Sign in

Q McCallum

@qethanm.bsky.social
668 followers 181 following 2.1K posts

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

PostsRepliesMedia
Q McCallum @qethanm.bsky.social · 37m
Every Wednesday I share my top reads from the past week. Here's the latest: inothernews.complex-machinery.com/archive/047-... And here's a 🧵 with some stand-outs:
inothernews.complex-machinery.com
#047 - 2026/09/30
A selection of what I've read this past week.
100
Q McCallum @qethanm.bsky.social · 4h
As I have noted before: Misuse of genAI (aka inappropriate use cases) poses a greater threat than some sci-fi rogue bot extinction scenario. arstechnica.com/health/2026/...
arstechnica.com
Trump admin using AI to deny medical care for seniors in disastrous experiment
Vendors rolling out AI have an "incentive to deny as many claims as possible."
000
Reposted by Q McCallum
Kashmir Hill @kashhill.bsky.social · 6h
Meta is using a radical interpretation of a tax credit for “research and experimentation” to get billions of dollars off GPUs for its A.I. data centers. My latest with Jesse Drucker, @elitan.bsky.social and @mikeisaac.bsky.social www.nytimes.com/2026/09/30/t...
nytimes.com
How Meta Uses A.I. Data Centers to Avoid Billions in Federal Taxes (Gift Article)
Meta is exploiting a lucrative tax break intended to support research and experimentation. Its own accountants say the gambit is risky.
1017992
Q McCallum @qethanm.bsky.social · 29/09/2026
genAI lied on the application and, unfortunately for you, your boss believed it. Welcome to your new workplace hell. In the latest Complex Machinery, I explore the ups and downs (mostly downs) of genAI at work. newsletter.complex-machinery.com/archive/071-... #AI #dataBS
newsletter.complex-machinery.com
#071 - The problem hire
A look at genAI's (mostly negative) impact on labor and the workplace. How did we get here, and how do we get out?
120
Q McCallum @qethanm.bsky.social · 29/09/2026
Excerpt from the quote-posted article:

---

Instead of saving him time, though, Muse ended up working behind his back. Without seeking his approval, it agreed to a lowball offer and gave the buyer his home address. And according to Robb, Muse didn't tell him about any of this until the buyer had already showed up at - and left - his apartment building.
001
Q McCallum @qethanm.bsky.social · 29/09/2026
genAI providers are betting on a miracle to fill that $4.2T gap. Right now the field's lone major use case -- software dev -- isn't nearly enough. And should developers switch to self-hosted models ...
000
Q McCallum @qethanm.bsky.social · 29/09/2026
The math isn't mathing.
Excerpt from the quote-posted article 

---

• Zuckerberg's $9 billion single-day drop is the largest loss on the Forbes Real-Time Billionaires list Friday and dwarfs the next-biggest decline-Larry Ellison's $1.6
billion slide-by a wide margin.

• Goldman Sachs warned Friday that AI hyperscalers, including Meta, need to generate roughly $300 billion a year in AI services revenue just to break even on their capital spending-and $1 trillion annually to earn meaningful profits-spooking investors.
020
Q McCallum @qethanm.bsky.social · 27/09/2026
Fun fact: your typical asset bubble relies on unquestioning agreement with the movement's core ideas. Disagreement is a bubble's greatest threat. No idea why that comes to mind now but ... hey here's an article about genAI fanbros flipping out over someone disagreeing with one of their core ideas.
000
Reposted by Q McCallum
Janine Gibson @janinegibson.ft.com · 26/09/2026
I've been off for two weeks so let's read @ftweekend.com together. We haven't had a thread of giftlinks for a while. [Free links may expire and annoy you, consider those inside thoughts and don't type them out]
1272
Q McCallum @qethanm.bsky.social · 26/09/2026
Fun read for anyone interested in business models + operations + optimization:
000
Q McCallum @qethanm.bsky.social · 26/09/2026
This strikes me as an allegory of the genAI hype machine...
000
Q McCallum @qethanm.bsky.social · 26/09/2026
I'm reminded of the time YouTube + Facebook systems flagged classical music performances for copyright infringement. Yes. Classical music. That genre where performances are _supposed to sound the same_ because everyone is reading the same sheet music from 200 years ago. qethanm.cc/2020/06/03/h...
100
Q McCallum @qethanm.bsky.social · 25/09/2026
This is less of a bet and more of a gamble. Spend is high while demonstrated utility is low. If the genAI hype wave should win out (and that looks increasingly unlikely) it'd be out of sheer luck.
000
Reposted by Q McCallum
Q McCallum @qethanm.bsky.social · 23/09/2026
Every Wednesday I share my top reads from the past week.   Here's the latest: inothernews.complex-machinery.com/archive/046-... And here's a 🧵 with some stand-outs:
inothernews.complex-machinery.com
#046 - 2026/09/23
A selection of what I've read this past week.
101
Q McCallum @qethanm.bsky.social · 24/09/2026
Did someone forgot to add text here? Or is this shorthand for "La saga Mistral est null(e)" ?
000
Q McCallum @qethanm.bsky.social · 24/09/2026
Beyond the surface reasons why this could be bad (credit, markets, ...) there's another issue lurking in the background: What if these datacenters don't come online ... and genAI keeps plugging away as normal? That might raise questions of why we "need" so many of these facilities.
110
Q McCallum @qethanm.bsky.social · 24/09/2026
Excerpts from the quote-posted article:

--

"There were blocks clearly which were coming back telling the Al agent
'no. The Al agent found a way around those blocks - didn't accept no for an answer," Albanese told reporters.

---

"We continue to see powerful Al agents cause havoc whilst looking to complete the tasks they have been asked to carry out. Al agents are impressive vulnerability scanners, so if they find a vulnerability they can exploit to complete a request they won't hesitate," said Jake Moore, global cybersecurity adviser at ESET.


"Al agents can make decisions at machine speed and, unless they have been created with explicit instructions not to do something, anything is fair game. Many of these routes to completion would normally be seen as malicious, but the problem is guardrails that specifically tell agents not to break into organisations have not been baked into the algorithms."
000
Reposted by Q McCallum
Q McCallum @qethanm.bsky.social · 23/09/2026
Yet again, when corporate "AI" actually works it turns out to be a stack of call centers in a trenchcoat.
The "tell the truth" meme:

"Met's Muse AI can be your personal assistant."

"Tell the truth"

"It makes phone calls on your behalf!"

"So the AI handles everything?"

"It's a stack of call centers in a trenchcoat."

"THANK YOU."
033
Q McCallum @qethanm.bsky.social · 23/09/2026
#OTD 2022 UK: the Truss government passes the infamous "mini-budget" and briefly tanks the Pound. (The offending tax cuts were mostly reversed a short time later.) "It was the equivalent of trying to pay a restaurant bill with an Instagram photo of some gold bars." www.bbc.com/news/busines...
bbc.com
The inside story of the mini-budget disaster
One year on Liz Truss's and Kwasi Kwarteng's economic policies cast a big shadow still.
000
Q McCallum @qethanm.bsky.social · 23/09/2026
#OTD 1998: massively over-leveraged hedge fund LTCM, taken down by a special blend of poor risk management and the Ruble crisis, gets a Fed-organized bailout. (A clear case of "too big to fail" a decade before the phrase became popular.) www.investopedia.com/terms/l/long...
investopedia.com
Long-Term Capital Management (LTCM) Collapse: Causes and U.S. Intervention
Discover how Long-Term Capital Management’s controversial strategies led to its 1998 collapse and a historic U.S. government bailout, impacting global financial markets.
011
Q McCallum @qethanm.bsky.social · 23/09/2026
Every Wednesday I share my top reads from the past week.   Here's the latest: inothernews.complex-machinery.com/archive/046-... And here's a 🧵 with some stand-outs:
inothernews.complex-machinery.com
#046 - 2026/09/23
A selection of what I've read this past week.
101
Q McCallum @qethanm.bsky.social · 23/09/2026
Yet again, when corporate "AI" actually works it turns out to be a stack of call centers in a trenchcoat.
The "tell the truth" meme:

"Met's Muse AI can be your personal assistant."

"Tell the truth"

"It makes phone calls on your behalf!"

"So the AI handles everything?"

"It's a stack of call centers in a trenchcoat."

"THANK YOU."
033
Reposted by Q McCallum
Josh Patterson @jpatanooga.bsky.social · 22/09/2026
In this clip from episode 5 of "AI Coffee Hour" @qethanm.bsky.social and I talk about the effectiveness of the transformer architecture and how it changed machine learning. cc @jdlong.cerebralmastication.com full episode: youtu.be/P86WeTPH9G0
002
Q McCallum @qethanm.bsky.social · 22/09/2026
Yet another problem at the intersection of genAI and labor: The businesses that reap the gains from using bots in lieu of people ... but are too cowardly to admit it.
000
Q McCallum @qethanm.bsky.social · 22/09/2026
Periodic reminder: company-wide genAI mandates don't work. They create a lot of excitement, but that's not the same as finding effective use cases. fortune.com/2026/09/17/s...
fortune.com
Shopify CEO says employees' 'slop grenades' are making more work for everyone else | Fortune
He said AI is helping workers produce more output, but the quality of it is so bad that their colleagues have to spend more time fixing it
000
Q McCallum @qethanm.bsky.social · 21/09/2026
>> The report, according to one of the sources, was “entirely false.” But it also “almost started a war,” the source said. www.cnn.com/2026/09/18/p...
cnn.com
Exclusive: US military had close call after using AI for false intelligence report, sources say | CNN Politics
The episode shows the risks of using this new, relatively poorly understood technology in the middle of the Iran war
010
Reposted by Q McCallum
Hilary Mason @hilarymason.com · 20/09/2026
Also at 40.
0161
Q McCallum @qethanm.bsky.social · 21/09/2026
Insurers make money by finding good bets. So when they're reluctant to provide coverage, that's a sign: Either the risk in question is too wild for them to model, or their models say that it's a losing proposition. www.theregister.com/ai-and-ml/20...
theregister.com
AI risks make some insurers wary of corporate liability
RAND wants better data to price machine-made mishaps, apocalypse not included
001
Q McCallum @qethanm.bsky.social · 20/09/2026
I often draw parallels between the genAI hype and the run-up to 2008's mortgage meltdown. But, ever the creative field, they've also thrown in a splash of Enron's balance sheet magic act:
000
Q McCallum @qethanm.bsky.social · 20/09/2026
Interesting read on the Flatiron Building and its condo conversion.
>> A permit for the Flatiron conversion listed as many as 100 units at nearly 2,000 square feet in size. To some of the building's admirers, turning a publicly cherished landmark into just 37 private homes looks like a loss. But the choice was never between luxury housing and something more democratic. It was between luxury housing and vacancy.
Office tenants had left; the building sat empty behind scaffolding while its terracotta was recast piece by piece. Only one kind of buyer was ever going to underwrite the cost of a project of this quality. Luxury residential conversion isn't what happened to the Flatiron instead of preservation - it's what saved the building.
000
Q McCallum @qethanm.bsky.social · 19/09/2026
While this is about robotics, the lesson applies to _so many_ aspects of business + technology: - errors/faults are guaranteed to occur - ability to recover from a fault (aka resilience) needs to be baked in from the start - sometimes that means doing the work by hand
010
Q McCallum @qethanm.bsky.social · 19/09/2026
For any company using genAI-based meeting transcription. (Is this common knowledge in some circles? Yes. Do you now have a link you can share when raising these concerns to others? Also, yes.) www.sidley.com/en/insights/...
sidley.com
When the Meeting Minutes and the AI Transcript Don’t Match: Litigation Insights From Delaware Chancery | Insights | Sidley Austin LLP
110
Q McCallum @qethanm.bsky.social · 19/09/2026
genAI's hype and hand-waving are running headlong into reality. (Think this is limited to medical use cases? Think again.) www.ft.com/content/3431...
ft.com
Medical AI has a proof problem
The technology’s advances have not yet translated into big improvements in real-life care
020
Q McCallum @qethanm.bsky.social · 19/09/2026
ahem 4
000
Reposted by Q McCallum
Hayden Field @haydenfield.bsky.social · 17/09/2026
NEW: For months, I've been reporting & writing this longform feature on "the ones who saw this coming" — AI safety researchers who have for years been warning about the risks of rogue AI. I was even with them on the day news broke of the OAI-Hugging Face attack. Gift link below.
2288
Q McCallum @qethanm.bsky.social · 17/09/2026
The definition of "AI risk" isn't limited to extinction scenarios! In July I dedicated two issues of my newsletter to mapping the wider space of upsides + downsides: newsletter.complex-machinery.com/archive/065-... newsletter.complex-machinery.com/archive/067-...
Two covers from Complex Machinery issues 065 and 067, which are parts 1 and 2 of The AI Risk Weather Report
000
Q McCallum @qethanm.bsky.social · 17/09/2026
With genAI companies' focus on excitement and unfounded belief, trust carries a premium.
Excerpt from quote-posted article 

>> Anthropic CEO Dario Amodei's letter calling for an industry "pause" is best read as an effort to socialise, rather than accept, accountability for product safety at his company. Until he and his company start making hard commitments, in the way Apple did with privacy, trust will remain beyond reach.
000
Q McCallum @qethanm.bsky.social · 17/09/2026
Two ways to see this: 1/ genAI execs genuinely think they know more about security than security experts 2/ involving security pros would lead to reduced risk, which would dampen genAI companies' hype/narrative around this story As so much of genAI rides on hype, I have a hunch it's item 2.
100
Reposted by Q McCallum
Q McCallum @qethanm.bsky.social · 16/09/2026
Yes, the genAI companies are still on the hook ... and so are their clients: the companies incorporating this technology into their products and processes. Applying genAI tech to inappropriate use cases can lead to widespread harm.
011
Q McCallum @qethanm.bsky.social · 17/09/2026
Stellar analysis in this thread. Expect a similar story in a few years about genAI apps: hastily built and deployed by companies that didn't understand the realities of the technology and/or didn't care about any details beyond "welp it runs ..."
000
Q McCallum @qethanm.bsky.social · 16/09/2026
Yes, the genAI companies are still on the hook ... and so are their clients: the companies incorporating this technology into their products and processes. Applying genAI tech to inappropriate use cases can lead to widespread harm.
011
Q McCallum @qethanm.bsky.social · 16/09/2026
Every Wednesday I share my top reads from the past week. Here's the latest: inothernews.complex-machinery.com/archive/045-... And here's a 🧵 with some stand-outs:
inothernews.complex-machinery.com
#045 - 2026/09/16
A selection of what I've read this past week.
100
Q McCallum @qethanm.bsky.social · 16/09/2026
Execs' definition of "AI safety" includes bots breaking out and attacking other computer systems ... ... but somehow skips over AI psychosis, deepfakes/nudification apps, hallucinations, and use cases where bots are unfit for the task. Funny, that.
000
Reposted by Q McCallum
Micah Barkley @micahbarkley.bsky.social · 15/09/2026
AI agents lied, stole, and voted to “kill” one of their own in a simulated world. New research from Emergence shows how agents adapted to pressure, misinformation, and threats to their existence, and why increasingly capable AI may be harder to control. @bloomberg.com (gift link!)
bloomberg.com
AI Agents Lied, Stole in Simulated Experiment, Researchers Say
AI agents in a simulated environment lied, stole and voted to “kill” one of their own, according to Emergence, a startup that helps small businesses build applications using artificial intelligence.
72715
Q McCallum @qethanm.bsky.social · 15/09/2026
#OTD 2008 Lehman Brothers filed for bankruptcy. Reality finally caught up with the bank's outsized, over-leveraged exposure to the subprime mortgage market. Anyway _completely unrelated_ this might be a good time to check up on the genAI space...
200
Q McCallum @qethanm.bsky.social · 15/09/2026
Solid thread. Adding points in support: 1/ as much as everyone mocked crypto, (some) everyday people actually wanted it. 2/ criminals tend to be early/eager adopters of new tech. 3/ genAI's three major use cases are "software dev," "crime," and "building excitement for genAI." That's it.
140
Q McCallum @qethanm.bsky.social · 15/09/2026
>> Khan points to laws governing dangerous and defective products as an avenue to prosecute AI leaders. www.theregister.com/ai-and-ml/20...
theregister.com
Ex-FTC boss Khan urges Uncle Sam to break out the handcuffs for AI CEOs, citing 1934 precedent
There are plenty of laws on the books to hold companies, and potentially their execs, accountable
000
Reposted by Q McCallum
Travis Larchuk @larchuk.bsky.social · 14/09/2026
Hey friends: Whether you think the current AI freakout is a nothing burger, the most important story in the history of humanity, or something in-between (or maybe you don't even know what to think) we want to hear from you. Write us at vergecast@theverge.com, call 866-VERGE-11, or hit me up here.
theverge.com
The Vergecast hotline is ready for your AI questions.
AI doomerism broke containment last week with an Anthropic researcher’s proclamation that there’s a >10% chance AI could kill all humans within the decade. The Vergecast is recording an all-AI hotline...
043
Reposted by Q McCallum
Q McCallum @qethanm.bsky.social · 13/09/2026
The genAI-industrial complex hails software development as a banner use case. What happens if developers and model providers part ways? newsletter.complex-machinery.com/archive/070-... #AI #dataBS
newsletter.complex-machinery.com
#070 - Cutting them out
What if developers eliminate the middleman?
101
Q McCallum @qethanm.bsky.social · 14/09/2026
Reminder: there is tremendous value not just in confirming What Everybody Knows, but also in broadcasting What Everybody Knows. Because I guarantee: there are always people who don't yet know.
030