Quotes

When I was curating my generated tweets, I estimated 30-40% of the tweets were usable comedically, a massive improvement over the 5-10% usability from my GPT-2 tweet generation. However, a 30-40% success rate implies a 60-70% failure rate, which is patently unsuitable for a production application.

The most ignorant are the most conceited. Unless a man knows that there is something more to be known, his inference is, of course, that he knows every thing. Such a man always usurps the throne of universal knowledge, and assumes the right of deciding all possible questions. We all know that a conceited dunce will decide questions extemporaneous which would puzzle a college of philosophers, or a bench of judges. Ignorant and shallow-minded men do not see far enough to see the difficulty. But let a man know that there are things to be known, of which he is ignorant, and it is so much carved out of his domain of universal knowledge. And for all purposes of individual character, as well as of social usefulness, it is quite as important for a man to know the extent of his own ignorance as it is to know any thing else. To know how much there is that we do not know, is one of the most valuable parts of our attainments; for such knowledge becomes both a lesson of humility and a stimulus to exertion.

Try looking at your mind as a wayward puppy that you are trying to paper train. You don't drop-kick a puppy into the neighbor's yard every time it pisses on the floor. You just keep bringing it back to the newspaper. So I keep trying to gently bring my mind back to what is really there to be seen, maybe to be seen and noted with a kind of reverence. Because if I don't learn to do this, I think I'll keep getting things wrong.

The Perl community has a long-standing love/hate-affair with making changes that impose "spooky action at a distance". They call it "black magic" and it is generally considered it a last resort. Black Magic that makes GLOBAL changes to things like inheritance is often characterised as being "Octarine" (see disk world novels), because it tends to work ok when there's only one person doing it, but start to mix a few together and KABOOM!

At the dark doorways they dinned and hammered; there was clang of swords and crash of axes. The smiths of battle smote the anvils; sparked and splintered spears and helmets. In they hacked them, out they hurled them, bears assailing, boars defending. Stones and stairways streamed and darkened; day came dimly—the doors were held.

Nego consequentiam! I don’t see at all why I should have respect for lies and frauds because other people are stupid. I respect truth everywhere, and it is precisely for that reason that I cannot respect anything that is opposed to it. My maxim is, Vigeat veritas, et pereat mundus, the same as the lawyer’s Fiat justitia, et pereat mundus. Every profession ought to have an analogous device.

Grandma’s secret cake recipe, passed down generation to generation, could be literally passed down: a flat slab of beige ooze kept in a battered pan, DNA-spliced and perfected by guided evolution by her own deft and ancient hands, a roiling wet mass of engineered microbes that slowly scabs over with delicious sponge cake, a delectable crust to be sliced once a week and enjoyed still warm with creme and spoons of pirated jam.

Given a week or two to try out ideas and search the literature, I’m pretty sure that Freek and I could’ve solved this problem ourselves. Instead, though, I simply asked GPT5-Thinking. After five minutes, it gave me something confident, plausible-looking, and (I could tell) wrong. But rather than laughing at the silly AI like a skeptic might do, I told GPT5 how I knew it was wrong. It thought some more, apologized, and tried again, and gave me something better. So it went for a few iterations, much like interacting with a grad student or colleague. [...] Now, in September 2025, I’m here to tell you that AI has finally come for what my experience tells me is the most quintessentially human of all human intellectual activities: namely, proving oracle separations between quantum complexity classes. Right now, it almost certainly can’t write the whole research paper (at least if you want it to be correct and good), but it can help you get unstuck if you otherwise know what you’re doing, which you might call a sweet spot.

What a peculiar pleasure it affords us to see any free animal looking after its own welfare unhindered, finding its food, or taking care of its young, or associating with others of its kind, and so on! This is exactly what ought to be and can be. Be it only a bird, I can look at it for some time with a feeling of pleasure; nay, a water-rat or a frog, and with still greater pleasure a hedgehog, a weazel, a roe, or a deer. The contemplation of animals delights us so much, principally because we see in them our own existence very much simplified.

I definitely think there is great art out there that was solely designed to give people what they want; in film, someone like [Charlie] Chaplin comes to mind. I mean, giving people what they want is an art unto itself, but I think the real challenge in that method is finding a way to give them what they want while giving them more.

In the official timeline, Peppa is appropriately reassured by a kindly dentist. In the version above, she is basically tortured, before turning into a series of Iron Man robots and performing the Learn Colours dance. A search for “peppa pig dentist” returns the above video on the front page, and it only gets worse from here.

