Quotes

When I first published the micrograd repo, it got some traction on GitHub but then somewhat stagnated and it didn't seem that people cared much. [...] When I made the video that built it and walked through it, it suddenly almost 100X'd the overall interest and engagement with that exact same piece of code. [...] you might be leaving somewhere 10-100X of the potential of that exact same piece of work on the table just because you haven't made it sufficiently accessible.

11. (A beginning ends what an end begins. 204. By looking for the origins of things we deceive ourselves about their inevitability. Things that did not happen also have origins. 209. Nature is not that prejudiced against the past. Yesterday is only a little less likely to happen next than any one of the zillion possible tomorrows that does. 217. All the falling rain is caught. 499. To feel an end is to discover that there had been a beginning. A parenthesis closes that we hadn’t realized was open).

GitHub’s slopocalypse – the flood of AI-generated spam PRs and issues – has made Jazzband’s model of open membership and shared push access untenable. Jazzband was designed for a world where the worst case was someone accidentally merging the wrong PR. In a world where only 1 in 10 AI-generated PRs meets project standards, where curl had to shut down its bug bounty because confirmation rates dropped below 5%, and where GitHub’s own response was a kill switch to disable pull requests entirely – an organization that gives push access to everyone who joins simply can’t operate safely anymore.

All the big guns want an iPhone killer. Even I, mad for all things Apple as I am, want an iPhone killer. I want smart digital devices to be as good as mankind’s ingenuity can make them. I want us eternally to strive to improve and surprise. Bring on the iPhone killers. Bring them on.

The primary problem is that while the answers which ChatGPT produces have a high rate of being incorrect, they typically look like they might be good and the answers are very easy to produce. There are also many people trying out ChatGPT to create answers, without the expertise or willingness to verify that the answer is correct prior to posting. Because such answers are so easy to produce, a large number of people are posting a lot of answers. The volume of these answers (thousands) and the fact that the answers often require a detailed read by someone with at least some subject matter expertise in order to determine that the answer is actually bad has effectively swamped our volunteer-based quality curation infrastructure.

The use of double underscores creates a separate namespace for names that are part of the Python language definition, so that programmers are free to create variables, attributes, and methods that start with letters, without fear of silently colliding with names that have a language-defined purpose.

Is it safe to say that LLMs are, in essence, making us "dumber"? No! Please do not use the words like “stupid”, “dumb”, “brain rot”, "harm", "damage", and so on. It does a huge disservice to this work, as we did not use this vocabulary in the paper, especially if you are a journalist reporting on it.

For, gentlemen, it is a fact, that every philosopher of eminence for the two last centuries has either been murdered, or, at the least, been very near it; insomuch, that if a man calls himself a philosopher, and never had his life attempted, rest assured there is nothing in him; and against Locke’s philosophy in particular, I think it an unanswerable objection...that, although he carried his throat about with him in this world for seventy-two years, no man ever condescended to cut it.

Building technology in startups is all about having the right level of tech debt. If you have none, you’re probably going too slow and not prioritizing product-market fit and the important business stuff. If you get too much, everything grinds to a halt. Plus, tech debt is a “know it when you see it” kind of thing, and I know that my definition of “a bunch of tech debt” is, to other people, “very little tech debt.”

He that lay in a golden Urne eminently above the Earth, was not likely to finde the quiet of these bones. Many of these Urnes were broke by a vulgar discoverer in hope of inclosed treasure. The ashes of Marcellus were lost above ground, upon the like account. Where profit hath prompted, no age hath wanted such miners. For which the most barbarous Expilators found the most civill Rhetorick. Gold once out of the earth is no more due unto it; What was unreasonably committed to the ground is reasonably resumed from it: Let Monuments and rich Fabricks, not Riches adorn mens ashes. The commerce of the living is not to be transferred unto the dead: It is not injustice to take that which none complains to lose, and no man is wronged where no man is possessor.

No man is an iland, intire of it selfe; every man is a peece of the Continent, a part of the maine; if a clod bee washed away by the Sea, Europe is the lesse, as well as if a Promontorie were, as well as if a Mannor of thy friends or of thine owne were; any mans death diminishes me, because I am involved in Mankinde; And therefore never send to know for whom the bell tolls; It tolls for thee...

Note that there have been no breaking changes since the [SQLite] file format was designed in 2004. The changes shows in the version history above have all be one of (1) typo fixes, (2) clarifications, or (3) filling in the "reserved for future extensions" bits with descriptions of those extensions as they occurred.

The general idea of an “Islands” architecture is deceptively simple: render HTML pages on the server, and inject placeholders or slots around highly dynamic regions. These placeholders/slots contain the server-rendered HTML output from their corresponding widget. They denote regions that can then be "hydrated" on the client into small self-contained widgets, reusing their server-rendered initial HTML.

A time is marked not so much by ideas that are argued about as by ideas that are taken for granted. The character of an era hangs upon what needs no defense. Power runs with ideas that only the crazy would draw into doubt. The "taken for granted" is the test of sanity; "what everyone knows" is the line between us and them.

