billti

If it’s predicting a next token to maximize scores against a training/test set, naively, wouldn’t that be expected?

I would imagine very little of the training data consists of a question followed by an answer of “I don’t know”, thus making it statistically very unlikely as a “next token”.

I did see a talk where someone was making the case that if the U.S. depends heavily on China for steel, and Taiwan for chips, then if China invades Taiwan the U.S. would very quickly be unable to wage a war with both supplies cut off. So the goal is to build some capacity for such products domestically.

They also tied that to Trump's desire to get out of Europe and deescalate the Middle East, being that if the U.S. was already stretched across Ukraine and say Iran, it would be way to stretched to also wage anything in the far east.

I'm no expert by a long way, but I can see some logic in the argument.

I've never understood why countries like the EU pass strict environment and labor laws and then DONT place some kind of tariff or tax on countries without those restrictions. Not only to offset making your own market less able to compete on cost, but to provide a financial incentive for the other countries to up their game on environmental and worker quality-of-life issues (if that is the overall aim).

Having recently gotten into quantum and listening to a lot of audiobooks on the history of it, that’s one of biggest takeaways for me. So many major advances in theory that languished for years because of the politics of the day of the personal opinions of their advisor, only for a physicist with greater standing to rediscover the same thing later and finally get it some attention. (Hugh Everett and David Bohm being two examples)

The sound designer goes over their use of this type of music in this excellent podcast. You'd really enjoy it. (Available Apple Podcasts, which is where I listened to it, but linking directly here).

The next one about the voices is excellent too, so linking both. The care they put into both the show and the people involved is almost as touching as the show itself.

https://www.20k.org/episodes/thesoundofbluey https://www.20k.org/episodes/thevoicesofbluey

I work on the quantum developer tools team at Microsoft. We put a lot of work into what we call the Quantum Katas to learn the basics via coding - https://quantum.microsoft.com/en-us/tools/quantum-katas

Our VS Code extension is trivial to install (https://learn.microsoft.com/en-us/azure/quantum/install-over...) or just try it entirely in the browser with Visual Studio Code online (https://vscode.dev/quantum/playground/)

To support that last scenario, where the language service, debugger, simulator, even package references, can run entirely in the browser, we built the whole thing using Rust compiled to WebAssembly, and our VS Code extension runs as pure JavaScript and Wasm. If interested you can dig into the implementation at https://github.com/microsoft/qsharp .

Happy to answer any questions!

If you know CUDA, then I assume you know a bit already about GPUs and the major concepts. There’s just minor differences and different terminology for things like “warps” etc.

With that base, I’ve found their docs decent enough, especially coupled with the Metal Shader Language pdf they provide (https://developer.apple.com/metal/Metal-Shading-Language-Spe...), and quite a few code samples you can download from the docs site (e.g. https://developer.apple.com/documentation/metal/performing_c...).

I’d note a lot of their stuff was still written in Objective-C, which I’m not that familiar with. But most of that is boilerplate and the rest is largely C/C++ based (including the Metal shader language).

I just ported some CPU/SIMD number crunching (complex matrices) to Metal, and the speed up has been staggering. What used to take days now takes minutes. It is the hottest my M3 MacBook has ever been though! (See https://x.com/billticehurst/status/1871375773413876089 :-)

Sorry for the delay in replying...

What do you hope to be doing in 5 years? Architecting quantum solutions? Reselling or consulting on cloud solutions? Building quantum applications?

I think the quantum space will have quite a bit of progress in 5 years, but I think most experts in the space (of which I'm NOT one) think we're still over 5 years out before there's broad adoption on running quantum programs with significant business value. (i.e, it'll still largely be researchers and bleeding edge adopters).

Opinions are my own, etc.

I gave a plug for this yesterday, but if you want to try the Quantum Katas from Azure Quantum, it runs in the browser and covers this stuff. See lesson 3 for linear algebra. <https://quantum.microsoft.com/en-us/tools/quantum-katas>

One thing I did forget to mention is that you can play with this stuff in a "familiar to software developers" way in our VS Code playground at <https://vscode.dev/quantum/playground/> . This is a 'code first' approach familiar to software developers leveraging VS Code integration. The playground is pre-populated with a bunch of common quantum algorithms.

