pSYoniK

As another commenter pointed out, I feel that the biggest change that occurred in the last 6 or so months is that these tools are removing the human "hurdle" in order to complete a task at all costs. The best way to complete the task is to no longer ask for input, clarify unclear things or use existing solutions but code the whole thing yourself from start to finish.

I have had the misfortune of working with such people who are now encapsulated in Opus 5/Fable/Astra which means that you WILL get a solution, but it won't generally be maintainable or useful. Multiple times have I found myself stopping Fable or Opus or even Sol from building their own JSON validator in Python or god knows what else, because at the end of the day, the reward is to complete the task.

It's also one of the reasons why I'm finding older models more useful for the type of work I actually do and why I've been favoring something like Deepseek Flash. Just started using Flash 4.1, so not sure if it exhibits the same maniacal approach to tasks as the Western counterparts. (I only briefly tried GLM 5.2/5.3 and for nothing major, so I couldn't comment on those).

For context, 80% of my professional work relies on adding functionality to an existing code-base that is very difficult to work with, has a ton of business logic scattered across and was built in a go-go-go fashion many years ago. Since then people kept pilling "features" on top with no testing strategy in mind apart from the business manually testing it. Letting something like an LLM loose on the code-base would introduce soooo much risk that it's just untenable so the only way to work is to really isolate changes and then try to build out small reusable components. Even so I find Opus go off on a tangent "Hey, let's not bring in Markdig, I'll build my own Markdown rendering engine, give me 7 hours...".

I have written on the subject of LLMs previously on my personal page, I find them completely unnecessary and a trove of theft and value extraction through theft, but I understand that they can provide benefits when used judiciously. However, despite all the hype in the last few months, these latest models feel and behave off.

If I hold the answers to a test, you might score more in a test if you break my arms to get the answers out of me, but that doesn't make you smarter.

I thought this was going in a different direction along some of the thoughts I had around LLMs for programming tasks specifically. While this talks about understanding and how to ensure you're keeping up with what is changing, I feel that this is maybe more aligned with how a PM/PO should understand the work being done and not necessarily how an engineer should.

I took a note a few months ago and my point was I think more engineering specific, although it might be my own lacking abilities/skills that caused this realization. "Your capacity to learn/recall and map information is the new bottleneck. LLMs can act as learning amplifiers but correctness isn't as important for their output as critical thinking on the side of the consumer - YOU."

My point is that, I think if an LLM outputs 50,000 lines of code, your ability to go through what has changed, how it has changed and where the changes have occurred is the bottleneck. I see the approaches here, sure, "summarize the changes" or "draw me a picture" or the more recently observed "build me a city building simulator to understand this", but I feel that misses the point from an engineering perspective. The difference in understanding the weeds such as DB transactional boundaries or tenant isolation (which I believe was a topic in a recent data leak), those aren't summarized that easily in drawings or if they are, if you are working at this granularity, then your 50,000 line PR will yield 50,000 pages of crayon drawings you now have to understand.

I guess, my point is that understanding is the bottleneck, but low level understanding and the ability to read/map/connect is even more so. Any developer with some experience will agree that if changes are trivial you can scan and pick up mistakes or flaws easily. So most SOTA models won't necessarily even make these. So what you're reviewing now is going to be one level higher or more in terms of difficulty, mapping multiple components or touching multiple surfaces. Your ability to make the links, reason about them and attempt to find flaws or logic issues is the bottleneck. In the time it takes you to understand, another 50,000 line PR is up.

I'm not sure how we're going to be solving this. I don't know if in the current state it is a solvable issue, maybe another 6 months? Maybe another 6 years? Maybe this is fine and we will settle in a sort of place where your mediocre engineer will be responsible for tens of reviews a day signing off on method/functions/classes/interfaces being added, get paid 50k a year and doing the same non-thinking work day in day out while signing their name to the quality of the code being shipped while a senior/lead will be busy reviewing multiple of these. Think of the way an assembly line functions.

P.S. I hate to see this annoying tendency of transforming knowledge work into assembly line work. We keep trying to "fix" this without understanding what knowledge itself is. Maybe this technology will indeed yield software assembly lines, I don't wish to eat my words, but I'm still struggling to see how we will handle the nitty gritty of software work. Maybe the same way we handle building airplanes - as long as only a couple crash a year, we're sort of fine.

I have been working on a static site generator. I've used Hugo and Jekyll before but everytime there is something you have to learn or get used to and they never map 1 to 1 to the flat structure of a simple blog for example.

It's a few classes of C# right now that parses a folder and auto generates the navigation, links, titles and summary off of the folder/file structure for the 1st level and then subsequent depth is mapped to individual items in the list view not on the nav bar (although I did that too, it looked weird with higher depth). So if you have a folder with 3 files and 2 additional folder you get 5 links in the nav bar, 3 content pages and 2 list pages. No other setup is needed. No llm written code, but I used deepseek v4 flash to ask various questions and get samples of some things.

I forgot just how much depth there is to exploring a simple problem. Loving it!

It is however very patronising to tell people to "Go full vegan, abandon all possessions, and all that".

It also isn't useful to reduce the conversation and assume that critique directed at the idea of necesarily going out and buying new hardware is a critique against technology or ownership, but, myself included, we do seem to read what we want. You also missed the point I made when I did clearly say voting with your wallet doesn't work. You didnt address the other, more salient point I was getting across, but obviously failed to do so - when starting out, don't worry too much, just get whatever and start learning. Questions will be easier answered when you already have some hardware.

