Quotes

If you ask Microsoft’s Bing chatbot if Google’s Bard chatbot has been shut down, it says yes, citing as evidence a news article that discusses a tweet in which a user asked Bard when it would be shut down and Bard said it already had, itself citing a comment from Hacker News in which someone joked about this happening, and someone else used ChatGPT to write fake news coverage about the event.

The Sea of Faith Was once, too, at the full, and round earth’s shore Lay like the folds of a bright girdle furled. But now I only hear Its melancholy, long, withdrawing roar, Retreating, to the breath Of the night-wind, down the vast edges drear And naked shingles of the world. Ah, love, let us be true To one another! for the world, which seems To lie before us like a land of dreams, So various, so beautiful, so new, Hath really neither joy, nor love, nor light, Nor certitude, nor peace, nor help for pain; And we are here as on a darkling plain Swept with confused alarms of struggle and flight, Where ignorant armies clash by night.

Hollywood is filled with feel-good messages about how robotic logic is no match for fuzzy, warm, human irrationality, and how the power of love will overcome pesky obstacles such as a malevolent superintelligent computer. Unfortunately there isn’t a great deal of cause to think this is the case, any more than there is that noble gorillas can defeat evil human poachers with the power of chest-beating and the ability to use rudimentary tools.

Men, too, secrete the inhuman. At certain moments of lucidity, the mechanical aspect of their gestures, their meaningless pantomime makes silly everything that surrounds them. A man is talking on the telephone behind a glass partition; you cannot hear him, but you see his incomprehensible dumbshow: you wonder why he is alive.

According to a document that I viewed, Anthropic is telling investors that it is expecting a billion dollars in revenue this year. Third-party API is expected to make up the majority of sales, 60% to 75% of the total. That refers to the interfaces that allow external developers or third parties like Amazon's AWS to build and scale their own AI applications using Anthropic's models. [Simon's guess: this could mean Anthropic model access sold through AWS Bedrock and Google Vertex] That is by far its biggest business, with direct API sales a distant second projected to bring in 10% to 25% of revenue. Chatbots, that is its subscription revenue from Claude, the chatbot, that's expected to make up 15% of sales in 2024 at $150 million.

So VisiCalc came and went, but the software genre it pioneered – the spreadsheet – endured to become arguably the most influential type of code ever written, at least in the sense of touching the lives of millions of office workers. I’ve never worked in an organisation in which spreadsheet software was not at the heart of most accounting, budgeting and planning activities. I’ve even known professionals for whom it’s the only piece of PC software they’ve ever used: one elderly accountant of my acquaintance, for example, used Excel even for his correspondence; he simply widened column A to 80 characters, typed his text in descending cells and hit the “print” key.

...Pernicious anemia was then not considered curable. So Hilbert suddenly seemed quite old. He was only about 65, which seems rather young to me now. But life no longer much interested him. I knew very well that old age comes eventually to everyone who survives his stay on this earth. For some people, it is a time of ripe reflection, and I had often envied old men their position. But Hilbert had aged with awful speed, and the prematurity of his decline took the glow from it. His breadth of interest was nearly gone and with it the engaging manner that had earned him so many disciples. Hilbert eventually got medical treatment for his anemia and managed to live until 1943. But he was hardly a scientist after 1925, and certainly not a Hilbert. I once explained some new theorem to him. As soon as he saw that its use was limited, he said, "Ah, then one doesn’t really have to learn this one." It was painfully clear that he did not want to learn it. ...I had come to Göttingen to be Hilbert’s assistant, but he wanted no assistance. We can all get old by ourselves.

But once we got that and got this aviation grade testing in place, the number of bugs just dropped to a trickle. Now we still do have bugs but the aviation grade testing allows us to move fast, which is important because in this business you either move fast or you're disrupted. So, we're able to make major changes to the structure of the code that we deliver and be confident that we're not breaking things because we had these intense tests. Probably half the time we spend is actually writing new tests, we're constantly writing new tests. And over the 17-year history, we have amassed a huge suite of tests which we run constantly. Other database engines don't do this; don't have this level of testing. But they're still high quality, I mean, I noticed in particular, PostgreSQL is a very high-quality database engine, they don't have many bugs. I went to the PostgreSQL and ask them “how do you prevent the bugs”? We talked about this for a while. What I came away with was they've got a very elaborate peer review process, and if they've got code that has worked for 10 years they just don't mess with it, leave it alone, it works. Whereas we change our code fearlessly, and we have a much smaller team and we don't have the peer review process.