You can also install the extension in VS Code directly (<https://marketplace.visualstudio.com/items?itemName=quantum....>), you don't need to run it in the browser, but even in the browser it has a fully working language service, debugger, evaluator, quantum simulator, package management, etc. It's all written in Rust and compiled to either Wasm for the browser and VS Code extension, or native code for the Python package. (I'm thinking about doing a video on how we build it, as I expect it will interesting to this type of crowd. Let me know if so).

Disclaimer: I work in Azure Quantum on the product mentioned. AMA.

If you're a software engineer, then the Quantum Katas might fit your learning style. The exercises use Q#, which is quantum specific programming language.

https://quantum.microsoft.com/en-us/tools/quantum-katas

The first few lessons do cover complex numbers and linear algebra, so skip ahead if you want to get straight to the 'quantum' coding, but there's really no escaping the math if you really want to learn quantum.

Disclaimer: I work in the Azure Quantum team on our Quantum Development Kit (https://github.com/microsoft/qsharp) - including Q#, the Katas, and our VS Code extension. Happy to answer any other questions on it.

I've been playing around with some low-level GPU stuff lately and trying to port some of my CPU-bound matrix multiplication. The challenge has been that in the space I work in (quantum) the matrices are of complex numbers, and for a certain size of problem floats don't cut it and you need doubles. Nearly everything I find for GPUs is targeted at matrices of real floats.

Any pointers to examples? I'd be fine sticking with floats as a first step, but would love to see some (reasonably optimized) low-level GPU code for working with matrices of complex numbers (preferably for Metal or WGPU, which is what I'm using).

It's funny to think back now that JSX had such a strong negative reaction on launch. That's the one thing from React I love - having a declarative and strongly typed markup for HTML in my JS/TS code.

I think React peaked with hooks around 2019. I write most of my front-end with the smallest 'React-like' library possible (Preact) using only functional components, and with as little 'magic' from useEffect etc. as possible.

I like to stay close to 'raw' HTML/CSS/JS when possible, but the web platform APIs are just a little to clunky and verbose for me without the JSX abstraction.

Why do companies hire junior devs? You still want a senior to review the PRs before they merge into the product right? But the net benefit is still there.

I don't think they had Space Grey on the M3 models either. That was initially my preference, but I went with the Black and quite like it.

Multiplying any two complex numbers 'a' and 'b' gives you a complex number z whose magnitude is the magnitude of 'a' times the magnitude of 'b' (that's covered in the article). I always thing of a 'complex conjugate' as a reflection across the real number line (i.e. has the opposite angle or 'phase'), so when a complex number and its conjugate are multiplied the angle disappears and you're left with no imaginary component, thus just the real part which IS the magnitude. (As a^2 + 0 = c^2)

I hadn't worked with complex numbers much for most of my life, but getting into quantum computing recently it is ALL complex numbers (and linear algebra). It's fascinating (for a certain mindset at least, which I guess I fall into), but it is a lot of mental work and repetition before it starts to feel in any way 'comfortable'.

> Apple did not help them design the CPU/Architecture

I believe they had a big hand in ARM64. Though best reference I can find right now is this very site: https://news.ycombinator.com/item?id=31368489

I’m no physicist/biologist, but I always find it odd when they look for water on other planets to see “if life could exist”.

Sure, maybe that’s a requirement for the type of life we on earth know about, but I don’t see why other elements couldn’t have also formed in just the right way to be able to reproduce, and maybe eventually “think”.

ARM64 has a 32-bit word, even though the native pointer size and general register size is 64 bits. To access just the lower 32 bits of a register Xn you refer to it as Wn.

I lived in Australia for quite a while. It's a terrible example of what can happen. There was recently a very good BBC article on it and the connection between sports, gambling, and marketing: https://www.bbc.com/news/articles/cx2yg3k82y0o

Similar to many other toxic things, once the genie is out of the bottle and the government gets used to the revenue it generates, there is a large disincentive to tackling the problem.

