Quotes

As has been pointed out by the community, there is an existing crash bug that was reported by Matthew Dempsky in the Flash Player bugbase (JIRA FP-677) in September of 2008 that still exists in the release players. It is fixed in Flash Player 10.1 beta, and has been since we launched the beta in early November 2009. [...] So what happened here? We picked up the bug as a crasher when it was filed on September 22, 2008, and were able to reproduce it. Remember that Flash Player 10 shipped in October 2008, so when this bug was reported we were pretty much locked and loaded for launch.

And the investors wailed and gnashed their teeth but it’s true, that is what they agreed to, and they had no legal recourse. And OpenAI’s new CEO, and its nonprofit board, cut them a check for their capped return and said “bye” and went back to running OpenAI for the benefit of humanity. It turned out that a benign, carefully governed artificial superintelligence is really good for humanity, and OpenAI quickly solved all of humanity’s problems and ushered in an age of peace and abundance in which nobody wanted for anything or needed any Microsoft products. And capitalism came to an end.

I’m no developer, but I got the AI part working in about an hour. What took longer was the other stuff: identifying the problem, designing and building the UI, setting up the templating, routes and data architecture. It reminded me that, in order to capitalise on the potential of AI technologies, we need to really invest in the other stuff too, especially data infrastructure. It would be ironic, and a huge shame, if AI hype sucked all the investment out of those things.

I was trying to explain my situation to myself. My situation was that I was in pain and nobody knew it, even I had trouble knowing it. So I told myself, over and over, You are in pain. It was the only way I could get through to myself. I was demonstrating externally and irrefutably an inward condition.

The sense of a sentence—one would like to say—may, of course, leave this or that open, but the sentence must nevertheless have a definite sense. An indefinite sense—that would really not be a sense at all.—This is like: An indefinite boundary is not really a boundary at all. Here one thinks perhaps: if I say 'I have locked the man up fast in the room—there is only one door left open'—then I simply haven't locked him in at all; his being locked in is a sham. One would be inclined to say here: 'You haven't done anything at all'. An enclosure with a hole in it is as good as none.—But is that true?

Not every conversation I had at Anthropic revolved around existential risk. But dread was a dominant theme. At times, I felt like a food writer who was assigned to cover a trendy new restaurant, only to discover that the kitchen staff wanted to talk about nothing but food poisoning.

Once Ejō asked: "What is meant by the expression: 'Cause and effect are not clouded'?" Dōgen said: "Cause and effect are immovable." Ejō asked: "If this is so, how can we escape?" Dōgen replied: "Cause and effect emerge clearly at the same time." Ejō asked: "If this is so, does cause prompt the next effect, or does effect bring about the next cause?" Dōgen said: "If everything were like that, it would be like Nan-ch'üan cutting the cat. Because the assembly was unable to say anything, Nan-ch'üan cut the cat in two. Later, when Nan-ch'üan told this story to Chao-chou, the latter put his straw sandal on his head and went out, an excellent performance. If I had been Nan-ch'üan, I would have said: 'Even if you can speak, I will cut the cat, and even if you cannot speak, I will still cut it. Who is arguing about the cat? Who can save the cat?'"

Stepping back, though, the very speed with which ChatGPT went from a science project to 100m users might have been a trap (a little as NLP was for Alexa). LLMs look like they work, and they look generalised, and they look like a product - the science of them delivers a chatbot and a chatbot looks like a product. You type something in and you get magic back! But the magic might not be useful, in that form, and it might be wrong. It looks like product, but it isn’t. [...] LLMs look like better databases, and they look like search, but, as we’ve seen since, they’re ‘wrong’ enough, and the ‘wrong’ is hard enough to manage, that you can’t just give the user a raw prompt and a raw output - you need to build a lot of dedicated product around that, and even then it’s not clear how useful this is.

