nurettin

To be fair, models are more accurate than how they were earlier this year. In march I was still frustrated and was writing my own event loops because the model simply wasn't built for it. By January I was still hand checking statistical outputs to make sure it wasn't making any errors, and it was. After august it was pretty much over. There were no stats errors, code generated was shit but it could do what is asked. Instruction following is still a problem, and I sometimes catch silly things in the scrolling diffs, but it has been much easier since then.

I just tell it to work until described result. It works unless it gets stuck at a necessary step I had explicitly forbade earlier, or an unreachable port.

What's your opinion?

I expected it to be the other way around. This is the best time to self-host and decentralize now that the computer can do whatever you want and warn you about security best practices.

Honorable mention: thanks to red letter media for keeping the memory alive. And for being dorky and entertaining.

I rewatched tng 4 times and I was skeptical of strange new worlds but I just watched s2e3 and mind = blown. I couldn't believe that was an episode from a series. Recommended.

What have they done to earn 100% of your trust?

Once I started parsing example messages with the above method, the format became apparent. I even made a viewer for it which was useful because people we were integrating with were also making mistakes and I was the one correcting them thanks to the viewer.

The hardest part was europeans taking 15 day holidays every couple of months leaving work to someone else who had to start over.

Once you know the tree, edi is trivial to parse. You can even do it with a loop and a state to store which level has which tag and what the current count is.

The real challenge was to discover what the tree was. Like wayfair order listings, cargo delivery messages, which are also similar to how Rhenus does them. Every vendor has their own format and standards.

This is not an aside at all. It is on point. It seems like whatever statistical trick they used also inflated gemini 3.8 to absurd scores as well.

delphi 1 here. Didn't even know perl existed.

TOTP has always felt incomplete to me. The key is right there unencrypted in plain sight.

Don't forget the "favorable" game reviews.

So am I

No, crazy would be owning the router and cables/satellites as well as producing your own power and having a manufacturing plant for hardware capable of running the software.

Germany is pretty famous for tagebaurekultivierung.

I didn't know it was so good. That was a gut punch. But I'm pretty sure the market already priced this in. And people are rightfully concerned about the owner of the data. Anything that is concerned with social, political or financial data goes out of your country to another one and is maybe even kept as a potential weapon.

After seeing the cancerous monster AI "proofs" it became obvious as day that we need better abstractions in several fields. The good part is: there is still a lot left to do for humans as they actually have the abstract thinking ability, as opposed to crazy cancerous token generation.

In my experience, LLMs instantly find the reason for the crash and fix. They don't even need to go through the debugging phase anymore. It doesn't matter if you're generating c++ lisp or cobol.

It was the best part of turbo pascal as well. Before IDEs decided to hide information.

With these numbers, I'm holding my breath for the pelicanbench.

LXC is a jail and qemu is an extra layer of emulation. Why pay that extra cost? Just because it serves a wsl image isn't really an answer. How is that worth the extra layer?

We still have developers because we can't outsource responsibility to AI. If shit breaks, we still need someone to fix it. Maybe you think that can also be automated. I have a counter-example. AI will rarely use the existing state when solving a bug. It will tack on it's own event queue, it will make up it's own messaging system, it is a cancerous growth. And this observation is based on both opus5.5 and gpt5.6-astra. The results are way past magical at this point, but you still have to weigh in pretty heavily for orderly code growth where solutions are succint, focused and preserve the existing structure. They are still noobs at fixing/adding stuff.

Same here

That very similar history is shared by us dinosaurs. I especially remember tuning the 1530 datasette to play differently recorded tapes, then moving to DOS and discovering how much faster it was to run something compiled by turbo pascal. Then discovering "infinite memory" since you could create a pointer pointing at a place in memory which had... another pointer!!! (I was a very happy kid after discovering that)

I don't have a voice or likeness in my code. Sure, I have a very specific style in all the code I wrote for the past couple of decades, but you can achieve the same outputs with completely orthogonal styles as well. (like AoS vs SoA) so what would I even negotiate for when everything I ever thought of got distilled into neural network weights to be high-pressure sprayed to the masses?

Used claude since 7/2025. Switched to codex after fable got blocked. It was still 5.5 but I knew they had to come up with something. As soon as I switched, wow. It wasn't super intelligent, but it was stable. Every day it was the same performance. This consistency is definitely worth paying for.

Oooh real 64bit ide instead of the crippled 32bit junk they gave as a demo for the past 10 years. Good improvement.

I used to get wrist pain from using mouse and keyboard back 20 years ago. I did no exercises. No stretching, no rock climbing, no triathlons, it just slowly subsided and went away. Maybe I just became aware of positioning and strain to such an expert degree that I just instinctively do the correct moves.

At that revenue, why not make your own filesystem, database and index structure? There is no way mysql is the best possible software for this use case. Why stop innovation and hand everything over to ops?