HNHacker News
TopNewBestAskShowJobs

godelski

28,841 karma · joined September 22, 2016

[ my public key: https://keybase.io/godelski; my proof: https://keybase.io/godelski/sigs/ITXd1PN0tKNiQYxLgpyMs4xX2HIQy0uyX-09Z4DbV48 ]
submissionscomments
godelski··on Anti-patterns in software blogging

  > At least in machine learning
You should read math papers lol.

Honestly, I feel the opposite. But maybe that's because my doctorate is in ML. But we might also have a different definition of what the hallmark papers are. Many of the well known papers are ones that model scaled an existing method (ViT and ddpm are good examples). Important papers, but in which sense? I do think we have space for all types of papers and written to different audiences, but I don't think we should punish papers just because they're written to their niche domain. In the past decade there's been a huge increase in citation farming and I don't think that helps. Math is probably too much on one extreme but I think ML is on the other. You have to write your papers so a first year PhD can understand, because they're the ones reviewing your work. I don't think that's a desirable outcome

godelski··on Anti-patterns in software blogging
Papers are written to peers. It shouldn't be a "murder mystery" for them (that doesn't mean it shouldn't tell a story). You have to write to your audience. If a paper was written to a general audience, instead of peers, every paper would have to be a novel.

I'll admit that sometimes I'm frustrated with HN comments on arxiv articles. Despite being a nerdy forum most papers aren't written for this audience. Hell, I have a PhD and most papers aren't written for me. Nor should they. I'm perfectly okay if I don't understand a paper. Arguably the hardest thing about a PhD is learning to be comfortable with being dumb and willing to work on becoming less dumb (you're never not dumb, only less dumb)

godelski··on Anti-patterns in software blogging
My PhD advisor said something similar, but he also said papers tell a story. There are a lot of stories that aren't murder mysteries nor suspense novels. Your paper shouldn't be a novel and especially not a suspense one
godelski··on Anti-patterns in software blogging

  > education is not storytelling 
I actually disagree. We're humans and we love stories. They keep is engaged and are a good way to compact information. But it also is highly dependent on how it is done. Truth be told, we're constantly telling stories and I think part of the problem is not being cognizant of this.

  > People want to save "twists" and "revelations" for maximum impact and it's harmful.
Maybe I'm misreading you. "Plot twists" would certainly be the wrong implementation. But when teaching one of my goals is to get students to be trying to ask a question that I'm about to answer. That's the story part. I'm leading them down the path and they should have suspicions about what lies around the next corner.

So basically a goal is that I'll stop for questions at some point and if I'm successful there'll be a student that asks a question we've been building up to and that I'm about to ask and/or answer. But there shouldn't be a "plot twist". I'm not trying to surprise students (usually :), but rather the story should make sense. As an educator[0] my goal is to get students to remember certain things and ask the right questions (a skill that's non-obvious yet people are overly confident about their level).

A good example of story driven writing is Cropper's The Quantum Physicists: And an Introduction to Their Physics. I really enjoyed this in my undergrad because it not only taught the physics but taught it in a historical manner. I said there's compression in story telling and this style of teaching did a great job at motivation of why these physicists were asking these questions. It gave context. I honestly think it had a big impact on me and helped push me into research (I do ML, not physics).

Certainly there's a time and place for being terse and hyper fixated on the information you're trying to teach but often we're trying to do a lot more than that.

FWIW, I think LLMs are terrible at this. They make everything seem "exciting". When everything is "load bearing" then nothing is. They're really bad at the context and understanding the bigger picture. They're especially terrible at responding to a question with something like "keep that question in mind, we'll come back to it in next week's lecture"

[0] I'm currently in industry

[1] https://archive.org/details/bwb_S0-BVP-068

godelski··on CS240 AI Cheating Retrospective

  >> Do not look at other students’ homework files and do not work with other students. At all.
  > This is a horrific policy. We work better with other people.
I think that line was in connect with quizzes, which are presumably online. Near the top of the article it says

  > You may discuss assignments in a general way with other students, but you may not consult anyone else's work.
I'd be with you if they made everyone work in "clean rooms" but that is appropriate for quizzes and tests.

