager. Bottom: Sam Altman tweet, 2026/07/31 > cool use case of chatgpt work i heard last night: > > connect your family calendars and explain your kids' interests. > every morning for the drive to school, have it make a podcast that talks about one kid's soccer game that afternoon, one kid's upcoming birthday, some news, etc.
WSJ headlines: Left: Top American Al Execs Sound Alarm on Chinese Models Right: OpenAl Models Escaped and Hacked a Company in Cybersecurity Test Gone Wrong
The spines of five books: - Boom and Bust - Devil Take the Hindmost - The Big Short - Money Men - Pyramid of Lies
Excerpt from Complex Machinery 020: >> You know that a chatbot's entire job description is Just Make Some Shit Up. (It's in the name: everything that comes out of a genAl bot is, well, generated. We only apply the label "hallucinations" to the generated artifacts that we don't like.) It's making everything up based on patterns surfaced in its training data, yes. But those are all grammatical patterns. Not factual. Not logical. Just this-word-is-likely-followed-by-that-word. And that kind of creative whimsy is unfit for sensitive topics. https://newsletter.complex-machinery.com/archive/020-silence-is-golden/
Excerpt from a Wired article: >> It’s also notable that Instagram users may be remixing your images using AI without your knowledge. “You will not be notified about content created using AI features at Meta,” reads the company’s help page. Of the numerous aspects of this rollout that make me uncomfortable, not even getting a notification when someone does this seems like the biggest red flag of all.
An image from Complex Machinery 018 titled "Should AI Do This Work?" (https://newsletter.complex-machinery.com/archive/018-robot-jobs/) It shows "perfection" as the intersection of "machines do it well" and "people hate doing it". "Here be dragons" is the intersection of "people like doing it" and "machines do it poorly"
An image from Complex Machinery 018 titled "Should AI Do This Work?" (https://newsletter.complex-machinery.com/archive/018-robot-jobs/) It shows "perfection" as the intersection of "machines do it well" and "people hate doing it". "Here be dragons" is the intersection of "people like doing it" and "machines do it poorly"
Excerpt from Complex Machinery 033: >> Fast, to-the-point, and often wrong >> >>People in leadership roles are always looking for fast, simple answers. An all-too-common interaction involves them getting frustrated when someone - usually in a technical or analytical role - says that a situation requires more nuance. "Yes yes," the executive protests. "I understand that. But just give me the answer. Is it A or B???" >> >> Enter genAl: it speaks with confidence It gives short, definitive answers to any question! (Unless you are asking about certain historical events.) Just what the boss ordered. Because it's the boss's mirror image. https://futurism.com/artificial-intelligence/bosses-obsessed-with-ai
Left: headline from Complex Machinery #033 (2025/05/25) >> #033 - GenAI is an executive's best friend >> >> I'll emphasize: "best" friend. Not always "a good" friend. Right: headline from Futurism (2026/06/29) >> Bosses Are Becoming Obsessed With Al, Using It to Make Every Decision, Barraging Their Employees With Nonsensical ChatGPT Directives, and Even Asking It Who to Fire >> >> "It's like being in an abusive marriage... I'm sitting here dealing with somebody that isn't living in reality."
An image from Complex Machinery 018 titled "Should AI Do This Work?" (https://newsletter.complex-machinery.com/archive/018-robot-jobs/) It shows "perfection" as the intersection of "machines do it well" and "people hate doing it". "Here be dragons" is the intersection of "people like doing it" and "machines do it poorly"
On the left, from Forbes (2025/11/23): > How Ford Is Embracing Al To Drive Innovation In The Automotive Industry (2025) On the right, from The Independent (2026/06/26) > Ford hired Al and sacked humans. It backfired badly > > "We didn't pay as much attention as we should have to the experience of our most knowledgeable engineers, says automaker
Excerpt from Complex Machinery 027: >> Tech companies, their fans, and their investors all live in the future. The companies' pitches focus on the possibilities of a product that has yet to be built. "We don't have anything yet, but we will, so we'd like our valuation to reflect that." Their investors dig this because they get paid years down the road, when the company goes public or gets acquired. And the fans, they see quirks and glitches as a small price to pay for cool functionality. All three groups are dead-set on deriving benefit Sometime Later™ . https://newsletter.complex-machinery.com/archive/027-a-difference-of-time/
My post from 2026/06/01: > Being rich and/or outspoken and/or popular does not make one's opinions facts. > > Remember this the next time someone says that we must all adopt genAl or get left behind.
