noone_youknow

I’ve been trying out various LLMs for working on assembly code in my toy OS kernel for a few months now. It’s mostly low-level device setup and bootstrap code, and I’ve found they’re pretty terrible at it generally. They’ll often generate code that won’t quite assemble, they’ll hallucinate details like hardware registers etc, and very often they’ll come up with inefficient code. The LLM attempt at an AP bootstrap (real-mode to long) was almost comical.

All that said, I’ve recently started a RISC-V port, and I’ve found that porting bits of low-level init code from x86 (NASM) to RISC-V (GAS) is actually quite good - I guess because it’s largely a simple translation job and it already has the logic to work from.

Definitely going to build this, I’ve been worried I might miss a nuke going off for a while now and this looks like just the thing!

For me, it’s not the typing - it’s the understanding. If I’m typing code, I have a mental model already or am building one as I type, whereas if I have an LLM generate the code then it’s “somebody else’s code” and I have to take the time to understand it anyway in order to usefully review it. Given that’s the case, I find it’s often quicker for me to just key the code myself, and come away with a better intuition for how it works at the end.

… and introduce something like virtual memory?

IMO the mystification section is right on the money, I couldn’t have said it better myself (and believe me, I’ve tried).

This is simply amazing work, I take my hat off to you. And not just the threading in an emulator (which is a genius concept all by itself) but for the article itself - I have spent years trying to find the right way to articulate exactly your points in hopes of engaging people to learn more about the systems they’re building on top of.

From now on, I’ll simply send out a link to this. Thank you!

Another great article! I always enjoy reading these.

That additional type ‘a’ register, supporting copying directly from the type ‘b’, is intriguing. I can well imagine the article to be correct that it’s a place to stash ESP during interrupt stack setup, specifically if there’s a privilege transition since I would think it’d be easier from an implementation point of view to “simply” copy ESP to that special register, load the new ESP from the TSS, and then stack that special register after stacking SS using the usual stacking logic.

Though of course that then raises the question of how SS itself is handled, since something similar would be needed there I think.

Sounds like you’re doing some interesting stuff and have a good, varied skill base to build on.

My advice would be to jump in and start working on kernel level stuff, or writing your own - IMO there’s no finer way to really “get” the low level concepts and the understanding you’ll build will really help with any other system-level stuff you do.

Not to plug, but if you were interested in getting involved in an existing project, my own toy kernel project[0] is at a point where there’s still lots of fun stuff left to do (both design- and implementation-wise) but a lot of the basic “project plumbing” and one-time machine setup stuff that people often get stuck on is already done, and I’d be glad to have the opportunity to share knowledge.

[0] : https://github.com/roscopeco/anos

The docs for intel’s integrated offerings seem pretty good, I’m planning to do a driver for those on my toy OS. Anything else is hit and miss in terms of availability of information unfortunately.

I work at one of those companies (not google) and Polypane is actively banned as “not safe, shouldn’t be used at all”.

Not to take away from the product, which does look cool, but the “customer list” is definitely sketchy.

Everyone has to start somewhere, and many who try don’t even get this far (I’m not sure I would have at OP’s age either).

You could build your own Motorola 68k computer:

https://rosco-m68k.com

(Caveat: I’m the founder of the open source project and I own the company, so obviously biased - other options in this space are available :) ).

Well, what I had in mind by “certain” is probably really “not me”. I’m fully aware how easy it is to fall foul of the ad-friendly guidelines, and have had more than one video demonetised for “reasons” myself. I’m also very aware that tying one’s entire revenue to a single platform isn’t a good strategy in _any_ business, it’s not limited to YouTube (but I can see an argument for it being worse there specifically).

I really do recognise the issue, being in it myself. I do have patreon (and others) for other projects and it’s another revenue stream, which is great. But for my YouTube main channel I believe the content itself has value, and having to pour time and resources into building a value-add package devalues it - both in the immediate (since I would now have less time to devote to content creation) and longer term (since it makes it essentially a leader for my value add packages).

(Some larger creators I know do manage to carve out some revenue on patreon etc without any “perk package” but I think for that to work it becomes even more of a numbers game, and won’t help small creators just getting started. I’m also putting aside the recent announcements ref. The App Store etc since they’re not directly relevant here).