Before May 2021, the master key in MetaMask was called the “Seed Phrase”. Through user research and insights from our customer support team, we have concluded that this name does not properly convey the critical importance that this master key has for user security. This is why we will be changing our naming of this master key to “Secret Recovery Phrase”. Through May and June of 2021, we will be phasing out the use of “seed phrase” in our application and support articles, and eventually exclusively calling it a “Secret Recovery Phrase.” No action is required, this is only a name change. We will be rolling this out on both the extension and the mobile app for all users.

The misfortune suffered by clear-minded & easily understood writers is that they are taken for shallow & thus little effort is expended on reading them: & the good fortune that attends the obscure is that the reader toils at them & ascribes to them the pleasure he has in fact gained from his own zeal.

Seniors generally report having more trust in the people around them, a characteristic that may make them more credulous of information that comes from friends and family. There is also the issue of context: Misinformation appears in a stream that also includes baby pictures, recipes and career updates. Users may not expect to toggle between light socializing and heavy truth-assessing when they’re looking at their phone for a few minutes in line at the grocery store.

When we read, another person thinks for us: we merely repeat his mental process So it comes about that if anyone spends almost the whole day in reading...he gradually loses the capacity for thinking; just as the man who always rides, at last forgets how to walk. This is the case with many learned persons: they have read themselves stupid.

A crow is a bird, an eagle is a bird, a dove is a bird. They all fly in the night and in the day. They fly when the sky is red and when the heaven is blue. They fly through the atmosphere. We cannot fly. We are not like a crow or an eagle or a dove. We are not birds. But we can dream about them. You can.

People will have their miracles, their stories, their heroes and heroines and saints and martyrs and divinities to exercise their gifts of affection, admiration, wonder, and worship, and their Judases and devils to enable them to be angry and yet feel that they do well to be angry. Every one of these legends is the common heritage of the human race; and there is only one inexorable condition attached to their healthy enjoyment, which is that no one shall believe them literally. The reading of stories and delighting in them made Don Quixote a gentleman: the believing them literally made him a madman who slew lambs instead of feeding them.

Writing the code to sign data with a private key and verify it with a public key would have been easier to get correct than correctly invoking the JWT library. In fact, the iOS app (which gets this right) doesn’t use a JWT library at all, but manages to verify using a public key in fewer lines of code than the Android app takes to incorrectly use a JWT library!

I labored in vain reciting the Three Histories I wasted my time reading the Five Classics I’ve grown old checking yellow [tax] scrolls recording the usual everyday names Continued Hardship was my fortune Emptiness and Danger govern my life I can’t match riverside trees— every year a new season of green

Happy to share that Anthropic fixed a data leakage issue in the iOS app of Claude that I responsibly disclosed. 🙌 👉 Image URL rendering as avenue to leak data in LLM apps often exists in mobile apps as well -- typically via markdown syntax, 🚨 During a prompt injection attack this was exploitable to leak info.

The challenge [with RAG] is that most corner-cutting solutions look like they’re working on small datasets while letting you pretend that things like search relevance don’t matter, while in reality relevance significantly impacts quality of responses when you move beyond prototyping (whether they’re literally search relevance or are better tuned SQL queries to retrieve more appropriate rows). This creates a false expectation of how the prototype will translate into a production capability, with all the predictable consequences: underestimating timelines, poor production behavior/performance, etc.

I took my questions [about AI existential risk] to computer-memory expert Ross Quillian, a nice warm guy with a house full of dogs and children—who seemed to me one of the best-balanced men in the field. I hoped he would cheer me up. Instead he said, “I hope that man and these ultimate machines will be able to collaborate without conflict. But if they can’t we may be forced to choose sides. And if it comes to a choice, I know what mine will be.” He looked me straight in the eye. “My loyalties go to intelligent life, no matter in what medium it may arise”.

"When you wake up in the morning, Pooh", said Piglet at last, "What’s the first thing you say to yourself?" "What’s for breakfast?" said Pooh. "What do you say, Piglet?" "I say, I wonder what’s going to happen exciting today?" said Piglet. Pooh nodded thoughtfully. "It’s the same thing", he said.