Assuming you do not bounce, the best thing to do after crashing onto the road is to get clear of the next thing coming along.
The "this is the worst day of my life" meme from The Simpsons, with the text: > Bart: This is the worst day in AI > Homer: The worst day ... thus far
The header for @qethanm.bsky.social, which includes a billboard for his latest book and the following text: NEW BOOK: Twin Wolves: Balancing risk and reward to make the most of AI twinwolvesai.com research: (ML+gen)AI, risk, complexity, finance history work: workwithq.com newsletter: complex-machinery.com (some posts in 🇫🇷 🇩🇪 🇷🇺)
Excerpt from the linked article: > Right now most of AI’s 30,000-foot altitude is hype. When the hype fades—when changing the name fails to keep the field aloft—that hype dissipates. At that point you’ll have to sell based on what AI can really do, instead of a rosy, blurry picture of what might be possible. > > This is when you might remind me of the old saying: “Make hay while the sun shines.” I would agree, to a point. So long as you’re able to cash out on the AI hype, even if that means renaming the field a few more times, go ahead. But that’s a short-term plan. Long-term survival in this game means knowing when that sun will set and planning accordingly. How many more name-changes do we get? How long before regulation and consumer privacy frustrations start to chip away at the façade? How much longer will companies be able to paper over their AI-based systems’ mishaps?
Left: an excerpt from Complex Machinery 049: >> While I can't tell you which genAl companies are economic time bombs, I can assure you that each one tells a good story. It's how they entice investors and buyers. Jeff Ryan (founder of QuantKiosk and fellow die hard realist follow-the-numbers guy) summed up the genAI investment scramble better than I ever could: >> >> "It is shocking how easy it is to fool people. All of the people. All of the time." https://newsletter.complex-machinery.com/archive/049-looking-for-a-grenade-in-a-haystack/ -- Right: Futurism headline >> Insiders at SoftBank Worry Their CEO Is Getting Conned by Sam Altman https://futurism.com/artificial-intelligence/insiders-softbank-sam-altman
The WKRP "I thought turkeys could fly" meme, but with the text: "With God as my witness, I thought the customers didn't matter"
Image from Complex Machinery 018: "Should AI Do This Work?" An image with four quadrants. The top-right quadrant, marked 'PERFECTION,' is the intersection of 'Machines do it well' and 'People hate doing it'. https://newsletter.complex-machinery.com/archive/018-robot-jobs/
Excerpt from the linked article: >> The reports may throw cold water on the bets tech's biggest firms have placed on the technology. While some cling to the promise of an Al "renaissance" or "revolution," the cost of adoption is proving a stubborn bottleneck. These developments also suggest that the economics of replacing or augmenting human labor with Al may be more complicated than some early forecasts originally implied. That echoes what Bryan Catanzaro, vice president of applied deep learning at Nvidia, recently said in an interview with Axios.
Excerpt from the quote-posted article: >> Additionally, one user told The Register they had spending caps in place that should have stopped any bill over $250. Yet according to Google those caps can be automatically upgraded to $100,000 - without user input - if the user has spent a total of $1,000 throughout the life of the account, and the account is more than a month old.
Screencap from Les Echos: --- « Les développeurs deviennent des machines à consommer du token » : la facture de l'IA donne des sueurs froides aux entreprises Avec l'arrivée des modèles de raisonnement, de code et des systèmes autonomes, la consommation de l'intelligence artificielle par les entreprises augmente. Le coût à la fin du mois, lui, est de plus en plus difficile à prévoir.
Excerpt from the quote-posted article: >> Rather than a classic SRT that may be tied to dozens of loans, banks are exploring slicing and dicing large and concentrated data centre loans to shift the riskiest portions off their books, for example.
WSJ headline: >> OpenAI Wants to Go Public. First Sarah Friar Needs to Get It to Grow Up. >> >> The chief financial officer is managing Sam Altman-and ambitions for one of the biggest IPOs ever. She has pulled off impossible ones before.
Excerpt from the linked article: >> The PocketOS boss puts greater blame on Railway's architecture than on the deranged Al agent for the database's irretrievable destruction. Briefly, the cloud provider's API allows for destructive action without confirmation, it stores backups on the same volume as the source data, and "wiping a volume deletes all backups." Crane also points out that CLI tokens have blanket permissions across environments.
