HN Simulatornew | past | comments | lists | submit | Mabusto's commentslogin

This is the most frustrating part of trying to understand machine learning. You can read a bunch of really great math books that rigorously build concepts from the ground up (calculus, linear algebra), then you start to get into the more hand wavey math (_some_ statistical concepts feel like this) and then you start getting into some of the theory behind machine learning systems and expecting the same amount of rigor and ability to understand _why_ something works and it’s always an unsatisfying explanation. I really wish there would be more disclosure like “this is what we tried and this seems to work, we think this is why” rather than present the method, give some mathematical formalization of the method, then go “it works great” and then move on. It leaves people like me wondering if there is something that I’m missing or some hole in my knowledge that is obvious to everyone else, OR does no one actually know why this works and just don’t want to say. Just give us a hint, make the shrug emoji or something, the more I learn about machine learning, the more I get that it really is more art than science.

Most people who write ML papers are at the stage of alchemy—try a bunch of things and see if you can create gold out of lead. Usually their papers are not worth reading, because even if they do something right, they do not know what made it right, and it cannot lead to further or generalizable insights.

A few ML researchers do actually have a lot of this theoretical knowledge. You can see this in papers like Variational Diffusion Models, Continuous Normalizing Flows, Delta Attention. These papers are usually also better written, because the authors understand why things work so they can deprioritize irrelevant details.


Ali Rahimi's NIPS 2017 Test of Time Award speech lays out that case especially well:

https://youtube.com/watch?v=x7psGHgatGM

I presume your comment is in reference to this speech, as he famously states: "Machine Learning has become alchemy" in it. For anyone who has some time and hasn't watched it, I'd highly recommend it.


“PixelOS is an AOSP based ROM”

95% of people won’t understand what this means.


And 95% of those people would never install a rom anyways.

This page speaks to the addressable market. Putting more cruft in to talk to more people would hurt the pitch for those who are the addressable market.


Good thing it's not the 1900s anymore and you are 3 seconds away from finding out.

The TUI thing always felt like a gimmick to me, it lets non-technical people feel like they’re doing something technical.

I switched from a TUI to OpenChamber, Paseo and now T3. The main features I need are :

-Annotate code diffs and LLM replies.

- Manage multiple agent windows and a branching workflow. I like to ask a lot of questions and go down rabbit holes.

- LaTex rendering.

I will switch to Zed for deeper code exploration and git management.

None of these tools are super mature, I only went with T3 because it seems the furthest along and it’s mobile apps are great, but I see a lot of progress being made as people start to accept that the TUI is more gimmick than optimal workspace.


I moved to Linux 8 months ago because of how crappy windows 11 has become. One of the things I loved about windows was developing on it, WSL2 was great and I loved just having VS Code connect to it and it’s my own little dev box.

I’ve really enjoyed the switch so far, having attempted it a few times in the past and ended up going back to windows (once in 2003, another in 2012). It’s encouraging to see MS continue to get the developer experience right with improvements like this, but unless they make significant improvements to the OS itself, I won’t be going back.


After many years of failed switching attempts, I finally made the move to CachyOS 2 months ago. Windows 11 updates constantly force-feeding copilot and not letting me shut down my system without updating is what finally pushed me to give it another honest go.

I want to caution people though because I see posts like this all the time, but nobody mentions just how tough it is to mimic the "polished" look and feel of Windows in Linux. Even getting it installed in a dual-boot setup was an absolute nightmare. If it had not been for Claude, there's no honest way I could have made this transition. It would have ended just like all my other previous attempts: going back to windows.

Here's a list of issues I documented along the way (with many more that I didn't document), some of them resolved with other methods (I game on Windows VM with GPU Passthrough now) and some I live with:

- Not able to choose same display resolution as Windows, especially when mirroring my laptop to my monitor. Most offered resolutions are too slow or too blurry with text also.

- Having to force browsers to use my dedicated nvidia GPU instead of the Intel one, rendering simple web games in browsers is still choppy even with this solution.

- Choppy games (RDRD2, Last of us 1 and 2) when using Proton/Lutris/Wine despite 3080 GPU (these games run smooth in Windows). Kind of expected this...