Yeah, I’ve never got the “it’s stealth layoffs” or “because we need to use the offices we pay for” or “for the tax breaks” arguments. The math just doesn’t work out.

Staff are such a huge part of costs, and the difference between your good staff and bad staff can break a company. Adopting any policy that would encourage the staff who can go elsewhere (the best ones) to leave and be left with the ones who can’t (the less good ones) is such a huge risk, you wouldn’t take that position for a few million dollars, but only if you believed it’s the best strategy long term.

It is interesting to me though that with so many companies taking this stance lately, many of them heavily data driven (like Amazon), that I’ve not yet seen any actual data or research that says “Here are some quantitive numbers on what we think having folks back in the office is worth”. Anyone got any data on this?

> Who are you to tell neurodivergent employees to suffer?

I found this line telling. I’m wondering if the Hacker News crowd is biased towards the “hacker” stereotype - the socially awkward coder who likes to be left alone and crank out largely independent work.

I have a couple of folks like this in my team, and they are absolutely as good (or better) working from home, as they don’t really talk to others much or contribute in meetings anyway.

But you can’t build a team/org out of those personality types. Much of the creative work and important decisions _does_ happen face to face, let alone the ad-hoc ideation and brainstorming from being in the same space.

I don’t think it’s a requirement that everyone be in the office 5 days a week (I don’t do that myself), but I do see the negatives of letting the team work from home whenever they want and expecting to get the same level of work done over Teams/Zoom calls and email.

Wow. I thought Reuters was an upper tier publisher. That is totally unreadable on my iPhone. The page is jumping all over the place as I try to scroll.

In that vein, posting comments to hackers news is just an auto-complete, or “what should the next token be” problem, right? :-)

Yet here we all are, reading the comments, looking for insights, knowledge, and entertainment.

> was foolishly violating the safety protocols by using a screwdriver to hold the two halves of the sphere apart. When the screwdriver slipped, the core dropped to form a critical mass

I always thought the material had to be forced together at high pressure for the chain reaction to start. Crazy that just dropping it had such dire consequences.

The hours I spent chasing down the edge cases with inline html! There’s some real nuance to the spec there.

You need inline html fairly often for anything non-trivial, and some of the rules certainly aren’t intuitive or easy to remember.

That said, like JSON, it won by being both simple to grok and good enough for most use cases.

I shudder to think of my kids doing the stuff I did as a kid, and sometimes wonder what angel was looking over me that I made it to adulthood.

I don’t worry at all about my 8yo boy however, as he’s very cautious and risk averse. My 6yo girl though already scares me to death with the stuff she does when she slips out of sight. I can’t wait for the teenage years! <sarcasm>

And likewise, it made for a front page news story on the BBC

https://www.bbc.com/news/articles/cp66xdxj0l3o

A click bait headline, but watching the video and what he said, everything seems somewhat reasonable.

If the experts don’t agree on risks, then clearly there’s a non-zero chance.

On some tasks, computing machines passed us by a while ago. I already ask an LLM several questions a day as it has a much “broader” intelligence than I do. Many times the code it suggests is exactly what I was about to write - sometimes better. The capabilities of generative AI today were in the realm of science fiction just a few years ago.

Artificial intelligence will continue to evolve faster than human intelligence. Sticking our head in the sand and acting like we will always be “superior” seems fool-hardy.

I use TypeScript and esbuild for all my stuff. Even then, I often spend a crazy amount of time getting modules working.

Between the various TypeScript module options and various package.json module options (and various code patterns used), modules make JavaScript way more painful than it should be.

I think most of the JS language standards work the past 10 years has been awesome, but modules was definitely rushed and poorly thought though, causing years of frustration.

Last time I tried Babylon even for something minimal my bundles (with esbuild) were just too big compared to alternatives.

I don’t see any mention of size optimizations in the blog post. Anyone have any knowledge on this they can share (else guess I’ll just try again with the new release).