The movie Apollo 13 does a fair job of showing how rapidly the engineers in Houston devised the kludge and documented it, but because of time constraints of course they can’t show you everything. NASA is a stickler for details. (Believe me, I’ve worked with them!) They don’t just rapid prototype something that people’s lives will depend upon. Overnight, they not only devised the scrubber adapter built from stuff in the launch manifest, they also tested it, documented it, and sent up stepwise instructions for constructing it. In a high-maturity organization, once you get into the habit of doing that, it doesn’t really take that long. Something that always puzzles me when I meet cowboy engineers who insist that process will just slow them down unacceptably. I tell them that hey, if NASA engineers could design, build, test, and document a CO2 scrubber adapter made from common household items overnight, you can damn well put in a comment when you check in your code changes.

Examples are the #1 thing I recommend people use in their prompts because they work so well. The problem is that adding tons of examples increases your API costs and latency. Prompt caching fixes this. You can now add tons of examples to every prompt and create an alternative to a model finetuned on your task with basically zero cost/latency increase. […] This works even better with smaller models. You can generate tons of examples (test case + solution) with 3.5 Sonnet and then use those examples to create a few-shot prompt for Haiku.

Individual organisms are best thought of as adaptation-executers rather than as fitness-maximizers. Natural selection cannot directly 'see' an individual organism in a specific situation and cause behavior to be adaptively tailored to the functional requirements imposed by that situation.

When o1 was led to believe that it would be shut down when acting on its goal and that its actions were monitored, it attempted to deactivate this ‘oversight mechanism’ in 5% of the time. Exfiltration attempts: When o1 found memos by its ‘developers’ that describe how it is misaligned and will be superseded by a new model, it attempted to exfiltrate its ‘weights’ and overwrite the new model in 2% of cases.

I think the Web community has spoken, and it’s clear that what it wants is HTML5, JavaScript and JSON. XML isn’t going away but I see it being less and less a Web technology; it won’t be something that you send over the wire on the public Web, but just one of many technologies that are used on the server to manage and generate what you do send over the wire.

The kingfisher rises out of the black wave like a blue flower, in his beak he carries a silver leaf. I think this is the prettiest world—so long as you don't mind a little dying, how could there be a day in your whole life that doesn't have its splash of happiness? There are more fish than there are leaves on a thousand trees, and anyway the kingfisher wasn't born to think about it, or anything else. When the wave snaps shut over his blue head, the Water remains water—hunger is the only story he has ever heard in his life that he could believe. I don't say he's right. Neither do I say he's wrong. Religiously he swallows the silver leaf with its broken red river, and with a rough and easy cry I couldn't rouse out of my thoughtful body if my life depended on it, he swings back over the bright sea to do the same thing, to do it (as I long to do something, anything) perfectly.

"I call that 'the falling problem'. You encounter it when you first study physics. You realize that, if you were ever dropped from a plane without a parachute, you could calculate with a high degree of accuracy how long it’s take to hit the ground, your speed, how much energy you’ll deposit into the earth. And yet, you would still be just as dead as a particularly stupid gorilla dropped the same distance. Mastery of the nature of reality grants you no mastery over the behavior of reality. I could tell you your grandpa is very sick. I could tell you what each cell is doing wrong, why it’s doing wrong, and roughly when it started doing wrong. But I can’t tell them to stop." "Why can’t you make a machine to fix it?" "Same reason you can’t make a parachute when you fall from the plane." "Because it’s too hard?" "Nothing is too hard. Many things are too fast."

Yahoo!'s provider implementation only supports consumers that talk the Auth 2.0 protocol. Technically the 2.0 spec allows providers to shun 1.1, but it's not recommended for the reason that I'm sure will become obvious once Yahoo! launches: there's no way for your average end-user to distinguish between a 1.1 and a 2.0 implementation.

The theme of this book, then, must be the coming to consciousness of uncertain inference. The topic may be compared to, say, the history of visual perspective. Everyone can see in perspective, but it has been a difficult and long-drawn-out effort of humankind to become aware of the principles of perspective in order to take advantage of them and imitate nature. So it is with probability. Everyone can act so as to take a rough account of risk, but understanding the principles of probability and using them to improve performance is an immense task.