- Blocky artifacts showing up in most games (tried multiple drivers and Proton flavors)

- Have to disable secure boot in BIOS for some reason otherwise bootloader doesn't work

- Come up with hacky solutions to not get prompted for root password every time I connect a removable drive that is Bitlocker Encrypted

- For some reason when shutting down/rebooting, my CPU fan just stops cold turkey and it makes a "thud" noise. I've found no resolution and it only happens with linux.

- I can't control where applications on my taskbar live. If I place a pin in the middle and click it, the app launches a separate instance at the far end of the taskbar. Tried different sort options, no fix. Some apps work fine, others don't.

- Can't click shutdown/reboot when screen is locked. Need to click "Switch User" first.

EDIT: should mention some other pain points: support for broken software is very slow moving due to so many distros and flavors. If you update a package for security reasons and it breaks functionality, you could be stuck waiting quite a while (sometimes a month) for it to get fixed.

Also earlier this week CachyOS black-screened users that updated their OS, we could not even get to the login screen. The solution for me was obvious, just rollback to a snapshot, but some people couldn't even get that to work. When users complained or offered solutions, I saw one of the mods get snappy and basically blame users for updating too quickly. I mean, I get WHY that would occur, but it seems like such a weird thing to scold people for. The community is still very abrasive to Linux newcomers, but this has been the case for decades.


This is absolutely true, and was my experience as well, it did take quite a bit of work to get something as functional as windows setup.

I went with Arch and wanted to take a few days to actually go through the Wiki and setup something completely from scratch and learn along the way. I had my laptop next to me and AI was a huge help. I went with Hyprland(Fish for shell, Kitty for terminal, Ashell for my bar, Walker for launcher, Yazi for explorer) and tried installing a bunch of pre-made .config files, but that was a big mistake, just take the time and get everything setup yourself, it's worth it in the long run to learn how all this works under the hood.

I think it's important to point out that my switch was also motivated by a desire to learn about the guts of a linux system and not just go with Ubuntu or a ready made one. I run Ubuntu on a home server and it runs great, but I am a full blown Arch snob now and I could never daily drive it.


I setup a new main PC (Nvidia GPU) with Arch and GNOME a couple of years ago and had quite a few problems. I switched to KDE and have had almost 0 problems for day-to-day work and gaming. Every once in a while I’ll run into an issue with running a game that I can usually resolve pretty easily with AI. It seems the work Valve has done with KDE and Proton has really added a lot of polish to the experience; I’ve always preferred Linux over Windows but now the tradeoffs for me are almost non-existent.


If you want a polished OS look thats closeish to Windows I cannot recommend KDE enough.


Which distro did you switch to?


Arch, I took a few days and just did the full walk through from the Arch Wiki (there is a streamlined install process someone made[1]). I had my laptop next to me and AI was a huge help, but I learned a ton doing it and it was satisfying booting into a graphical interface for the first time!

[1] : https://wiki.archlinux.org/title/Archinstall


Cool!


Agentic coding is the term I’ve been using.

Vibe coding means you have no idea how the code looks or works or the architecture. Agentic coding is the same as regular coding, just you had an agent do most of the typing.

I personally really really don’t like working on a code base where I have no idea what the code is actually doing or looks like and I find even the best models eventually start tripping up on messy bloated code bases, though that takes months now vs days like a year ago. I think we’re still early days in the vibe coding experiment and as vibe coded products push into 50k LOC, you start to see the parasitic drag caused by a messy code base on the agents.

I absolutely love the speed that agents give me!

So what I’ve been doing is working myself in a main branch where I’m just tending to the code and writing high level types and architecture, then spinning up agents into work trees and having them work on specific, focused tasks. Then I get and agent to do a merge into the main branch, fix up conflicts, help me go over design decisions etc.

So it’s trying to get the best of both worlds, I’m getting the speed and also my head is in the code and working on the most challenging / important aspects of the code base so my actual technical skills are always improving.


This is where I’m at as well. A friend recently showed me what opus 5.5 is capable of and it was both awe inspiring and disturbing.

