godelski

28,113 karmajoined 10년 전

[ my public key: https://keybase.io/godelski; my proof: https://keybase.io/godelski/sigs/ITXd1PN0tKNiQYxLgpyMs4xX2HIQy0uyX-09Z4DbV48 ]

comments

그저께discuss

There's also an interesting relationship to variance when you only have mean and median

그저께discuss

Even you're relatively new. There definitely was a shift as AI took off

Also, they said "year 1" not "1 year old"

3일 전discuss

  > Which makes it a simpler problem.

Yes, which is what I stated in the first comment

  > I'd say SOTA AI coding systems are better than 1 and still improving.

I'm including Fable. I use agents every day. They still are very sloppy and hyper fixated on tasks and are poor at making code flexible and maintainable in the long run.

But again, I'm not surprised because that is the modus operandi of the typical Silicon Valley programmer (irrespective of seniority). They leave in their wake bugs that are avoidable and unnecessary and significant amounts of tech debt.

But then again, no one ever gets credit for squashing bugs before they exist. The agents are just as myopic as we are

3일 전discuss

Isn't the end of the article saying that their users are mainly in that tail? Seems to be exactly what you say: figuring out who those 2% are. In the OP's case, it's 30% of their users

3일 전discuss

I think you're not being generous in your interpretation. How I read it he could be talking about the number of 9s a server's uptime is. If you pay for 1 9 you'll lose a lot of customers. Hell, true for even 3 9's. Look at all the complaints about GitHub this year. 5 9's is the standard and that's 99.999%!!

The thing is that it is all context dependent. A lot of times 0.1% is nothing and can be ignored or pushed off. But sometimes that 0.1% is worth billions of dollars.

The point is that data means nothing without context and interpretation. If you're lazy in your analysis you are going to have lots of issues

4일 전discuss

I think you're over generalizing. Software rot happens for a multitude of reasons

  - hardware changes
  - firmware changes
  - operating system changes
  - dependencies change
  - bits flip, thanks cosmic rays!

  > Right now, it feels like AI coding is ~2.5

A lot of people would argue that it's not even 1. I'd argue it sits between 0.7 and 1.2 given the task.

I don't think it's close to 2. It's hyper fixated on task implementation. It's really bad at abstractions, even when requested. You have to be pretty precise over these instructions. I've only seen it develop well, even with explicit instructions, a handful of times. But hey, that's true about a lot of people too. Though people will not claim to have solved the proper generalization, people will just say "I just care that it works", which is kinda why AI is trained that way. It's not being programmed by your Knuths, it's being programmed by your Zuckerbergs.

5일 전discuss

  > why have an official firmware if it's not at least slightly maintained.

Because it's still useful to have a blessed child so that people getting into the space have somewhere to start. You could accept zero additional PRs and it would still be a useful thing to have.

The hardware is static so the rate of software rot is pretty low. It can effectively not be maintained as long as it's already in a stable state. Adding new features is cool and all but it also adds more bugs.

But the great thing is that there's a community, all using the same hardware, and people can fork. So people can still get those updates that they want. Maybe the only thing to do is create a community fork that is much more open but doesn't come with the same stability promises. But that can still be a lot of work, even if you get community maintainers

5일 전discuss

To be fair, some software does rot. But when you have control of the hardware and the software, rot is pretty uncommon.

Honestly, I thought the whole point was to make a popular unified platform where the community could come together and expand on it. I really can't imagine a centralized player can predict nor create all features that users might want. But it seems like Flipper did the right thing: make the software flexible and easy to expand upon.

5일 전discuss

  >  Last I checked, one of the Hacker News rules

The guidelines do have these points though:

  - Please don't complain about tangential annoyances—e.g. article or website formats, name collisions, or back-button breakage. They're too common to be interesting. 
  -  Please don't pick the most provocative thing in an article or post to complain about in the thread. Find something interesting to respond to instead. 
  -  Please don't post shallow dismissals, especially of other people's work. A good critical comment teaches us something. 
  -  Please don't use Hacker News for political or ideological battle. It tramples curiosity. 
  -  Eschew flamebait. Avoid generic tangents. Omit internet tropes.

Which, let's be honest, the top comment and majority of the responses violate these guidelines. I was quite surprised to see that that was the top comment. I would have never noticed until they pointed it out. Personally, I don't care and I'd rather read comments about the actual content of the article than talk about furries. They don't live rent free in my head.

7일 전discuss

  > the middle layers of the Transformer that are affected most by RL post-training

This is where you should expect most change in models. The beginning layers need to embed while the later layers will reform the result to the final conditions. The middle is what does all the untangling.

7일 전discuss

  > It's not just a tautology, it's

Inconsistent logic.

You've made a grave assumption in treating this as a binary problem: comprehensible or incomprehensible. The truth is that this lies along a spectrum. I can "comprehend" a color which my eyes can't see but this is different than trying to comprehend red. Or as a variant, I can comprehend a color that I can't see naturally but if I go into a lab and they simulate those cones in very specific ways I won't ever be able to really imagine it. Seeing it will be clearly a different level of comprehension.