That’s great, I’m glad that you’re supporting creators directly and getting value from it too. But unfortunately you’re in the minority in my experience, for every person who does this, there are hundreds who wouldn’t even consider it.

For creators making certain kinds of content the “uncensored” and “non-ad-friendly” topics are a great argument for direct sponsorship etc, I definitely agree.

> But another reason why sponsor reads annoy me is that it breaks the youtube premium deal.

I totally get that, and I feel the same way when I see yet another read as a viewer and premium subscriber.

I don’t really have an answer (and if I did, I’d be doing it already), but I will say that my (subjective, based on my ad placement strategy and viewer profile) experience is that premium views are worth less than non-premium - although YouTube cleverly don’t actually give me enough data to _know_ that as a fact (and it would go against their stated position, which I guess they would never do).

Well, that’s your call, of course. And when it comes to regular YT ads I don’t really blame you, “the algorithm” and the way monetisation works encourages us to set up aggressive mid-roll placements etc that must be incredibly annoying if one doesn’t pay for premium.

One of the nice things about sponsor segments is that they don’t involve YouTube, so the creator gets more benefit from the deal, but of course done badly (and I assume this must be the case with many of the generic irrelevant VPN ads for example) they will harm retention and thus limit reach.

Your “whole video” suggestion is really “advertise smarter” IMO, which I completely agree with. Personally I’ve never done a “reading a 30-second script about how great product X is” type segment, but I have done videos where I try out “product X” in some way that’s relevant to my audience. It’s more product placement than direct advertising, but I guess even that is unpalatable to some.

Sure, I totally get that. I’m no fan of being advertised to myself and as a premium subscriber I do find sponsor segments - especially poorly-places ones - just as annoying as everyone else when watching YouTube - which is why I said I was conflicted in my earlier comment.

However as I mentioned in another reply in this thread, removing routes to monetisation and devaluing content in general (by making it be effectively a loss-leader for value-add sponsorships or memberships) will only have the effect of making YouTube non-viable for many, and especially those who necessarily have higher production values to make better quality (I’m thinking more thoroughly-researched, more interesting, that sort of thing) content.

> Don’t do your videos for money.

This is of course a valid suggestion, and there are many, many creators that do this. However I think the world would be a poorer place if we lost all the creators that do need to make _some_ money for their channels to survive, which IMHO is the natural endgame if we remove or block all routes to passive monetisation.

I do get the issue with premium, as a premium subscriber myself I too find it annoying to be interrupted by yet another 30-second (or increasingly, more) read for some shady VPN or whatever.

Channel memberships, like patreon etc., are an option, but have a vanishingly small rate of uptake, and people expect some sort of value-add in return (early access to videos, a discord, and so on). Without other routes to revenue this just devalues the content itself, which I feel may be part of the problem here - we no longer value attach value to quality content. Rick Beato made a great video on the effects of this (in the music industry) recently, and it’s not great - but it does feel like all media is going a similar way.

As a YouTuber, I’m conflicted about this. My main channel (non-tech) is small, but is monetised, and YouTube see fit to throw me a _very_ variable amount of money every month. CPMs are down right now so revenue has tanked along with it, it’ll pick back up at some point, but the variability is itself the pain point. My videos are relatively expensive and time consuming to make, but people seem to find them useful, and even enjoyable. The occasional (relevant) sponsor read or similar has been a huge help in providing some stability in the past, and I know for many channels it’s the main source of income since YPP revenue share can be so volatile.

I do worry that if this takes off it will just result in those sponsors pulling their budgets for this type of advertising, and it’ll be another nail in the coffin for creators. Sure many of us also do patreon etc but that’s never really sat right with me personally (and see also the post on HN just today about Apple coming for a revenue split there for another creator-hostile storm brewing).

On the other hand, I totally get the hatred of “the usual suspect” sponsors (VPNs, low-quality learning platforms etc) that get done to death because of their aggressive sponsor budgets and not-unreasonable deals. Those get shoehorned into a ton of videos and it’s a shame, but a blunt instrument like this is likely to kill off sponsorships as a whole, not just those bad ones.

Sounds to me like the author has the “expectations problem”. I’m not a huge fan of type hints (though they are useful especially with modern IDEs) but it seems a stretch to say a thing is inherently flawed simply because it doesn’t work the way I expect it to…

The trouble with CPLDs in my experience is the software environments and (for some at least) programming hardware.

