The question we should continually insist on asking ourselves, both as makers of tools and users of them, is how they can help us improve the flow of love. Do they connect us with each other, and reinforce our instincts to treat each other as moral ends? Do they help us convey our love out into the world through information and interfaces, or help us know when we are being offered any?

I get what he’s saying, and there’s some nice prose here, but to me it seems like a whole lot of jibber-jabber. AI is making even the smartest people in my blogroll a bit loopy.

It’s been an interesting few weeks for counterexamples. This post is basically my perspective of what has been going on in the world of formalization, AI tools and, in particular, counterexamples.

Crazy shit, dog.

Pretty much every organization finds it easy to reward “productivity” as expressed by lines of code emitted, but the benefits of doing thorough and thoughtful design, analysis, and code review very difficult to reward. In this world, an LLM is an invaluable tool for the sociopathic ladder-climber, particularly if your legacy organization is still structuring their workflows as if the person prompting the bot is “writing” the code, and then they get to foist off the act of “reviewing” the code onto someone else. / Here, the prompter effectively externalizes the cost of the LLM’s failures but internalizes any benefits.

When a student is asked to write an essay, the teachers see the effort as both intrinsically worthwhile for the student, as well as useful as a pedagogical tool to evaluate and react to the student’s progress. The student, by contrast, sees a stumbling block designed to knock them off the path to success and into a permanent underclass.

via: https://liamoc.net/forest/2026-W26/index.xml

With reference to SNAP as an illustration

I’m a software engineer and (reluctant) policy wonk with an interest in complex human problems, often having to do with institutions.

Met him at an event he hosted in summer 2023 while he was doing an Interact Fellowship. He stopped posting regularly to his blog that summer, and now irregularly writes a newsletter. After moving between various gigs, he’s now at Anthropic. Damn, does this company swallow up everyone.

see: https://daveguarino.substack.com/

Moral conviction predicts not only a desire for distance but a greater acceptance of violence as a way to resolve conflict. I want to leave that thought hanging over everything that follows—because resistance to AI has already begun producing more than op-eds and think-pieces. It has produced threats, break-ins, and worse. Quietly, moralization sorts a society into camps that would rather not share a room—or a planet. And we already have another name for that: polarization.

I understand where the author is coming from, but I really feel like this article reveals why consequentialism struggles to work in practice. At some point, it actually becomes inhumane to force the talking out of problems and to make value judgements. Morality at some point must be brought into the picture when we are making sweeping decisions about the lives of millions with an extreme lack of consent from the majority.

see: https://osf.io/preprints/psyarxiv/5mwre_v11

Agents can write code faster than we can absorb it. Here’s why it still matters for humans to understand what they build — and some techniques for doing that efficiently: explainer docs, quizzes, micro-worlds, and shared spaces.

You get HTML, trackers, cookie banners, popups, and JavaScript. This needs a browser built by thousands of engineers just to show you a paragraph.

Machines get the clean version.

see: https://blog.cloudflare.com/markdown-for-agents/

How Rust prevents data races, and how languages like it could influence future code generation.

LLMs are built to automate a complex web of reification. They are designed to tell us what we want to hear. As we write, “Trained on an enormous corpus of human writing, speech, and code, and tuned to refine responses around context and memory as user interactions unroll, a model of this kind is designed to provide the sense that one’s expectation is being exceeded.”

Artificial intelligence is a bizarre technology that participates in its own mythmaking.

see: https://www.theideasletter.org/essay/reify-this/

Clever Hans was a horse that could do maths, or was it using some other trick? Is AI music classification working like a ‘Clever Hans?

see: https://arxiv.org/abs/2601.16675

For however futuristic the new genies seem, there is never actually progress in genies. Magic is not a measurement of their capability, but a description of our irresponsibility. Effortless wish-fulfillment is a perpetual hypothetical, no matter how perilously it is ever approximated.

I want you to explore. I want you to want to explore. If you walk into the darkness with a lamp, you will come upon all sorts of things you did not already know were there. One of them might be the thing you would have told the genie to fetch, but the more you explore the darkness, the better you will understand what else you might want from it, or from yourself. Everything you do changes you: exploring cultivates your curiosity and discernment and perspective and humility; settling for wish-fulfillment impoverishes your experience and then even your wishes.

I don’t know what will happen to us. I don’t trust the people in charge of the future, but humanity isn’t going to change any time soon. I refuse to be incurious about magic just because we can’t be trusted with it.

How the Midjourney ultrasound project started a war between medicine and tech.

I side with doctors on this one, though for a reason that they didn’t seem willing to advance. We shouldn’t obsessively overmedicalize ourselves. There is an inherent harm in always seeing yourself as a patient, ailing unless proven healthy. More information can make you sicker.

What lets the same algorithm work across all three domains is that they share two key properties:

1. Closure under negation and intersection — so the difference between two candidates is itself expressible.
2. Sampling from that difference — so the system can show the user concrete cases where the candidates disagree.

This is just Ought’s Factored Cognition paradigm again, just now contrasted with the currently popular agent paradigm. A bit disappointed that he didn’t cite prior art.

Apparently, he’s now doing AI systems research at a showy VC fund? :/

How do we navigate the world of LLMs, and how they understand (or rather, don’t) the concept of truth? How can we use them for what they’re good for, rather than throwing them at every problem? Let’s discuss ways of constraining them, and ways to improve how they’re used.

via: https://www.aeracode.org/2026/06/01/constraining-llms/

You probably recognise these things as good, but a surprising number of managers seem to think they can just chastise people until quality improves. They talk themselves into this because they believe line workers are fully responsible for their actions.

But even those managers will find it very hard to convince themselves quality improves when they scream at the code-generating robot. It’s a robot! It can’t be responsible for its actions. We have to adopt the lean philosophy for building systems around robots. When something goes wrong, we have to blame the process, not the robot.

We always had to do that, even with people, but with robots it’s painfully obvious.

“We show that a tiny snippet—just 13 words—of retrieved text on a UGC website like Reddit, Wikipedia, Quora, or Facebook can change AI agents to output spam / scam content pretty consistently.”

“I think based on the comment content itself, it’s just hard to distinguish between the poisoned text and an actual user’s text,” Zhang said. “Let’s say if you want to find the best restaurant, it could be possible that some [human] users post about good restaurants—you can’t really say [as a moderator] ‘You cannot post this comment because it’ll poison an LLM.’”

see: https://arxiv.org/abs/2605.24245