But the intellectually interesting part for me is something else. I now have something close to a magic box where I throw in a question and a first answer comes back basically for free, in terms of human effort. Before this, the way I'd explore a new idea is to either clumsily put something together myself or ask a student to run something short for signal, and if it's there, we’d go deeper. That quick signal step, i.e., finding out if a question has any meat to it, is what I can now do without taking up anyone else's time. It’s now between just me, Claude Code, and a few days of GPU time. I don’t know what this means for how we do research long term. I don’t think anyone does yet. But the distance between a question and a first answer just got very small.

CSRF is not a security issue for the Web. A well-designed Web service should be capable of receiving requests directed by any host, by design, with appropriate authentication where needed. If browsers create a security issue because they allow scripts to automatically direct requests with stored security credentials onto third-party sites, without any user intervention/configuration, then the obvious fix is within the browser.

we cannot simply presume that we know instinctively why people do what they do no matter how emotionally satisfying that may be, because humans are often generally unaware of the reasons for their thoughts and actions in the first place...In most cases, our thoughts and actions simply make sense at the time.

When you start a creative project but don’t finish, the experience drags you down. Worst of all is when you never decisively abandon a project, instead allowing it to fade into forgetfulness. The fades add up; they become a gloomy haze that whispers, you’re not the kind of person who DOES things. When you start and finish, by contrast — and it can be a project of any scope: a 24-hour comic, a one-page short story, truly anything — it is powerful fuel that goes straight back into the tank. When a project is finished, it exits the realm of “this is gonna be great” and becomes instead something you (and perhaps others) can actually evaluate. Even if that evaluation is disastrous, it is also, I will insist, thrilling and productive. A project finished is the pump of a piston, preparing the engine for the next one.

Frankenstein is a terrific book partly based on how concerned people were about electricity. It captures our fears about the nature of being human but didn’t help anyone really come up with better policies for dealing with electricity. I worry that a lot of AI critics are doing the same thing.

...Let me add a certain virile reply recorded by De Quincey (Writings XI, 226). Someone flung a glass of wine in the face of a gentleman during a theological or literary debate. The victim did not show any emotion and said to the offender: 'This, sir, is a digression; now, if you please, for the argument.' (The author of that reply, a certain Dr. Henderson, died in Oxford around 1787, without leaving us any memory other than those just words: a sufficient and beautiful immortality.) A popular tale, which I picked up in Geneva during the last years of World War I, tells of Migel Servert’s reply to the inquisitors who had condemned him to the sake: 'I will burn, but this is a mere event. We shall continue our discussion in eternity.'

One interesting quirk of Pinboard is a complete absence of unit tests. I used to be a die-hard believer in testing, but in Pinboard tried a different approach, as an experiment. Instead of writng tests I try to be extremely careful in coding, and keep the code size small so I continue to understand it. I've found my defect rate to be pretty comparable to earlier projects that included extensive test suites and fixtures, but I am much more productive on Pinboard.

This is when I pull out “we don’t do that here.” It is a conversation ender. If you are the newcomer and someone who has been around a long time says “we don’t do that here”, it is hard to argue. This sentence doesn’t push my morality on anyone. If they want to do whatever it is elsewhere, I’m not telling them not to. I’m just cluing them into the local culture and values.

When attention is being appropriated, producers need to weigh the costs and benefits of the transaction. To assess whether the appropriation of attention is net-positive, it’s useful to distinguish between extractive and non-extractive contributions. Extractive contributions are those where the marginal cost of reviewing and merging that contribution is greater than the marginal benefit to the project’s producers. In the case of a code contribution, it might be a pull request that’s too complex or unwieldy to review, given the potential upside

For the same cost and similar speed to Claude 3 Haiku, Claude 3.5 Haiku improves across every skill set and surpasses even Claude 3 Opus, the largest model in our previous generation, on many intelligence benchmarks. Claude 3.5 Haiku is particularly strong on coding tasks. For example, it scores 40.6% on SWE-bench Verified, outperforming many agents using publicly available state-of-the-art models—including the original Claude 3.5 Sonnet and GPT-4o. [...] Claude 3.5 Haiku will be made available later this month across our first-party API, Amazon Bedrock, and Google Cloud’s Vertex AI—initially as a text-only model and with image input to follow.

It’s interesting to me how much [Closure] feels like a more advanced version of Dojo in many ways. There's a familiar package system, the widgets are significantly more mature, and Julie and Ojan's Editor component rocks. The APIs will feel familiar (if verbose) to Dojo users, the class hierarchies seem natural, and Closure even uses Acme, the Dojo CSS selector engine.