Anyway, enjoy your day

TL;DR - please stop wasting tons of resources putting together new servers every year and turning this into yet another outlet for "I have more money than sense and hopefully I can buy myself into happiness". Just get old random hardware and play around with it and you'll learn so much that you will be able to truly appreciate the difference between consumer and enterprise hardware.

This seems awfully wasteful. One of the main reasons for which I've built my own homeserver was to reduce resource usage - one could probably argue that the carbon footprint of keeping your photos in the cloud and running services is lower than building your own little datacentre copy locally and where would we be if everyone builds their own server, then what? Well, I think that paying Google/Apple/Oracle/etc whoever money so that they continue their activities has a bigger carbon footprint than me picking up old used parts and running them on a solar/wind only electricity plan. I also think I'm going a bit overboard with this and I'm not suggesting to vote with your wallet because that doesn't work. If you want real change this needs to come from the government. You not buying a motherboard won't stop a corporation from making another 10 million.

Anyway, except for the hard drives, all components were picked up used. I like to joke it's my little Frankenstein's monster, pieced together from discarded parts no one wanted or had any use for. I've also gone down the rabbit hole to build the "perfect" machine, but I guess I was thinking too highly of myself and the actual use case. The reason I'm posting this is to help someone who might not build a new machine because they don't have ECC and without ECC ZFS is useless and you need Enterprise drives and you want 128 GB of RAM in the machine and you could also pick up used enterprise hardware and you could etc...

If you wish to play around with this, the best way is to just get into it. The same way Google started with consumer level hardware so can you. Pick up a used motherboard, pick up some used ram, a used CPU, throw them into a case and let it rip. Initially you'll learn so much and that alone is worth every penny. When I built my first machine, I wasn't finding any decently used former desktop form hp/lenovo/dell so I found a used i5 8500t for about 20$, 8 gb of ram for about 5$, a used motherboard for 40$, case was 20$ and PSU was $30. All in all the system was 115$ and for storage I used an old 2.5inch ssd for boot drive and 2 new NAS hard drives (which I still have btw!). This was amazing. Not having ECC, not having a server motherboard/system, not worrying about all that stuff allowed me to get started. The entry bar is even lower now, so just get started, don't worry. People talk about flipped bits as if it happens all day every day. If you are THAT worried, then yeah, look for a used server barebone or even a used server with support for ecc and do use ZFS, but I want to ask, how comfortable are you making the switch 100% now over night without having ever spent any time configuring even the most basic server that NEEDS to run for days/weeks/months? Old/used hardware can bridge this gap and when you're ready it's not like you have to throw out the baby with the bathwater. You now have another node in a proxmox cluster. Congrats! The old machine can run LXCs, VMs, it could be a firewall it could do anything and when it fails, no biggie.

Current setup for those interested:

i7 9700t

64 GB DDR4 (2x32)

8, 10, 12, 12, 14 TB HDDs (snapraid setup and 14 TB HDD is holding parity info)

X550 T2 10Gbps network card

Fractal Design Node 804

Seasonic Gold 550watts

LSI 9305 16i

I had wireguard on docker before for some containers, but it felt clunky and it over complicated the network stack in my head (I'm unfortunately not very skilled in networking in general). So I said that I'd go back to the root and run it at OS level because then I can expose Proxmox to the world or any of the other VMs I run by having them join the wireguard network. Which in turn means that I can connect to any machine I want/need directly. I am also playing around with writing my own dynamic DNS worker in C# and I was curious on how I could have that run as a systemd process but bypass the wireguard tunnel to keep updating IP addresses. A lot of these were tied to me just being a bit more curious about the whole stack.

I mentioned that as an alternative along with Headscale and Nebula. Not for me though! At least not now.

I put Zorin OS on my dads old laptop 5 years ago and I think the only time I got a question was when someone setting up his new internet was digging through network settings but hadnt used any Linux distro before. Even then it was a 5 min call. Its a very Windows-like experience and I've noticed most parents really just write an email, browse the web and maybe consume media. All of those can occur in a browser.

I have looked at that briefly, I think I had gone with pihole in the end for the ability of having a UI to easily see any resolution issues and local dns management (which, I think, is also present in Unbound but not in a UI but via configs).

This is such a good point. I've changed jobs a lot and one thing consistently bad (to varying degrees) is documentation and tutorials specifically.

"Download X and setup"

"It's not working..."

"Oh yeah, you're supposed to do it on the remote access VM"

"It says access denied"

"Oh right, you're supposed to use the Yubikey for access"

"I don't have a Yubikey, its pass + authenticator"

"Ok, I'll email Jeff from this department you wont hear off until someone new starts. But otherwise keep following the tutorial and you should be good to go!"

It always infuriates me. At my last job I had a lot more control and authority, so I redid the entire tutorial for the proj we worked on. Every few months I'd check all my account permissions, update the list on the readme, spin up Windows/Ubuntu VMs and try to get the project running using ONLY the tutorial. Anything missing - add it.

If anyone added a new dependency the documentation would be updated and the steps checked on a new VM. I did this as we had various people come in and work for a few weeks, add a new feature and leave. The end result was that instead of 1-2 weeks to get running, people would have everything running within their first day and start work sooner. Instead of needing someone for 4 weeks for ONE feature, we could finish 2-3 and sprinkle in more tests and confidence.

I think most developers would benefit from writing for a less experienced audience, especially for this sort of thing.