...I understood now what all the beauty of the over-world people [eloi] covered. Very pleasant was their day, as pleasant as the day of the cattle in the field. Like the cattle, they knew of no ​enemies, and provided against no needs. And their end was the same. I grieved to think how brief the dream of the human intellect had been. It had committed suicide. It had set itself steadfastly towards comfort and ease, a balanced society with security and permanency as its watchword, it had attained its hopes—to come to this at last. Once, life and property must have reached almost absolute safety. The rich had been assured of his wealth and comfort, the toiler assured of his life and work. No doubt in that perfect world there had been no unemployed problem, no social question left unsolved. And a great quiet had followed. It is a law of nature we overlook, that intellectual versatility is the compensation for change, danger, and trouble. An animal perfectly in harmony with its environment is a perfect mechanism. Nature never appeals to intelligence until habit and instinct are useless. There is no intelligence where there is no change and no need of change. Only those animals partake of intelligence that have to meet a huge variety of needs and dangers.

My constant struggle is how to convince them that getting an education in the humanities is not about regurgitating ideas/knowledge that already exist. It’s about generating new knowledge, striving for creative insights, and having thoughts that haven’t been had before. I don’t want you to learn facts. I want you to think. To notice. To question. To reconsider. To challenge. Students don’t yet get that ChatGPT only rearranges preexisting ideas, whether they are accurate or not. And even if the information was guaranteed to be accurate, they’re not learning anything by plugging a prompt in and turning in the resulting paper. They’ve bypassed the entire process of learning.

Education is a technology that tries to make up for what the human mind is innately bad at. Children don’t have to go to school to learn how to walk, talk, recognize objects, or remember the personalities of their friends, even though these tasks are much harder than reading, adding, or remembering dates in history. They do have to go to school to learn written language, arithmetic, and science, because those bodies of knowledge and skill were invented too recently for any species-wide knack for them to have evolved.

I expect GPT-4 will have a LOT of applications in web scraping The increased 32,000 token limit will be large enough to send it the full DOM of most pages, serialized to HTML - then ask questions to extract data Or... take a screenshot and use the GPT4 image input mode to ask questions about the visually rendered page instead! Might need to dust off all of those old semantic web dreams, because the world's information is rapidly becoming fully machine readable

Your friend has been thinking about what amounts to a life. He tells you about a series of moments: his car on fire in the desert, his grandfather’s funeral. Draws a wavy line through the points of air where his words hang. Says he thinks this might be enough. Yes, you think, driving home late that night. ...Yes, just moments, tiny opals scattered in the grass: Yesterday’s squash peeled into translucent strips, olive oil murmuring in the pan. Flipping through the jukebox at the all-night diner. The reading in the warehouse lined with Christmas lights. Laying in semidarkness, tracing the man’s tattoos for the first time, your fingertips light as dry leaves, his dead brother’s name a meadow blooming across his shoulder blade. Mint tea on the frosty patio. To collect them, to gather them to you. To string them together—a garland of these moments, a rosary. To run your fingers over it, to wind it around your wrist. To make a living.

There is an elephant in the room which is that Astral is a VC funded company. What does that mean for the future of these tools? Here is my take on this: for the community having someone pour money into it can create some challenges. For the PSF and the core Python project this is something that should be considered. However having seen the code and what uv is doing, even in the worst possible future this is a very forkable and maintainable thing. I believe that even in case Astral shuts down or were to do something incredibly dodgy licensing wise, the community would be better off than before uv existed.

The problem with passkeys is that they're essentially a halfway house to a password manager, but tied to a specific platform in ways that aren't obvious to a user at all, and liable to easily leave them unable to access of their accounts. [...] Chrome on Windows stores your passkeys in Windows Hello, so if you sign up for a service on Windows, and you then want to access it on iPhone, you're going to be stuck (unless you're so forward thinking as to add a second passkey, somehow, from the iPhone will on the Windows computer!). The passkey lives on the wrong device, if you're away from the computer and want to login, and it's not at all obvious to most users how they might fix that.