Especially for the Atmel 15xx parts, the software is windows-only, old and annoyingly crash-prone. The GALs are at least programmable from modern OSes with commodity programmers.

That said, the CPLDs themselves are good chips, and I don’t think especially inappropriate for a vintage project

For a simple 8-bit system such as this, even if “no GALs” is a requirement, all the chips one would need can still be bought today, brand new, and are still in production in one form or another (e.g Alliance SRAM, SST flash, 74HCT logic etc).

Worth noting that 16V8 and 22V10 GALs are still being made too, and I’d argue that programmers for them aren’t exactly “outdated”.

Not that I have anything against this design, I’m always happy to see a new SBC, though it could be argued that it still “hides” more of the bus inside the MCU than would perhaps be ideal.

Nice! I’d be interested to see how it handles https://github.com/rosco-m68k/rosco_m68k , it’s a mixed software / hardware repo, with a lot of code in assembler and C (for an old platform). Might be a challenge?

Thanks I couldn’t agree more, the m68k is a great architecture, powerful enough yet simple enough that a person can keep the whole thing in their head.

That said, we _do_ have a slightly-related 6502 too https://github.com/rosco-6502/rosco_6502

Probably not exactly what you’re looking for (since you mention “outdated”) but there are similar projects with 16 and 32 bit architectures, e.g. https://github.com/rosco-m68k/rosco_m68k

(Disclosure: I’m the designer and lead developer on that project).

I would second this, if you really want to learn hardware then Arduino (etc) won’t help as much as jumping into something a bit lower level. Ben’s kits and videos are a fantastic resource, and you can progress to more in-depth stuff like rc2014.co.uk or rosco-m68k.com (yeah, shameless self-plug at the end there :D).

For me, it’s still the remote thing. I’m a principal, couple decades of experience, I’m still seeing quite a few approaches on LinkedIn but almost all are wanting a few days a week / month in some “fantastic” office in some city. Nope.

This kinda comes off like you’re trying to ridicule the comment or minimise their concerns, but the fact is that, if you’re going to do a data grab, you absolutely should have a way for people to have you remove that data.

The fact you freely admit there’s no unsubscribe link (so I guess no way to have you remove an email address from whatever list you’re building) is not a good look at all IMO.

Personal data (even if it’s just email addresses right now) are probably a thing you want to respect, and be seen to respect. Personally I’m glad my spidey-senses kicked in and made me hit “cancel” on the oauth before I came here to read the comments.

I don’t know if I qualify for this, I’m a dev in the day job but my channel is non-tech - after years of tech-only side projects I wanted to do something different. Here’s what I’ve learned so far.

* you have a lot of great advice about audio on the comments here - and they’re right, decent audio is a must. I use a DJI wireless lav for most stuff, but also have a Rode NT-USB for tabletop stuff and their shotgun mic on-camera for sync audio and quick pieces to camera when out and about. I usually record main audio into audacity via a behringer audio interface.

* get a good editing program, and get good (which means efficient) at using it. You’ll spend a lot of time in it. Personally I use the studio version of davinci resolve, other opinions are available.

* it can be feast and famine. I started my channel in Jan this year, and hit 1000 subs / 4000 hours at Easter. Monetised a couple days later. I’ve earned about twenty-five pounds since then - it’s gone quiet, but it’ll pick up again.

* don’t pay too much attention (or really, any) to the VidIQs and Channel Makers of this world - they just want to sell you snake oil (but their videos are often entertaining and you can learn stuff from what they do - just not so much what they say).

* don’t bother with VidIQ or tube buddy or all that other stuff - spend the money on storyblocks, a better camera, lights, audio gear instead.

* decent lights are worth their weight in gold. Aputure are great, but expensive. Smallrig do very similar kit at half the price.

* probably more specific to me, but filming long-take 4K in a hot kitchen with all the equipment going will challenge even the best heat-management system in your camera

Also FWIW I don’t do shorts any more. I experimented with them and the dopamine hit is great, but I’m a long-form guy. Shorts just skewed my analytics and brought in low-quality subscribers…

How is this even a “show HN”? Literally just a thinly disguised money/data grab.

Nope.