As someone who's graded lots of undergrads I can tell you that cheating is prolific. It's easy to identify who is copying code, but it is hard to prove. My wife teaches in another department and says the same thing. AI use has grown exponentially and it's really destructive to the education process. Though I have more complex thoughts on this and that it is just exacerbating larger problems that no one wants to actually solve because they'd be very hard to solve in a way you can have clear metrics for the bureaucrats (including industry). Proper evaluation is unfortunately always going to be fuzzy. Metrics are in contention with making a meritocracy, especially if you don't have a deep understanding of said metrics

godelski··on America.gov

  > To enhance the status quo you don’t need LLMs, you need people who care.
Pournelle's Iron Law of Bureaucracy seems to always come into play. The second group tends to win over time as their objective is easier to satisfy
godelski··on Dots: Always-on agents
Is everyone here actually a nerd? Or do they just work in tech. You might think they're the same but plenty of people wouldn't. There's certainly different types of nerds
godelski··on How to keep enjoying programming in a world of LLMs

  > You have a previous generation of cars where people just enjoy working on with hand tools, and there’s modern cars where people like to tune with software patches.
I think this is a prettier picture than reality suggests, it's overly simplified and a statement people like because it's a variant of "everything is the same"

I think the analogy with cars is apt, but we should also like at what happened with cars. People are being locked out of fixing them, even for simple jobs that would usually be task that introduces someone to the hobby or career. Fixes like changing your oil or brakes. Remember that the Ford CEO said that cars are "too dangerous" to fix. Probably connected with the other thing they talk about... a $100 bn repair industry that they want a bigger piece of. There's an easy way to do that, John Deer and Apple have clearly shown that it's not hard, even when we get the right to repair. Is it any wonder far fewer people work on their cars these days? Ask the people with electric vehicles... there are people out there, just as there always will be, not the numbers do matter

Weirdly, I think LLMs would be much more useful if everything was open source. No need to hack your microwave to give it a firmware update to the door doesn't lock, just patch. More training for the LLMs and turns what would be a specialized task into one anyone can do if they point the LLM at the code. Plus, people submit patches to the source code, getting companies free work. There is a way everyone can win in this. Sometimes (often) being greedy prevents you from getting a bigger fill

godelski··on U.S. appeals court upholds designation of Anthropic as supply chain risk

  > Just because you disagree with a power of the government does not mean the government does not have that power.
Sure, but we live in a democracy, not a dictatorship. The government doesn't just get to do whatever it wants. People and corporations have certain freedoms. One of those freedoms is not being compelled to work with the government. The government can't force a company to work for them except under extreme circumstances. Which currently no extreme circumstances have been claimed and no one is talking about nationalizing (the act of forcing the company to work for the government) anthropic
godelski··on U.S. appeals court upholds designation of Anthropic as supply chain risk
And then another company got the contract, claiming they got all the conditions (and more!) that the original contractor was sour about.

I don't know how anyone can look at the situation and see anything but an abuse of power. The government could have just as easily contacted with another provider. This whole thing is vindictive

godelski··on California is chasing wealth that has feet

  > A forest is much less efficient than a mall and yet, the forest might still be more important.
This is a good example of a very common problem: data isn't objective, it needs to be interpreted. Metrics will give you information, but they aren't the full story. That's why Goodhart's Law is so prolific. You can't just look at data and act on it without context. It depends what your actual goals are. And a huge part of that is that we have to consider how much we value things, especially things that haven't already been assigned monetary value. Sure, we can assign monetary value to things like a forest (economists do this), but it would also be wildly inappropriate to just accept those estimates as cold hard facts void of interpretation too. What's the saying? Reality has a surprising amount of resolution.

In a weird twist of irony our efforts to be lazy end up costing us a lot of work. But that's also because there's two types of lazy: short term and overall work. We used to say we want to hire programmers that are lazy because they'll find the most efficient way to do something. But now we don't revere that kind of lazy, we like the kind of lazy that procrastinates. Do the quick cheap thing now, telling ourselves that we'll make it better in the future, knowing that's a lie. That pattern isn't unique to programming, it's just marshmallows.

godelski··on Early rogue AI agent activity and attempts to hack found on urlquery.net
Let's not forget that there's tons of people who are in, or have gone to, jail because they created some computer worm that ended up doing way more damage than intended.

Not to mention while causing *FAAAAR* less damage[0]

[0] https://en.wikipedia.org/wiki/Samy_(computer_worm)

godelski··on Early rogue AI agent activity and attempts to hack found on urlquery.net

  > you are already assuming that OpenAI is intentionally using unaligned agents in these evals or training runs
Uhh... yes. By definition. They are training. That is part of the alignment process.

But also none of that really matters. They clearly weren't monitoring what should obviously be monitored. I mean one of the hacks was performed by the agents editing /etc/hosts. That makes nearly every linux user a "hacker" by that metric. I don't think anyone technical can look at the postmortems and not come away thinking that their sandboxes were woefully inadequate. I wouldn't even consider myself a security person but simply as a long time linux user I can say that it is insane to just let agents have superuser access in their containers. That's asking for trouble.

Look at the rogue wiki stuff too. This was supposedly done during the agent's "down time". And you're not monitoring and there's no flags being raised when agents keep making requests to some random site? If you were training these things responsibly you'd be watching them like a hawk.