Consider Bitcoin a grand middle finger. It’s a prank, almost a parody of the global financial system, that turned into a bubble. “You plutocrats of Davos may think you control the global money supply,” the pranksters seem to say. “But humans will make an economy out of anything. Even this!”

We show for the first time that large-scale generative pretrained transformer (GPT) family models can be pruned to at least 50% sparsity in one-shot, without any retraining, at minimal loss of accuracy. [...] We can execute SparseGPT on the largest available open-source models, OPT-175B and BLOOM-176B, in under 4.5 hours, and can reach 60% unstructured sparsity with negligible increase in perplexity: remarkably, more than 100 billion weights from these models can be ignored at inference time.

The Oxen once upon a time sought to destroy the Butchers, who practiced a trade destructive to their race. They assembled on a certain day to carry out their purpose, and sharpened their horns for the contest. But one of them who was exceedingly old (for many a field had he plowed) thus spoke: "These Butchers, it is true, slaughter us, but they do so with skillful hands, and with no unnecessary pain. If we get rid of them, we shall fall into the hands of unskillful operators, and thus suffer a double death: for you may be assured, that though all the Butchers should perish, yet will men ever want beef." Moral: Do not be in a hurry to change one evil for another.

To evaluate the model’s capability in processing long-context inputs, we construct a video “Needle-in- a-Haystack” evaluation on Qwen3-VL-235B-A22B-Instruct. In this task, a semantically salient “needle” frame—containing critical visual evidence—is inserted at varying temporal positions within a long video. The model is then tasked with accurately locating the target frame from the long video and answering the corresponding question. [...] As shown in Figure 3, the model achieves a perfect 100% accuracy on videos up to 30 minutes in duration—corresponding to a context length of 256K tokens. Remarkably, even when extrapolating to sequences of up to 1M tokens (approximately 2 hours of video) via YaRN-based positional extension, the model retains a high accuracy of 99.5%.

...the spacing that has made for the most successful inductions will have tended to predominate through natural selection. Creatures inveterately wrong in their inductions have a pathetic but praiseworthy tendency to die before reproducing their kind...In induction nothing succeeds like success.

When we speak of improving the mind we are usually referring to the acquisition of information or knowledge, or to the type of thoughts one should have, and not to the actual functioning of the mind. We spend little time monitoring our own thinking and comparing it with a more sophisticated ideal.

Eventually, however, HudZah wore Claude down. He filled his Project with the e-mail conversations he’d been having with fusor hobbyists, parts lists for things he’d bought off Amazon, spreadsheets, sections of books and diagrams. HudZah also changed his questions to Claude from general ones to more specific ones. This flood of information and better probing seemed to convince Claude that HudZah did know what he was doing, and the AI began to give him detailed guidance on how to build a nuclear fusor and how not to die while doing it.

O ye, whom wrath consumes! who, passion-stung, Writhe at defeat, and nurse your agonies! Shut up your senses, stifle up your ears, My voice is not a bellows unto ire. Yet listen, ye who will, whilst I bring proof How ye, perforce, must be content to stoop: And in the proof much comfort will I give, If ye will take that comfort in its truth. We fall by course of Nature's law, not force Of thunder, or of Jove. Great Saturn, thou Hast sifted well the atom-universe; But for this reason, that thou art the King, And only blind from sheer supremacy, One avenue was shaded from thine eyes, Through which I wandered to eternal truth. And first, as thou wast not the first of powers, So art thou not the last; it cannot be: Thou art not the beginning nor the end.

Why weekly? You want to keep your finger on the pulse of what’s really going on. When 1:1s are scheduled bi-weekly, and either of you have to cancel, you’ll likely be going a month between conversations and that is far too long to go without having a 1:1 with your direct report. Think of how much happens in a month. You don’t want to be that far behind!

They [Newtonians] saved appearances by forging expressly occult qualities or faculties which they imagined to be like little demons or goblins capable of producing unceremoniously that which is demanded, just as if watches marked the hours by a certain horodeictic faculty without having need of wheels, or as if mills crushed grains by a fractive faculty without needing anything resembling millstones.

Results are only valuable when the amount by which they probably differ from the truth is so small as to be insignificant for the purposes of the experiment. What the odds should be depends: On the degree of accuracy which the nature of the experiment allows, and On the importance of the issues at stake.

Please, fanboys, don't send me dumb notes averring that Apple's failure to police this use of its mark will lead to the end of its ability to stop manufacturers from producing rival MP3 players and calling them iPods. That's a fairy tale that trademark lawyers tell their kids when they want to reassure them that they'll have a healthy college fund.

You remember the colleague who, in a sad letter, confessed to me that, at last, his department too had bought a research computer, thus almost ensuring that its output at large would fall down to the general level of mediocrity?... God save us from the "research vehicle"—SRC terminology (UK—for, if you are not very careful, you will find yourself fully absorbed by ‘driving’ it. This, by the way also explains the almost certain scientific sterility of computing science research establishments primarily manned by electronic engineers, when building something is the best they can think of to do.