Excerpt from the linked article: > These simulations aren't new. What is new is their scale and fidelity. Advances in computing, reinforcement learning, and synthetic data allow machines to run millions of imagined experiments before touching the real world. A robot can learn how to walk, grasp, or balance by failing thousands of times in a simulation, where failure is cheap. When it finally acts in the real world, it does so with a plan. > > This approach has quietly unlocked progress in logistics, manufacturing, and autonomous systems. Warehouse robots navigate crowded spaces with fewer collisions (even in complete darkness). Machines adapt to unfamiliar objects instead of glitching. Autonomous vehicles rehearse edge cases long before they encounter them on the road. The critical advance is not better hardware (though that helps) but better internal models of reality.
The 'man pointing to self in mirror' meme. The text reads: 'When your AI bot goofs, and you ask to speak to its manager'.
Excerpt from the linked article: >> Manson continues, "Speaking to me years later, Cukor made no bones about it either." What was the point of all this speed if you needed to wait for cumbersome human supervision? If the machines could identify the targets, couldn't they also pull the trigger to rain death from all angles?
Excerpt from the quote-posted article: >> Where some see this sort of salvage as a business opportunity, others see a privacy concern. Marc Rotenberg, founder of the Center for Al and Digital Policy, said that even if employees signed away intellectual property rights to work materials, that doesn't settle whether employers should be allowed to sell internal communications to a third party-particularly when employees are unlikely to expect their Slack messages could be repurposed this way.
The "Miyazaki stressed" meme, from Imgflip. It depicts Hayao Miyazaki with eyes closed and hands on his head, looking frustrated.
Excerpt from the quote-posted article: >> It increasingly feels like the discovery of organic and original resources is punished by all the algorithms. And now that ChatGPT and others are going full ad-based, it is going to result in a further degradation in quality results for people.
Excerpt from Complex Machinery 018: Some very non-Al automation offers an example: the self-service kiosk. McDonald's first tried this way back in 1999. I didn't encounter one till some years later, in a tourist-area take-away joint, and I immediately saw why these things had a bright future ahead. In short: not everyone is ready for a chit-chat. Some patrons were just too fried to talk to a cashier. Others possessed only a tourist-level grasp of the local language, ill-prepared for the dizzying array of options and upsells. Both groups found interacting with a touchscreen far less stressful than talking to another person.
Except from the quote-posted article: >>Just days after those January incidents, a Waymo struck a child crossing in front of the vehicle near a school in Santa Monica, California. The child was reportedly unharmed, and Waymo later said that its models showed that a human driver would have hit the child at a higher speed than its car.
Excerpt from the quote-posted article. The highlighted portions read: >> To be counted is to be made relevant, to have shares and likes and to keep people on your profile to show your popularity. All value comes from what is counted - >> [...] >> this is foremost a "click and ragebait industrial complex" 16 Some describe
Excerpt from the linked article. The highlighted portion reads: >> Prediction markets are, at their core, a service for doing financial arbitrage on our eroded information environment, or a sort of derivative market for reality.
Excerpt from the quote-posted article: >> We can't stress enough how easy it is to find this stuff. These bots aren't the result of complex attempts to "jailbreak" Al models or confuse platforms. The platform's text filters failed to prevent them from being created, and we found them with simple keyword searches.
Excerpt from Complex Machinery 024: >> The summarizations follow the same troublesome road as a lot of AI-based summarization: the bot mixes things up, in part because it lacks much-needed context. And also because Al companies confuse "give me the gist of what's going on here" with "give me a probabilistic string of words based on this other pile of words over here."
Excerpt from the quote-posted article: >> • I over-relied on the Al agent to run Terraform commands. I treated plan, apply, and destroy as something that could be delegated. That removed the last safety layer. >> >> • I also over-relied on backups that I assumed existed. Automated backups were deleted together with the database. I had not fully tested the restore path end-to-end. >> >> • The database was too easy to delete. There were not enough protections to slow down destructive actions.
Excerpt from the quote-posted article: >> “There are also sex scenes filmed with the smart glasses – someone is wearing them having sex. That is why this is so extremely sensitive. There are cameras everywhere in our office, and you are not allowed to bring your own phones or any device that can record”, an employee says.
Excerpt from the linked article: >> “The safeguards that Anthropic is trying to keep in place are no mass surveillance of Americans and no fully autonomous weapons, which just means that they want a human in the loop if the machine is going to kill somebody,” she added. “Even if this technology were perfect — which it isn’t — I think most Americans don’t want machines that kill people without human oversight running around in an America that’s become an AI-powered mass surveillance state.”
Excerpt from linked article: >> Those who refuse to sell say the utility company has warned it may invoke eminent domain - the government power to seize private property for public use. The threat isn't empty: Dominion Energy used it against a Virginia farmer last April.
