>Two weeks ago, Rich had posted a video on installing Windows 11 25H2 with a local account. YouTube removed it, saying that it was "encouraging dangerous or illegal activities that risk serious physical harm or death."
> A weeks ago, my friends and I were talking about the inner workings of Zen 5. We were talking about how the CPUID instruction works, and how AMD MSRs are technically editable if you ask the processor nicely.
I don't know about your friends, but as we're on HN, I'm sure others here have friends like mine, who absolutely have conversations about how the low level shit that facilitates our world works.
1. Start building stuff that is hard to build that requires touching these niche topics. Especially stuff you don't know how to build
2. As you encounter problems, you'll have to scour for solutions (AI doesn't know these things due to lack of training data). In the process you will find people who are also working on these problems. Ask these people well-formed, intelligent questions.
I am skeptical of your second claim here… if you can “scour for solutions”, and you find something about it on the internet, then AI could find it the same way.
Just cause the AI could find the info definitely does not mean it will find and apply that knowledge correctly to solve a problem.
I find AI shockingly bad ad searching the web, as SEO blogspam sites heavily pollute AI context windows, while relevant and important resources are typically very densely presented reference material which must be constantly revisited.
It doesn’t need to. It has already all the fundamental knowledge it needs. Just set it up on a system with an editable proc file system and it would be able to figure it out.
A lot of the solutions are buried in places AI can't scrape or train on. Like inside people's brains or inside private codebases or chatrooms not open to bots. However you can find these people and the products and services they're making and start talking to them.
Most LLMs can't even count parentheses properly to build basic Lispy stuff. Building something niche like a logic solver in Scheme macros only? Forgetaboutit.
By 'talk' I suspect he means discord and by friends he means display names. Not that there's anything wrong with that. I catch myself saying 'talk' when I'm talking about something a friend told me over chat.
>> A weeks ago, my friends and I were talking about the inner workings of Zen 5. We were talking about how the CPUID instruction works, and how AMD MSRs are technically editable if you ask the processor nicely.
> As do we all.
I think they interpreted “as do we all” as pointing out humorously that this is an unusual friend group. So, speculation that it might have formed online makes sense, because online spaces can sometimes facilitate that sort of thing.
Presumably because they all traveled there for a temporary part of their lives. And after university, they presumably scattered to the places where they built their careers and families.
They are two different situations but why is the distinction meaningful here? I rarely even remember the venue of most conversations, just that it happened.
Because for a lot of us it’s hard to imagine finding a half dozen or so people who can talk in person like that outside of a conference or workplace. Discord facilitates that because it assembles people based on interests and such. You’re just going to more easily have that kind of conversation than you would “out in the real world.”
My guess is they are functionally saying “this probably happened on discord if anybody is wondering how this is even possible and not made up for effect” but I might be interpreting too much
My friends and I share interests but they can’t all talk about the relative pros/cons of full frame vs. cropped sensors in digital cinema with me. That’s kind of the framing here if that makes sense. We share a lot of interest and can talk in depth about certain topics, but there are plenty of topics that I am interested in or just know a lot more about that none of them can really discuss with me, so I have to find those communities elsewhere
OP here: Yep, it's Discord. This all happened on a Discord server for a tech youtuber, I'm not sure if they want to be "outed" but they're an 800K subscriber-ish youtube channel with a five-ish thousand member private discord. There were a bunch of people involved, but you might see Arae around in the Reddit thread, she was the one with the actual 9700X in question. None of us have met IRL and we do stay fairly anonymous, but we do chat quite regularly (I've known Arae in particular for at least six months now).
If you're looking for similar discords, I might recommend the discords for things like Bazzite, LTT, Mint, or any number of other small-tech-youtube-discords, or discords for technical video games (eg. Turing Complete, BeamNG, PCBS, Factorio). Discord has no algorithm, you have to find the content yourself!
I'd just close it without comment. Or maybe if I'm feeling really generous I'll make a FAQ.md that gives a list of reasons why we'll close PRs without review or comment and link that in the close comments. I don't owe anyone any time on my open source projects. That said, I haven't had this issue yet.
That's fine for an open source project, but many many companies are mandating AI use, they're putting it in performance reviews, they're buying massive Cursor subscriptions. You'd be cast as an obstructionist to AI's god like velocity ™.
This is the same process used to port GHC to new architectures, like ARM. It is not easy to do, I don't know how many people can actually do it. But its possible and has been done multiple times.
Most mainstream languages have a fairly straightforward bootstrapping process that doesn't rely on a trusted binary. And yes, most distrubutions ignore that, but nonetheless it is possible to use those languages in a high-sec environment if you put the work in.
I'm not sure that I agree that GHC can't be bootstrapped though. There is a process for porting to other architectures; its not an automated process and perhaps no one outside the GHC team can actually do it, but if for some insane reason NSA decided they want to use Haskell I'm not sure that they actually can't, if they put a lot of work in and hire GHC committers with high security clearances.
The burden of evidence here is on you. They don’t need to prove LRMs can’t scale to meet these problems; their only claim is current models can’t handle these problems. Others will take this up as a challenge - and chances may be good they will overcome it. This is how science works.
They can’t claim current models aren’t able to handle these problems if they didn’t use a setup similar to coding agents like Claude Code and OpenAI Codex. Using a suboptimal setup is akin to verbally telling a person the whole reasoning problem without letting them write down notes and expecting them to memorize and solve it after only hearing it once.
The thing they are testing for is reasoning performance. It makes sense to not give tool access.
This is same as the critiques of the LLM paper by apple where they showed that LLMs fail to solve the tower of hanoi problem after a set number of towers. The test was to see how well these models can reason out a long task. People online were like they could solve that problem if they had access to a coding enviornment. Again the test was to check reasoning capability not if it knew how to code and algorithm to solve the problem.
If model performance degrade a lot after a number of reasoning steps it's good to know where the limits are. Wheather the model had access to tools or not is orthogonal to this problem
If the models can’t do it they can make that claim. If you want to make claims about agents then design that experiment, collect the data and write a paper. That is how science works.
Its an entire desktop environment, its not as simple as choosing between two different apps. Although people who make this complaint should probably just use KDE, maybe they've used Gnome for a long time and don't want to change.
reply