torginus · HackerLangs

il y a 9 heuresdiscuss

I thought linear algebra was pretty much the poster child of applied mathematics - the entire field was invented to represent computations in a regularized form to feed into computers. Well not really, but much like Boolean algebra or the Fourier Transform, it was pretty much a curiosity until computers came along.

torginus

il y a 10 heuresdiscuss

Huh. I never knew that was a thing. Although sounds cool. Though if you do that, or take something like a ncurses UI, and start turning it into something nicer, at what point do you have something that's not a terminal?

But I guess there's promise in having something that's like a Jupyter notebook, but for files.

torginus

il y a 14 heuresdiscuss

I use the terminal that happens to be at hand, which is usually the one that ships with the OS, or the one in VSCode. I don't have high demands on the terminal, and I don't understand what problems are there to solve, as long as it handles unicode and control sequences correctly.

torginus

il y a 14 heuresdiscuss

Its academia. Rust is admirable for having been able interesting programming language research into a mainstream language, but this had the implication of bringing the people who did said research with it into the communities.

It used to be that programming languages were mostly boring and predictable, with maybe questionable semantics (const etc.), but generally that messiness meant they were good enough at getting the job done.

PL research and theory focused on mostly FP, Ocaml family and other functional languages, with things like advanced type inference system, based on postgraduate category theory. These people have fought endless and bitter mental battles with each other, a glimpse of which occasionally leaked to HN. Some paper about a noteworthy discovery in solving a problem incomprehensible to the general public. Some guy complaining about how he tried to educate average programmers about how unsound their programs were, and being taken aback at how these people didn't want to be saved from their own stupidity. Some article complaining about how if every programmer was just 15 IQ points smarter, they'd be all doing FP. But mostly this community kept to themselves.

Thanks to Rust, all these ideas have found purchase in real practical software. Now the academics get to torment themselves with the moral duty of saving the everyman from using less theoretically sound programming languages.

(Disclaimer - I don't hate Rust, I think it's great they made this breakthrough from academia to regular boring programming, but they need to respect the nuance of the world that exists outside the walls of research institutions)

torginus

il y a 15 heuresdiscuss

I remember watching this urban exploration video, the guy went into an abandoned Russian nuclear bunker, deep underground. Watching this titanic effort of engineering, made by people who were both highly intelligent and had vast resources at their disposal, yet felt it necessary to build it to have an answer to the unthinkable horrors the future held, all of this reflected in the sturdy but utilitarian design of the concrete complex, rooms filled with all sorts of pumps, air filters, and electronic equipment necessary to sustain human civilization after the bomb fell. As the guy walked from room to room, he noticed that in one of the rooms was a set of old PCs. The power was on. He switched it on, and suddenly the familiar bootup chime of Windows 95 played, and he was looking at a desktop. He sat down, started clicking around, opened Solitaire, started playing. Suddenly all the tension dropped, I forgot where I was. The whole thing felt comfortable, even pedestrian. I had to actively remind myself that the guy was many stories underground in an abandoned bunker, likely patrolled by active military.

Computers even at their crudest have a hypnotic ability to bring you into their world, and somehow make you forget about the reality you live in. This is not the only mechanism of society that does this, but certainly one of the most powerful we found in recent history.

torginus

il y a 15 heuresdiscuss

This is my opinion as well. Powershell and object shells are much better and the way to go imo. Also Powershell code is fast owing to running on the .NET VM.

But the actual UX is just convoluted and horrible. For example, the person who decided that commands without a ; at the end of the line should just vomit their outputs to the enclosing function's clearly hasn't heard of the principle of least surprise.

PS just does so many plain weird and unintuitive things, which is worsened by the fact that it looks like a boring old programming language, while clearly not being one.

torginus

il y a 18 heuresdiscuss

It's just how LLMs work - these are three completely separate models trained in parallel, with different numbers of parameters and using different amounts of compute.

They could hide this behind a harness that picks the correct model for you, but devs don't seem to like that.

There's also the 'effort' slider, which I guess how many experts in the MoE are evaluated and how long reasoning chains are allowed to go on, which is the 'smooth' scaling you are thinking of.

torginus

il y a 18 heuresdiscuss

I keep thinking about the cyborg chicken mechs in Metal Gear Solid 4. If stuff like Neuralink gets advanced enough, we might find that cutting up animals and sticking electrodes in their brains is the best way to make robots.

Yeah, people might object, but it can be argued that we are already subjecting scores of animals to horrors beyond comprehension just to get a bucket of chicken wings. And even if we manage to get silicon to do what brains do, it will likely cost 1000x as much and consume 1000x the power like you said.

It's hell of an economic incentive.

torginus

il y a 19 heuresdiscuss

The irony being, that youtubers and Twitch personalities have almost certainly played the games they are reviewing.