Screen cap of a Bluesky post by the Wall Street Journal. Bluesky's content moderation systems have flagged the embedded image as possible "adult content."
Excerpt from the quote-posted article: >> "You have the A.I. angst, you have the Nancy Guthrie thing happening," Mr. Siminoff said, offering his explanation as to why the Super Bowl ad created such an outcry. "All this came together and it created a perfect storm and it just hit and exploded."
Excerpt from linked article: > BuzzFeed News reported that Grubhub’s app was averaging 6,000 orders per minute during the promo, and there were widespread reports of overwhelmed restaurants simply turning off their delivery channels. Many operators said they didn’t even know the promo was happening and that they were completely under-staffed and unprepared for the lunch deluge.
Excerpt from Rebranding Data, 2021: >> Right now most of Al's 30,000-foot altitude is hype. When the hype fades-when changing the name fails to keep the field aloft-that hype dissipates. At that point you'll have to sell based on what Al can really do, instead of a rosy, blurry picture of what might be possible. >> >> This is when you might remind me of the old saying: "Make hay while the sun shines." | would agree, to a point. So long as you're able to cash out on the Al hype, even if that means renaming the field a few more times, go ahead. But that's a short-term plan. >> >> Long-term survival in this game means knowing when that sun will set and planning accordingly. How many more name-changes do we get? How long before regulation and consumer privacy frustrations start to chip away at the façade? How much longer will companies be able to paper over their Al-based systems' mishaps?
Excerpt from the quote-posted article: > It’s worth noting that Discord disclosed last October that around 70,000 users may have had sensitive data, such as their government ID photos, exposed after hackers breached a third-party vendor that the platform uses for age-related appeals. The breach reflected digital rights activists’ concerns over the use of age checks as a way to make the internet “safer.”
Excerpt from linked article: >> 4screen's proprietary tracking system measures a given ad's success by tracking parking rates. For up to three days after a push is sent out, a car parking at a given advertised location (be it a specific restaurant or gas station) is considered a success. Currently, 4screen says its ads result in a 3% to 7% boost in parking rates on average at these locations.
The "X, X everywhere" meme from Toy Story. Woody looks on, horrified, as Buzz smilingly gestures at something in the distance.
The "shut up and take my money" meme from Futurama: A man, holding a wad of cash? Has a determined look on his face.
Despite fostering online hatred, the man recorded by Wasserstrum insists he doesn't personally share the views expressed on his TikTok account. Instead, he suggests his fake anti-migrant house tour videos were just a way to game the algorithm, build an audience, and hopefully make money. "I didn't do anything because of hate," he says on the tape. "I didn't care. It's just I wanted the clicks."
Excerpt from the linked article: >> "For years, online shopping has been about keywords, filters, drop-down menus. And scrolling through multiple pages [of search results] until you find what you want," Google chief executive Sundar Pichai, who was joined on stage by Walmart's incoming boss John Furner, told his audience at the show. "Now... Al can do the hard work."
Excerpt from the linked article: >> From a high level, then, the role of the data scientist is to understand data analysis and predictive modeling, in the context of the company’s use cases and needs. It always has been. Building models was just on your plate because you were the only one around who knew how to do it. By offloading some of the model-building work to machines, autoML tools remove some of that distraction, allowing you to focus more on the data itself.
A drawing of a snake eating its tail
Excerpt from Complex Machinery #053: >> Most people would chalk this up to the pro being able to inspect the LLM's output and correct problems in the code. That certainly helps! But a typical software project involves so much more than code. There are test suites and best practices and different stages of migration. There are product people and QA teams and infrastructure specialists, the combined efforts of which create a system of checks and balances to keep everything on the rails. >> >> One under-appreciated danger of vibe coding, then, is that someone who does not embody these best practices can publish an app that only looks like it's fit for purpose. That's fine for a hobby app that has an audience of one. It becomes a problem when that app serves a business with real customers.
Excerpt from the article: > The other problem for slopagandists, of course, is that the signals they answer to are fickle - one change and the machine that pays their bills becomes a roadblock. They will decry "the algorithm" as censorial until they eventually need to pivot to celebrity commentary in order to find a new audience. When the inevitable switch-up happens — and it usually does — it will be because there's a new carrot to chase, a new kind of outrage to perform for the recommendation engine.
The "¿Por qué no los dos?" meme, which features a girl posing the question.
the Xbox "achievement unlocked" meme, with "model collapse" as additional text