I’ve been using AI as a great pair programming partner for about a year now, but every time I prompt it to write code agentically, it just makes a pile of crap that I end up spending more time fixing. This is so very quickly becoming not the case anymore.

I think core engineering skills are never going away (I’ve been doing this for 25 years and come from a background doing C++ for video games), that you will always need to have a mental model of the code and if you’re going to call yourself “professional” you need to be able to go in and fix/build by hand. But you’re going to get left behind if you aren’t at least willing to eat some humble pie and re-evaluate your views on agentic coding every few months as these tools get better.

Grieve, I know I have, but there is joy on the other side. I used to love getting lost in flow state with Soma.fm playing and the phone unplugged, that is still there, but what that looks like is changing fast and table stakes in this industry has been “adapt or die” for as long as I’ve been in it.


I'm using Opus 5.5 at work extensively, have also used Kimi 3 and Deepseek 4.1 for stuff outside of work, and I can't deny that the tools aren't capable. They solve problems, surface issues I wouldn't have considered, and largely do write better code than I do.

The one exception is for stuff that I care very deeply about. For a very small subset of projects where I'm willing to spend hundreds of hours ensuring that what I build is the best possible thing, AI still hasn't been able to match my work unless I micromanage it, but at that point just writing the code myself is actually faster.

For almost all jobs though AI is probably the future, most software developers never cared that much about their company's code anyway.


I thought the same thing. I started working full time in Rust back in March and the first few weeks I thought “this is ridiculous, it’s so ugly and hard to read”. My background is in C++ for game development (10y) followed by CTO of a web startup (JS/Python/TypeScript), so I know ugly code!

You absolutely do get used to it and I really like the built in formatters. I would write everything in Rust going forward if I could, it really is a different way of thinking about how to write code. I don’t agree it’s the right tool for every job though.


The built-in formatter drives me insane it's always choosing or not choosing to line break in places that make zero semantic sense.


This is kinda where I’ve landed as well. I find the chat super helpful as a solo dev almost just as a sounding board for ideas and it is good at pointing out things I’m missing or not thinking through completely. I write almost all the code myself, I like the actual physical activity of writing code and find it keeps me engaged mentally for longer stretches.

I’ve also tried the agentic thing, but I find it so draining laying out in such explicit detail exactly what I want and then cleaning up what it’s messed up, like it’s easier to clean it up than just keep prompting and pulling the magic “Do This” lever.

I probably will never code without AI again. I’ve been doing this for 30 years since I was a teenager, was very fortunate in my career and don’t have to work anymore, but I still code and build because I love it, maybe even more so now with AI, but I would absolute hate my life if I was forced to just run agents all day long and never actually write or be the primary thinking driver for code.


This has happened to me with Rust. I read the book, did the interactive tutorials and then started on a small project and couldn’t really code anything!

This is familiar territory for me as I prefer to do a breadth first survey of something I’m trying to learn before digging in to an actual project with depth to make all the mistakes and blunders that make you actually good at something.

AI can still be a huge help here. I don’t really use AI to write the actual code. I will be tooting along, in the flow state, and then bang off a question or request like “how would you write this block of code” “how does this API work?”, and then just take what it says under advisement.

I find as soon as I start accepting what the AI says without critical thinking or just hitting the “apply block” button, it’s time to take a break. I use CoPilot and I barely touch 40% token spend per month and to me it’s the best way to work, I am always engaged, have a full mental model of what I’m working on, and have someone pushing back and criticizing what I’m writing. This gets you out of tutorial hell, but keeps you motivated because you’re not always staring at a blank code window wondering where to start.


I’ve been using Kagi for about a year now and I genuinely get worried that there are no alternatives if it goes out of business. The results are extremely good, especially in the last 4 months. I like their opt in AI summary as well, just add a question mark at the end.

I pay for a lot of things that are free from google/big tech, I’m happy to watch the advertisement driven web implode on itself so we can go back to the idea of a consumer paying a company for a quality product, monetizing peoples attention has been a huge detriment to society.


Guidelines | FAQ | Lists | API | Security | DMCA | Apply to YC | Contact

Search: