Links: Sep 2026
I remember this. It was glorious.
Pon this day 6 years ago, we witnessed the greatest ever day of live tweeting
https://www.patreon.com/ShivRamdas/posts/rice-truck-74959374-
As I wrote in RFD 576, to use an LLM to write is to void the social contract between writer and reader: we readers shouldn’t be expected to labor to understand a sentence that the writer themselves didn’t work to create.
-
Readers can detect LLM words in the parts per trillion. However much work you put into scuffing up and humanizing it, an LLM paragraph will register to much of your audience not as writing but as output. So, first the bad news: you have to write for yourself.
But LLMs are still extraordinarily useful. It’s just you need to use them like a copyeditor rather than a ghostwriter. So, step one of my method: write your piece. Then, step two: feed it to a good model to find flaws.
But before we talk about how that works, there are two rules you need to understand. They’ll ward off LLM-creep that will knock you into the uncanny valley between expression and output and knock you out of your reader’s attention.
The online shopping trend where you buy nothing
On South Korea’s “dopamine sites,” the pleasure comes from aspirational shopping rituals like browsing, curating, and tracking — not from a delivery.
-
Consider nuclear power. Overnight construction costs for early U.S. demonstration reactors fell 81 percent between 1954 and 1968. For reactors begun between 1967 and 1972, costs rose 187 percent. The 1971 Calvert Cliffs decision and then the reaction to Three Mile Island in 1979 accelerated the cost increases and slowed construction even more.
What might have happened had the earlier learning and deployment trends continued? Peter Lang’s 2017 paper in Energies estimates that nuclear power would have cost about one-tenth as much by 2015. The additional generation could have avoided as many as 9.5 million premature deaths.
-
I could have (and perhaps would have?) taken this whole shameful incident to the grave, but I feel compelled to confess it now, because I have never seen fear sown so irresponsibly by putative technologists as I have now with respect to AI and the threat of human extinction (!). Amazingly, that is not hyperbole: this week, ex-Anthropic employee Jacob Coxon claimed — and then Anthropic Alignment Science lead Evan Hubinger agreed — that the probability that AI will "kill all humans" is ">10% in the next decade" (!!).
- AN
The loudest voices stoking fears about AI dangers have made tremendous headway in the past two weeks. AI technology has not taken some unexpected, dangerous turn, but the hype around it — propelled by what appears to be a well orchestrated PR campaign — has drummed up considerable fear. I worry that it represents a setback for our field.
I have written frequently that fears of AI are overhyped. AI’s capabilities can be uncannily human-like and unpredictable, and it’s rational to worry when people who are directly involved express concerns. But I see the problems as a sign of the engineering work that ahead, rather than insurmountable barriers or the sky falling. AI technology continues to advance — which is a good thing! — but technical advances, poorly understood by the public, give those who seek to generate hype repeated opportunities to do so.
First, I don’t see any step up in the risk of human extinction from AI compared to a few months ago. The theories about this remain the same fantastical, science fiction scenarios as a few months ago. The biggest change in AI risk is its cybersecurity capabilities — a topic which we should take seriously — but this, too, will not lead to the end of the world.
The most notable recent event leading to increased fear was when an OpenAI team deployed an agent swarm that hacked into Hugging Face. Much of the popular press contained significant hype. For example, some publications reported that a swarm of 1,200 agents carried out the attack. While this was technically accurate, as I write this, I have about 1,300 processes running on my laptop. Yes, the ability to get large swarms of agents to work in parallel on a task is a significant technical advance, And, in computing, many processes run at the same time. So this shouldn’t be seen as some magical capability.
Additionally, OpenAI’s buggy sandboxing and monitoring processes were key to enabling this incident. Fixing these bugs and putting in place improved monitoring would be appropriate fixes, not pausing AI. There are many well known ways to attack software systems. The main advantage of AI agents is that they are relentless. They will tirelessly try many tactics — and have the patience to chain vulnerabilities together — that previously would have taken an infeasible amount of human effort. But in the long term, I believe the advantage will lie with defenders (because they have more information with which to identify bugs, which they can fix), but the cyber-threat landscape has changed significantly. There are still bottlenecks to identifying and exploiting a vulnerability. AI agents still have to try a lot of things to see what works, and taking these actions takes time and might be detected by defenders. This is why, even though it is now easy to obtain versions of leading open weight models that have had their guardrails removed or weakened, so they will not refuse to try to execute cyber attacks, the world has not ended.
I am also concerned about the anthropomorphization of AI in a lot of reporting, where LLMs and agents are unnecessarily treated as if they were people. If I wield a hammer, miss a nail, and accidentally dent the wall, it’s not the fault of the hammer. The problem lies in how I used the hammer. Similarly, if I prompt an agent and it hacks into someone else’s system, the responsibility lies with me, not the agent.
Of course, we want to build systems that are as safe and predictable as possible. (For example, an unsafe hammer would be one whose head randomly flies off under normal use.) Today’s agentic systems are not predictable, but I see no reason why, by applying sound engineering practices, we won’t be able to make them extremely safe to use. One new element in the forecasts of AI-enabled doom is AI companies disclaiming responsibility for their own products. “I didn’t do it; my out-of-control agent did!” There’s a balance to be struck between the responsibility of the tool maker and the tool user, but when something goes wrong, let’s hold the people building and/or using the hammer responsible, rather than the hammer. (By the way, if you’re worried about AI bioweapon risk, David Bellamy has a great post on why this, too, is overhyped. Briefly, the bottleneck in building a bioweapon is not intelligence, but lab work and manufacturing.)
Pausing AI progress will create much more harm than benefit. First, our adversaries will certainly not slow down. Second, engineering requires discovering problems empirically so we can fix them. If we pause AI by a decade, we will also delay finding and implementing safety engineering fixes by about the same duration.
Of course, the incentive to stoke fears — for regulatory capture, to garner attention, or to make one’s technology seem more powerful — remains the same as before. Disclaiming responsibility is a new one. Taking a hard technical look at the actual risks however, I see little factual basis for the degree of fear that’s been stoked up. We still have hard research and engineering work ahead to improve AI safety, but the beneficial applications continue to vastly outweigh the risks, and we should keep building.
[Original text (with links): The Tragedy of Nikole Hannah-Jones
The essay makes for hard reading and I acknowledge NHJ’s courage in writing about her mistakes and the cost to her child. The piece is raw and true when talking about her personal choices, but the larger story in which NHJ embeds her personal history is mostly false. The problem isn’t funding. The failures she documents are failures of management and accountability, not lack of resources. NYC spends upwards of $40k per pupil on average, the highest of any large district in the country. Moreover, schools in NYC with lower-income students get more than the average. PS 307, where NHJ initially sent her child, is currently funded at just over 52k per student, far above the national average.
Selective breeding has pushed horses to the limits of biology. But at what cost?
In June 1973, chestnut colt Secretariat ran the Belmont Stakes so fast that no horse has matched him since. In less than two and a half minutes, Secretariat flew along the track, finally pulling away to win by 31 lengths, nearly three quarters of a football field. The other horses were so far behind they weren’t even in the television frame. Fifty years later, billions of dollars have been poured into breeding and training, yet Secretariat’s record still stands.
The Odyssey, AGI and the bicameral mind
Yet in his 1977 book, The Origin of Consciousness and the Breakdown of the Bicameral Mind, Julian Jaynes argued that the two epics were divided in time by a change in the human brain. In The Iliad, people heard gods talking to them directly, instructing them and guiding them. In The Odyssey, the gods have gone into hiding and humans are guided by their own consciousness. It’s worth quoting a passage from Jaynes here.
and
Reading Jaynes’ words today, it is almost as if the heroes of The Iliad were AI agents being trained and instructed by their human masters, all the way up to the godlike personages of Sam Altman and Dario Amodei. We will get to them shortly, but first, The Odyssey. As Jaynes suggests, shortly before 1000 BC, something called the Dorian invasion shattered the Mycaenean civilisations around the eastern Mediterranean. Nolan picks up on this idea in his film, with references to mysterious ‘people from the sea’. This social breakdown somehow rewired the brain, argues Jaynes. The voices in peoples’ heads receded and consciousness took its place.
- BI
Interesting NYT story about the most popular Uber Eats delivery restaurant in the world - an Indian restaurant in a Houston strip mall. Some of the drivers only deliver from there,
Reality has a surprising amount of detail
That’s how I came to spend a substantial part of my teenage years replacing fences, digging trenches, and building flooring and sheds. And if there’s one thing I’ve learned from all this building, it’s that reality has a surprising amount of detail.
This turns out to explain why its so easy for people to end up intellectually stuck. Even when they’re literally the best in the world in their field.
Links resume after more than a month. There was a big backlog, but as I skimmed through them, I realized that many of them didn't age well. After thinking about always posting with a lag or shifting to a monthly posting schedule, I am shifting to a monthly schedule. Hopefully this results in fewer, higher quality links. The risk is I forget to post altogether. So I am going to keep a weekly cadence of posting but only monthly for links. The remaining three weeks must be filled with something else.