But they are considered at least a 'rung' lower than games journalists when it comes to how much access they have, and how much their opinion matters professionally.

torginus

il y a 19 heuresdiscuss

> I'm not too mad about it not following my instructions

I'm a bit concerned about this - starting with GPT5, AI labs started doing this 'complete app from a prompt' sizzle demos. When I started working with GPT5 - which was supposed to be a qualitative jump, just like Fable is now, I tried to do a frontend, and discovered that it gave me a CSS-animated purple-blue interface with embossed buttons, gradient backgrounds and dropshadows.

It looked very cool, but it was a bit overwhelming (also broken), and I was really looking for a pedestrian Bootstrap job.

It required not inconsiderable amount of wrangling for GPT5 to stop doing this. So I don't really like the idea that these models have tons of implicit and hidden behavior, to 'soup up' pedestrian prompts.

torginus

hierdiscuss

Dunno, personally, I've had 5.5 lie to me. Which was a common occurence with weaker models, but I haven't really seen it that much with strong ones.

Basically what happened was there was a task which could either be solved in a simple way that yielded poor performance, but there was also a highly involved solution. Both Opus and 5.5 identified the steps necessary for the hard solution, but Opus gave up.

5.5 claimed it did it, but looking at the code, there were parts that were missing based on the plan. Basically 5.5 gave up halfway, did the easy solution, and claimed it was done.

torginus

avant-hierdiscuss

Because you are paying for silicon, and processes, not transistors. Wafers have a certain cost, and litographic processes can illuminate a certain X mm2 of dies in an hour. If a transistor gets smaller, more of them fit in a certain area.

Granted the machines that make them become more expensive, but that's capital expenditure, which gets amortized as time goes on.

So there are two forces here working against each other.

torginus

avant-hierdiscuss

With regards to RAM price I never understood the following: A 16GB RAM stick has 16*8=128 billion bits, with 1 transistor per bit, thats still 128B, yet its supposed to cost like $60 before the price hikes? In contrast, a 5090 GPU was $2000 (true it has RAM, but you're paying for the GPU ASIC really, I guess the rest of the GPU was less than $500), it had 93B transistors.

GPU transistors are smaller due to the more advanced process node (cost per transistor metrics aren't really clear, if they improve on advanced node or not, but I'd say they get cheaper as they get smaller, as technology costs are amortized).

I'm sure both RAM and logic use a process that is quite similar in both inputs and manufacturing steps. So while RAM is a commodity product, this insane price difference didn't make any sense.

So I guess when those fundamental inputs become a constraint, it would make sense for $/transistor move closer for both, which is a massive hike for RAM.

torginus

avant-hierdiscuss

From playing quite a bit of KSP, I'd say an issue is that realistic space travel consists of a lot of waiting to get to the right part of the orbit. You have time dilation for that in SP, but not in MP.

torginus

avant-hierdiscuss

I have a cheeky idea - imagine you're a company (especially a games company, whose engine is a small part of the whole product). You want to cut dev costs by AI - problem is your 100k codebase is completely unknown to most LLMs. So you release all your code, and labs will gobble it up and train on it for free.

So come a year maybe, suddenly ChatGPT or Claude will know the code like the back of its hand.

torginus

il y a 3 joursdiscuss

What's the legal status of 'your' car spying on you if its a company or private lease? After all its legally not yours, and they can put a lot of things in the contract.

torginus

il y a 3 joursdiscuss

I distinctly remember that there was an article years ago about this. Automotive bodies have been mandating features for decades, and were doing followup studies.

The introduction of seatbelts, ABS, ESP all came with noticeable drops in accident rates.

However it was noticed, that these new driver assist features didn't reduce the number of accidents accordingly. These systems are not new. We know they don't really help.

Why do this?

torginus

il y a 3 joursdiscuss

What happens if you bypass these systems? Does it have a separate speaker or does it use the infotainment?

If the former, you can bypass it electronically. If the latter, I'm sure you can mess with the software of the multimedia head unit to silence the chimes.

Will your car fail inspection if you do this?

torginus

il y a 3 joursdiscuss

Dunno, if I were to backdoor a piece of my code, I would definitely put in an exploit instead of a deliberate bypass.

Plausible deniability is important.

A lot of the stuff I worked on already had glaring issues like that without me having to add it..

torginus

il y a 3 joursdiscuss

I have done this accidentally at least once - we shipped a full-stack app, and telemetry started lighting up that on certain older phones and browsers (no points for guessing which brand and browser), the release version didn't load. The minifier did something in the release build that it didn't like.

So after a quick test, it was decided to deploy the debug version of just the frontend as a bandaid. Next day we saw we managed to deploy the debug version of the backend with admin stuff like this as well..

comments