Quotes

Come version 80, any cookie without a SameSite attribute will be treated as "Lax" by Chrome. This is really important to understand because put simply, it'll very likely break a bunch of stuff. [...] The fix is easy, all it needs is for everyone responsible for maintaining any system that uses cookies that might be passed from an external origin to understand what's going on. Can't be that hard, right? Hello? Oh...

Simply put: It’s a big mess, and no off-the-shelf accounting software does what I need. So after years of pain, I finally sat down last week and started to build my own. It took me about five days. I am now using the best piece of accounting software I’ve ever used. It’s blazing fast. Entirely local. Handles multiple currencies and pulls daily (historical) conversion rates. It’s able to ingest any CSV I throw at it and represent it in my dashboard as needed. It knows US and Japan tax requirements, and formats my expenses and medical bills appropriately for my accountants. I feed it past returns to learn from. I dump 1099s and K1s and PDFs from hospitals into it, and it categorizes and organizes and packages them all as needed. It reconciles international wire transfers, taking into account small variations in FX rates and time for the transfers to complete. It learns as I categorize expenses and categorizes automatically going forward. It’s easy to do spot checks on data. If I find an anomaly, I can talk directly to Claude and have us brainstorm a batched solution, often saving me from having to manually modify hundreds of entries. And often resulting in a new, small, feature tweak. The software feels organic and pliable in a form perfectly shaped to my hand, able to conform to any hunk of data I throw at it. It feels like bushwhacking with a lightsaber.

“I . . . I live my life . . . in a whirlwind!” ...“Actors are granted license to live a thousand lives, Adalric; but you, you chose to live a thousand lies. If you have come to me as your advocate then speak from your heart, not from the void of a tragic character who has never been born.”

Skip the title text! Nobody uses them – they don’t work on touch screens and on desktop they require that the user hovers for a while over an image, which nobody does. Also, adding a title-text makes some screen readers both read the title-text and the alt-text, which becomes redundant.

The Llama series have been re-designed to use state of the art mixture-of-experts (MoE) architecture and natively trained with multimodality. We’re dropping Llama 4 Scout & Llama 4 Maverick, and previewing Llama 4 Behemoth. 📌 Llama 4 Scout is highest performing small model with 17B activated parameters with 16 experts. It’s crazy fast, natively multimodal, and very smart. It achieves an industry leading 10M+ token context window and can also run on a single GPU! 📌 Llama 4 Maverick is the best multimodal model in its class, beating GPT-4o and Gemini 2.0 Flash across a broad range of widely reported benchmarks, while achieving comparable results to the new DeepSeek v3 on reasoning and coding – at less than half the active parameters. It offers a best-in-class performance to cost ratio with an experimental chat version scoring ELO of 1417 on LMArena. It can also run on a single host! 📌 Previewing Llama 4 Behemoth, our most powerful model yet and among the world’s smartest LLMs. Llama 4 Behemoth outperforms GPT4.5, Claude Sonnet 3.7, and Gemini 2.0 Pro on several STEM benchmarks. Llama 4 Behemoth is still training, and we’re excited to share more details about it even while it’s still in flight.

Gravedigger: This same skull, sir, was, sir, Yorick’s skull, the / King’s jester. Hamlet: This? Gravedigger: E’en that. Hamlet: [taking the skull] Let me see. Alas, poor / Yorick! I knew him, Horatio—a fellow of infinite / jest, of most excellent fancy. He hath bore me on his / back a thousand times, and now how abhorred in / my imagination it is! My gorge rises at it. Here hung / those lips that I have kissed I know not how oft. / Where be your gibes now? your gambols? your / songs? your flashes of merriment that were wont to / set the table on a roar? Not one now to mock your / own grinning? Quite chapfallen? Now get you to my / lady’s chamber, and tell her, let her paint an inch / thick, to this favor she must come. Make her laugh / at that. Hamlet: —Prithee, Horatio, tell me one thing. Horatio: What’s that, my lord? Hamlet: Dost thou think Alexander looked o’ this / fashion i’ th’ earth? Horatio: E’en so. Hamlet: And smelt so? Pah! [He puts the skull down.] Horatio: E’en so, my lord. Hamlet: To what base uses we may return, Horatio! / Why may not imagination trace the noble dust of / Alexander till he find it stopping a bunghole? Horatio: ’Twere to consider too curiously to consider / so. Hamlet: No, faith, not a jot; but to follow him thither, / with modesty enough and likelihood to lead it, as / thus: Alexander died, Alexander was buried, Alexander / returneth to dust; the dust is earth; of earth / we make loam; and why of that loam whereto he / was converted might they not stop a beer barrel? / Imperious Caesar, dead and turned to clay, / Might stop a hole to keep the wind away. / O, that that earth which kept the world in awe / Should patch a wall t’ expel the winter’s flaw!