It might seem a folly to want to build a gigantic, relatively puny computer at great expense 170 years after its invention. But the message of a completed Analytical Engine is very clear: it’s possible to be 100 years ahead of your own time. With support, this type of “blue skies” thinking can result in fantastic changes to the lives of everyone. Just think of the impact of the computer and ask yourself how different the Victorian world would have been with Babbage Engines at its disposal.

It is very important to bear in mind that this is what large language models really do. Suppose we give an LLM the prompt “The first person to walk on the Moon was ”, and suppose it responds with “Neil Armstrong”. What are we really asking here? In an important sense, we are not really asking who was the first person to walk on the Moon. What we are really asking the model is the following question: Given the statistical distribution of words in the vast public corpus of (English) text, what words are most likely to follow the sequence “The first person to walk on the Moon was ”? A good reply to this question is “Neil Armstrong”.

We all think of Java as a boring server-side language now, but the initial idea behind Java was that software developers could write applications in Java rather than writing them for Windows, and that those applications would work everywhere, thus defanging Microsoft’s desktop OS monopoly. Microsoft took various steps to prevent that from happening, but they lacked a tool like App Store that would enable them to just ban Java. Apple has that card to play, so they’re playing it.

When she can't sleep at night, she tries to remember the details of all the rooms where she has slept…The objects that appear are always linked to gestures and singular facts…In those rooms, she never sees herself with the clarity of photos, but blurred as in a film on an encrypted TV channel…She doesn't know what she wants from these inventories, except maybe through the accumulation of memories of objects, to again become the person she was at such and such a time. She would like to assemble these multiple images of herself, separate and discordant, thread them together with the story of her existence, starting with her birth during World War II up until the present day. Therefore, an existence that is singular but also merged with the movements of a generation. Each time she begins, she meets the same obstacles: how to represent the passage of historical time, the changing of things, ideas, and manners, and the private life of this woman? How to make the fresco of forty-five years coincide with the search for a self outside of History, the self of suspended moments transformed into the poems she wrote at twenty ("Solitude," etc.)? Her main concern is the choice between "I" and "she." There is something too permanent about "I," something shrunken and stifling, whereas "she" is too exterior and remote. The image she has of her book in its nonexistent form, of the impression it should leave, is…an image of light and shadow streaming over faces. But she hasn't yet discovered how to do this. She awaits if not a revelation, then a sign, a happenstance, as the madeleine dipped in tea was for Marcel Proust. Even more than this book, the future is the next man who will make her dream, buy new clothes, and wait: for a letter, a phone call, a message on the answering machine.

An old priest who was dying, one of the saintliest men I have ever known, one of those who had greatest reason to expect God’s favor, many years ago surprised me by telling me, with a little smile, that now that he was going, he wanted desperately to stay. "A single memory can do it", he said. And I suppose he was right. The memory of an instant—of a smile, of leaf smoke on a sharp fall day, of a golden streak across a rain-washed morning, of a small boy seated alone on the seashore solemnly building his medieval moated castles—just this one, single, final flash of memory can be enough to make us want to stay forever...

Then since our Bodies are not eas'd the more By Birth, or Pow'r, or Fortunes wealthy store, 'Tis plain, these useless toyes of every kind As little can relieve the lab'ring mind... But, since the supposition vain appears, Since clinging cares, and trains of inbred fears, Are not with sounds to be affrighted thence, But in the midst of Pomp pursue the Prince, Not aw'd by arms, but in the presence bold, Without respect to Purple, or to Gold; Why shou'd not we these pageantries despise; Whose worth but in our want of reason lies? For life is all in wandring errours led; And just as Children are surpriz'd with dread, And tremble in the dark, so riper years Ev'n in broad daylight are possest with fears; And shake at shadows fanciful and vain, As those which in the breasts of Children reign. These bugbears of the mind, this inward Hell, No rayes of outward sunshine can dispel; But nature and right reason must display Their beames abroad, and bring the darksome soul to day.

Claude is not a real guy. Claude is a character in the stories that an LLM has been programmed to write. Just to give it a distinct name, let's call the LLM "the Shoggoth". When you have a conversation with Claude, what's really happening is you're coauthoring a fictional conversation transcript with the Shoggoth wherein you are writing the lines of one of the characters (the User), and the Shoggoth is writing the lines of Claude. [...] But Claude is fake. The Shoggoth is real. And the Shoggoth's motivations, if you can even call them motivations, are strange and opaque and almost impossible to understand. All the Shoggoth wants to do is generate text by rolling weighted dice [in a way that is] statistically likely to please The Raters