I'm not saying "mistakes don't happen" but for a company whose CEO is constantly telling everyone that their product has a high likelihood of killing everyone in the world you think they'd have better security than your average high school.

godelski··on VSCode's SSH Agent Is Bananas (2025)
Related, netrw (vim) can do quite a lot of things that I think people don't realize. I mean not just that it supports file trees, but well... just open :h netrw
godelski··on VSCode's SSH Agent Is Bananas (2025)
It's just another messaging. I used them to automate messages about current disk usage and warn if the machine was actively being used.

But then again, I had people who would run jobs without checking if the machine is already in use. Obviously these people didn't check email or slack either...

godelski··on VSCode's SSH Agent Is Bananas (2025)

  >>> Assuming your users use the integrated terminal regularly
People don't use the integrated terminal regularly
godelski··on VSCode's SSH Agent Is Bananas (2025)
And then you put it in MotD because everyone ignores emails and slack
godelski··on VSCode's SSH Agent Is Bananas (2025)
Reality is you send multiple emails, slack messages, and MotD and people will still act like there was no warning
godelski··on VSCode's SSH Agent Is Bananas (2025)
Are all those processes 1 UID?
godelski··on VSCode's SSH Agent Is Bananas (2025)

  > but that's probably not a universal stance.
Sometimes you aren't really "the owner". For one example I was admining my group's server in grad school. I wanted to add quotas (we already had zfs) because people were abusing home directories but my advisor and a few members were very against it because I was "over complicating things". Their worry about me wasting time (20 minutes for all machines...) resulted in hours of yelling at people over the years. All because, surprise, a small number of people can't follow instructions and abuse systems, ruining it for everyone.

Another frequent problem we had was people using the systems while others were. They wouldn't check the machine's status. And of course you can guess that I wasn't allowed to add a scheduler.

A lot of groups do things in janky ways. Often not because they don't know any better but because leadership doesn't and is assertive

godelski··on VSCode's SSH Agent Is Bananas (2025)

  > they are too low for modern development...
Do the math, 256 processes 50 MB each. How much RAM is that?

Too low? 256 reads as *pretty* generous to me.

godelski··on VSCode's SSH Agent Is Bananas (2025)

  > The max limit on 64 bit systems is what, 4,194,303?
What a weird framing... I'm not sure what you're even trying to argue. I mean a single process can overload the machine. Just because you can label 4m processes doesn't mean you can actually run that many programs. Just think about that for a minute. 256 processes is pretty generous
godelski··on VSCode's SSH Agent Is Bananas (2025)

  > what do people put there?

  \033[1;31mCLEAN YOUR FUCKING DRIVE OR I'LL DO IT FOR YOU!\033[0m
You know, typical admin stuff
godelski··on VSCode's SSH Agent Is Bananas (2025)
Tried that. But I suspect you know why it didn't work
godelski··on Meta VR Glasses
Why? Does his model not get firmware updates anymore?
godelski··on VSCode's SSH Agent Is Bananas (2025)
Exactly! Like if I'm admining a server what am I supposed to do? Message on a big slack channel and have everyone ignore me? It's easy when people are just logging in through normal ssh as I can put a big bright warning message on their screen that they can't ignore.

Also, timeouts...

Also, does anyone know if VSCode supports mosh?

godelski··on VSCode's SSH Agent Is Bananas (2025)
As a Linux user I've hated VSCode's ssh. There's lot of annoying things that make it harder to admin for. Like it doesn't pick up the MotD, preventing me from showing users important messages. I've found that it also doesn't reuse sessions (at least by default. TBF, neither does ssh) and I'll find that there's just dozens of open sessions over months from users. I literally had to write a script to boot people...

It would be one thing if the plugin was just a wrapper and people were still expected to know ssh but the plugin abstracts away all that and is intended to make it a "use VSCode on remote machine" tool. So it needs to do more than just handle creds, otherwise it creates a divergent experience while making people think it's just ssh

godelski··on Apple has added persistent 'ads' to iOS, and it's driving users crazy
I don't want to really downplay Apple but there's much better innovation examples that what you listed. One of Apple's great advantages is that it owns the hardware, unlike Linux or Windows. Those have to work in highly heterogeneous environments. I mean Linux runs on everything from a literal potato to a super computer and everything in between.

To highlight Apple's innovation maybe mention things like the iPhone or iPod. They weren't the first devices of their kind but Apple's magic was always the mixture of the two Steves. Technical but stylish. But modern Apple isn't the same. The magic is gone

godelski··on 'We hacked the FBI:' Hackers say they have data on all FBI employees
I don't think it's that no one can do it but rather we always use the cheapest option and then get surprised that we get what we pay for.

I'm not just talking about bids. The idea presets itself prolifically. We all work in tech, I'm sure you see it on a daily basis. Rush Rush, no time to think, just do it. 6mo later, "there's so much shit! How could we have ever prevented this?" Rinse, lather, repeat

godelski··on The senior engineer death spiral
I assure you I'm human. You can't really on a single flag or you'll just have tons of false positives and piss people off.
← PreviousPage 3 of 34Next →