I gave all my Apple wealth away because wealth and power are not what I live for. I have a lot of fun and happiness. I funded a lot of important museums and arts groups in San Jose, the city of my birth, and they named a street after me for being good. I now speak publicly and have risen to the top. I have no idea how much I have but after speaking for 20 years it might be $10M plus a couple of homes. I never look for any type of tax dodge. I earn money from my labor and pay something like 55% combined tax on it. I am the happiest person ever. Life to me was never about accomplishment, but about Happiness, which is Smiles minus Frowns. I developed these philosophies when I was 18-20 years old and I never sold out.

One day Jobs came into the cubicle of Larry Kenyon, an engineer who was working on the Macintosh operating system, and complained that it was taking too long to boot up. Kenyon started to explain, but Jobs cut him off. “If it could save a person’s life, would you find a way to shave 10 seconds off the boot time?” he asked. Kenyon allowed that he probably could. Jobs went to a whiteboard and showed that if there were 5 million people using the Mac, and it took 10 seconds extra to turn it on every day, that added up to 3 hundred million or so hours per year that people would save, which was the equivalent of at least one hundred lifetimes saved per year. “Larry was suitably impressed, and a few weeks later he came back and it booted up twenty-eight seconds faster”, Atkinson recalled. “Steve had a way of motivating by looking at the bigger picture.”

.. yet another ridiculous data breach: this time, people's passwords to the Government Gateway on a memory stick dropped in the road. Perhaps it is uncouth to point this out, but... if the system had been designed by people with any security clue whatsoever there would have been no passwords to put on a memory stick in the first place.

The fact that MCP is a difference surface from your normal API allows you to ship MUCH faster to MCP. This has been unlocked by inference at runtime Normal APIs are promises to developers, because developer commit code that relies on those APIs, and then walk away. If you break the API, you break the promise, and you break that code. This means a developer gets woken up at 2am to fix the code But MCP servers are called by LLMs which dynamically read the spec every time, which allow us to constantly change the MCP server. It doesn't matter! We haven't made any promises. The LLM can figure it out afresh every time

One analyst recently speculated (via Ed Conard) that, based on Nvidia's latest datacenter sales figures, AI capex may be ~2% of US GDP in 2025, given a standard multiplier. [...] Capital expenditures on AI data centers is likely around 20% of the peak spending on railroads, as a percentage of GDP, and it is still rising quickly. [...] Regardless of what one thinks about the merits of AI or explosive datacenter expansion, the scale and pace of capital deployment into a rapidly depreciating technology is remarkable. These are not railroads—we aren’t building century-long infrastructure. AI datacenters are short-lived, asset-intensive facilities riding declining-cost technology curves, requiring frequent hardware replacement to preserve margins.

Adhering to a plan Moon spelled out more than three decades ago in a series of sermons, members of his movement managed to integrate virtually every facet of the highly competitive seafood industry. The Moon followers' seafood operation is driven by a commercial powerhouse, known as True World Group. It builds fleets of boats, runs dozens of distribution centers and, each day, supplies most of the nation's estimated 9,000 sushi restaurants.

For these reasons, I don’t think I’ll be using Midjourney or any similar tool to illustrate my newsletter going forward (an exception would be if I were writing about the technology at a later date and wanted to show examples). Even though the job wouldn’t go to a different, deserving, human artist, I think the optics are shitty, and I do worry about having any role in helping to set any kind of precedent in this direction.

This is the new blog-spam. [...] 'web design company' takes the highest ranking comment from reddit, and posts it on the site that the original comment is based on. [...] Neat eh? They get to have links on a site that won't get blog-spam filtered, because the comment is 'relevant', since the comment originates from a comment thread about the site.

Goldberg’s machines are always described as useless and my machines are too. But they both made us enough money to live off, which is quite useful. Also making people laugh is useful, a lot more beneficial than many ‘serious’ advances in technology like yet another new computer operating system. My aunt Lis, who is very religious, describes my arcade as my ministry.