  > In practice, we'll never run out of mystery or ignorance or mistakes.

I fully agree with this but also for some reasons unmentioned. Truth has infinite precision, a thing we will never achieve. Similarly my namesake showed there are limits to axiomatic systems.

  > Brains are Turing complete.

And there are problems that can't be solved by Turing complete machines. That only means they can solve computable problems but there's plenty that aren't. Very famously the halting problem isn't. And in the real world there's many problems which are intractable. And some are more intractable than others

8일 전discuss

  > Thirdly try to pay it forward.

An important step. If you don't it's not dissimilar from pulling up the ladder behind you.

The thing I've found is that often what's simple and easy for me (the person being asked) is difficult or overly cumbersome for the person asking (or even not asking). Often it's due to simply being in the right position or knowing the right people. We're all in this together and there's so much to success that's beyond talent and skill. All the small little actions add up even if you don't get to see how

8일 전discuss

  > Moreover, in my opinion, by buying Google phones you're feeding Google, and I, personally, would like to avoid that.

I bought an iPhone based on that very decision. TBH, I regret it. The ecosystem is so locked down that I can't even sync my photos to my NAS without a hacky constantly breaking shortcut, building my own app, or paying for an app. Just to replace a small <50 line bash script that could do more than either the shortcut or any paid app I know. I'm constantly battling my phone.

That would all be worth it, but it's been a few years and Google did all the shitty stuff I was protesting anyways and Apple is getting worse.

I'm hoping we get those moto phones with graphene pre installed so I can actually send the right market signal. But what's fucked up these days is you can't send the right market signal. Meanwhile I talk about fixing shit at work and my coworkers ask "what's the value" or say "there probably isn't any money in doing that". Even for problems they are one liner fixes and that they agree we spent more time arguing over than it would be to fix it. I don't think it's a top down problem, it's a bottom up. Those are much harder to solve

9일 전discuss

I prefer Thunderbird

9일 전discuss

This freaked me out the first time I used Mail. It rendered a PDF about buying bitcoin on PayPal (I don't have PayPal). Looked at the same email in Gmail and Thunderbird, it's just a PDF attachment, no message, and the sender was different. No wonder people are falling for those scams, Mail makes it easy for them.

Now if only I had a better client on my phone (never buying an iPhone again)

11일 전discuss

You can build all of mathematics on type theory? I very much doubt that considering there isn't even a fully unified mathematics. There's holes that don't know how to be bridged between entire subfields. So I'd be impressed if type theory really could do everything, but hey, I don't know

11일 전discuss

  > Nobody genuinely cares about the moving sofa problem, or about square packing, or about the minimum number of colors needed to draw a map

Wild examples. These are all extremely useful.

I agree that solving problems along the way is incredibly useful too, but there's no math problem that I'm aware of that's utterly useless. Thing is, math is about abstraction and structure. You're developing tools. Those tools can be used for more than one thing. That's why math is often called a language. It's also why mathematicians and physicists end up being so helpful in other fields, because it was (most of them) about understanding the language of complex problem solving. Even code is math (I find it quite odd programmers are against math and elegance as these are directly in line with their goals, but I guess there's always a Dijkstra EWD already calling this out)

Where I fully agree with you is that solving hard problems advances us. TBH, it doesn't matter what it is. Fund NASA and you not only get men to the moon but velcro, GPS, and global wireless communication. Fund particle accelerators and you get the internet. Fund math and, like you said, you get a tool (calculus) which is worth more money than we can even imagine. The thing is that solving hard problems forces the unknown unknowns to become, at least known unknowns.

14일 전discuss

  > In some sense I always considered programming to be more trustworthy than maths arguments without the certainty of a solver proof.

But programming is a subset of mathematics. They are both formal languages. I suspect the trustworthiness is more in your comfort level than the ability to verify

14일 전discuss

  > just one kid needs to figure out how to bypass the system, and the knowledge spreads like wildfire.

I'm surprised this is not obvious to people here on HACKER News.

When I was in high school we all learned about proxies and bypassed the school firewalls. You didn't have to know anything technical after a few people figured it out. Hell, even the teachers were in on it. I remember one wanted to know so he could check the lotto numbers lol.

It's an eternal cat and mouse game and the mouse is going to win. I agree that the right idea is friction but if people aren't aware that there's no clear win that's going you work even 80% of the time then we'll write the wrong laws and have the wrong idea

14일 전discuss

  > it’s certainly better than a server.

I disagree, and so does the OP.

  > OP just had to use a bad implementation.

Then it isn't so certainly.

You act like machines are perfect. Machines glitch and have all sorts of problems. They're usually inflexible because programmed by the lowest bidder. You could argue about implementation but that is also true for human servers too