physicsguy

On 1) though, they're an inch deep and a mile wide.

It used to be more common in the U.K. but it’s pretty rare now. Stews and braising were the traditional way to cook it as older ones are kind of tough.

I think they do but only when it's applied to learn some specific spatial task.

We can see for e.g. that frontier models have got really good at CAD sorts of tasks recently. That suggests benchmaxxing RL to a degree but it does mean they've suddenly got quite good at forming simple geometries in a step by step way.

This misses the point entirely of why C is as it is.

It’s quirky because when you had kilobytes of RAM, storing the length of the string was memory you couldn’t afford. Arrays decaying into pointers similarly meant carrying a single pointer around rather than a fat struct. Magic integers for return codes were used because they’d fit into a single register. Compilers didn’t have the processing power or memory to track complex borrow rules or static analysis. They were slow even as they were! Old processor architectures were way more diverse and fragmented than now - and even 30 years ago endianness wasn’t a settled thing.

Things weren’t quirky because people “wanted” them to be, they were quirky because everything was way more constrained. Rust came into a way more rationalised environment and it’s a great language of course but the operating environment is so totally different it’s hard to compare.

Colin Furze - and at least for part of it he didn’t! The council sent him an enforcement notice and he had to get it done retrospectively

Very good!

> I did a PhD in magnetic materials

There are two of us on here!

I was going to say that people encounter ferrimagnets more commonly than pure ferromagnets I think, since as we both know pure ferromagnets tend not to have very high anisotropy.

I got Claude to design me a shed the other day. I asked it to make the door hinged in the CAD software I am using, and it did, but parts of the door weren't attached to each other so some were left floating when I set the door open.

People were definitely asking for it.

Oh this is great, it was one of my biggest bugbears about Go since you almost always have to link C/C++ code to get the appropriate performance.

The one negative I'd say is that often autovectorisation is 'good enough' and this doesn't really tackle that gap.

We're still quite some way off that in the UK. Copper lines still working though due to shut down by next year I think. Lots of older people with no internet left in the lurch with phone lines with that of course.

> I live in a European country, where the phone company hasn't bothered to come remove the old unused copper phone lines in the last 5 years. Thousands of dollars worth of cables running past my house alone, and no one has bothered to go strip it in the night and sell it

Isn't that because they're still flat out on rolling fibre out? There's not that many telecoms engineers and if they spend tons of time removing the old, that slows down the new.

I ran into the exact same wall when doing some SIMD + OpenMP optimisations.

Me: "Parallelise this loop without changing the results."

AI: "I can't do that, because of floating point accumulation."

Me: "So spin out a temporary fixed-length array, accumulate into that thread-locally, and then do an ordered summation of that at the end."

AI: "Oh, you're right!"

It's a fantastic tool for automating the implementation grind, but you still have to hand-feed it the strategy. I guess it will get there at some point.

That is true if you're using them as chatbots but for something behind a product it's largely OK as long as it meets the requirements.

I'm finding that executives at work are generating AI slop and sending it at each other and teams that are then meant to digest it, without any critical thought. Often these are sloppy and contain broad mistakes that aren't relevant to our company's setup.

Only about 15 years later than it should have been introduced

I wouldn’t personally refer anyone I’ve not worked with. What’s the point? Someone being smart isn’t enough, they have to be actually sane and normal to work with. You can be friends with people you wouldn’t want to work with. Someone I’ve not met would be a totally unknown quantity.

In the UK I had a panel interview with 4 people, in person. But for international students who often brought their own funding sponsored by their government that didn’t happen.

I recently tried to get Claude to use Codegraph in a repo rather than using grep/find all the time but I found it didn't follow instructions a lot of the time. I tried putting in a pre-tool call hook and explciitly blocking find/grep, and instead rather than using Codegraph like it was told, it started using Python to find/search instead.

> Growing up in Australia, every computer I’ve ever owned has defaulted to “US Letter” as the paper size.

I'm not sure how; the UK is an Anglosphere country that uses A4 and desktop machines as far back as I can remember (mid 90s) also correctly defaulted to A4 when printing. I suspect it's more likely to be that there's been English (UK) and English (US) options in OS's for a long time and Aus got defaulted to English (US) rather than English (UK) which carried some settings that weren't correct by default.

Na, was normal in the UK to have A4 too

> Anthropic and OpenAI don’t care about Math and the progress in that field: They just picked a domain with a lot of cultural capital that is so complex and abstract that people don’t really understand what’s going on.

I don't think this is true at all. A lot of people who work at places like OpenAI and Anthropic or any research lab really, aren't just CS people, there are a plenty of mathematicians and physicists, etc. and so the problems are interesting to them.

I thought you were going to say about "bob" being a slang word for a poo.

Turns out that's a particularly East Midlands thing!

This is one of those when you go "Huh, this is amazing!" or "Huh, I thought this trick was really well known!" depending on your background ;)

Here's a similar impl from 1980 written in Fortran that also computes other properties like centroid: https://calgo.acm.org/550.zip Algorithm 550: Solid Polyhedron Measures A. M. Messner and G. Q. Taylor ACM Trans. Math. Softw., 6(1), Mar 1980, pp.121--130 Keywords: polyhedron, graphics, numerical integration Language: Fortran 66/77; Shar Index: Z; Gams: P File size: 19.1 KB;

But Messner published it first in: A. M. Messner, "A surface Integral method for computer calculation of mass properties", Paper No. 852, 29TH ANNUAL CONF. OF THE SOCIETY OF AERONAUTICAL WEIGHT ENGINEERS, Washington, D.C., May 1970.

I think

CUDA used to target POWER9 architecture but it was so niche that you ended up compiling from source literally everything since no wheels were available for it. EasyBuild and Spack were full of half broken recipes for the magic incantations necessary to build fundamental packages for HPC like TensorFlow, etc.

One of the big things to think about is whether local LLMs will be things companies want to deploy.

If you think of for e.g. some proprietary piece of software that wants to embed an LLM they've fine tuned or trained, they will want to make back some of their research cost right. So they are not going to want to put this on-device even if the hardware is there, unless there's some way of locking it down. I suspect we'll need on-hardware validation/verification and a way of preventing extraction of weights for this move to happen for many use cases.

What is the performance like for symbolic maths?

I always preferred Mathematica for that to SymPy since it's baked in to the language much more. It was also way way faster than SymPy was for anything interesting.

> Small llms are still way more efficiently server on big GPUs.

Yes, but the privacy aspect means that for many, many applications slower local will still be preferable to faster remote so long as the actual model performance is the same.

Speed cameras are less liked for sure but ANPR is typically used to catch people who have not paid their vehicle tax or who have not got their annual car safety check done (MOT) but are continuing to drive

I find all the reporting on cameras in the US quite interesting given we’ve had ANPR cameras in the UK for 20+ years