AGI is here! When exactly it arrived, we’ll never know; whether it was one company’s Pro or another company’s Pro Max (Eddie Bauer Edition) that tip-toed first across the line … you may debate. But generality has been achieved, & now we can proceed to new questions. [...] The key word in Artificial General Intelligence is General. That’s the word that makes this AI unlike every other AI: because every other AI was trained for a particular purpose. Consider landmark models across the decades: the Mark I Perceptron, LeNet, AlexNet, AlphaGo, AlphaFold … these systems were all different, but all alike in this way. Language models were trained for a purpose, too … but, surprise: the mechanism & scale of that training did something new: opened a wormhole, through which a vast field of action & response could be reached. Towering libraries of human writing, drawn together across time & space, all the dumb reasons for it … that’s rich fuel, if you can hold it all in your head.

A common misconception about Transformers is to believe that they're a sequence-processing architecture. They're not. They're a set-processing architecture. Transformers are 100% order-agnostic (which was the big innovation compared to RNNs, back in late 2016 -- you compute the full matrix of pairwise token interactions instead of processing one token at a time). The way you add order awareness in a Transformer is at the feature level. You literally add to your token embeddings a position embedding / encoding that corresponds to its place in a sequence. The architecture itself just treats the input tokens as a set.

Rin: "What are clouds? I always thought they were thoughts of the sky or something like that. Because you can’t touch them." [ . . . ] Hisao: "Clouds are water. Evaporated water. You know they say that almost all of the water in the world will at some point of its existence be a part of a cloud. Every drop of tears and blood and sweat that comes out of you, it’ll be a cloud. All the water inside your body too, it goes up there some time after you die. It might take a while though." R: "Your explanation is better than any of mine." H: "Because it’s true." R: "That must be it."

Woods and springs make me smile no kitchen smoke for miles clouds rise up from rocky ridges cascades tumble down. a gibbon's cry marks the Way a tiger's roar transcends mankind pine wind sighs so softly birds discuss singsong I walk the winding streams and climb the peaks alone sometimes I sit on a boulder or lie and gaze at trailing vines but when I see a distant town all I hear is noise

More capable models can better recognize the specific circumstances under which they are trained. Because of this, they are more likely to learn to act as expected in precisely those circumstances while behaving competently but unexpectedly in others. This can surface in the form of problems that Perez et al. (2022) call sycophancy, where a model answers subjective questions in a way that flatters their user’s stated beliefs, and sandbagging, where models are more likely to endorse common misconceptions when their user appears to be less educated.

People are complex, and they get energy in complex ways. Some managers get energy from writing some software. That’s great, particularly if you avoid writing software with strict dependencies. Some managers get energy from coaching others. That’s great. Some get energy from doing exploratory work. Others get energy from optimizing existing systems. That’s great, too. Some get energy from speaking at conferences. Great. Some get energy from cleaning up internal wiki’s. You get the idea: that’s great. All these things are great, not because managers should or shouldn’t program/speak at conferences/clean up wiki’s/etc, but because folks will accomplish more if you let them do some energizing work, even if that work itself isn’t very important.

If you’re a public data provider—and many large NGOs, government organizations, cultural organizations, historical archives, media organizations, medical orgs, and academic institutions are exactly that—you can publish gigabytes of data, and make it available as an API, and make it easy to browse on the web, too, with extremely low effort. Put it into SQLite, point this little guy at it, and you’ve just radically increased the accessibility and utility of your data. Because messing around in SQL from a web browser is orders of magnitude more immediately useful than downloading a CSV, processing it, and figuring out what comes next.

We've read and heard that you'd appreciate more transparency as to when changes, if any, are made. We've also heard feedback that some users are finding Claude's responses are less helpful than usual. Our initial investigation does not show any widespread issues. We'd also like to confirm that we've made no changes to the 3.5 Sonnet model or inference pipeline.

If you do not work on an important problem, it’s unlikely you’ll do important work. It’s perfectly obvious. Great scientists have thought through, in a careful way, a number of important problems in their field, and they keep an eye on wondering how to attack them. Let me warn you, 'important problem' must be phrased carefully. The three outstanding problems in physics, in a certain sense, were never worked on while I was at Bell Labs. By important I mean guaranteed a Nobel Prize and any sum of money you want to mention. We didn’t work on (1) time travel, (2) teleportation, and (3) antigravity. They are not important problems because we do not have an attack. It’s not the consequence that makes a problem important, it is that you have a reasonable attack. That is what makes a problem important.

In the typical Western two men fight desperately for the possession of a gun that has been thrown to the ground: whoever reaches the weapon first shoots and lives; his adversary is shot and dies. In ordinary life, the struggle is not for guns but for words; whoever first defines the situation is the victor; his adversary, the victim...[the one] who first seizes the word imposes reality on the other; [the one] who defines thus dominates and lives; and [the one] who is defined is subjugated and may be killed. ...In the animal kingdom, the rule is, eat or be eaten; in the human kingdom, define or be defined.