nottheengineer

There are red license plates for this exact case. Car dealerships generally have them to allow for test drives and bureaucracy drives.

If you buy a car (which is currently unregistered) from a private person that doesn't have those, you just did something that the bureaucracy didn't foresee and at that point, the best course of action is to avoid as much of it as possible.

At this point I have no hope that any of this will getter unless the german state collapses completely.

The entire openAI site was overloaded constantly during working hours just before the last update to the models. After the update, it has had almost no issues at all.

It seems to me that those choices are made with low DPI in mind. Those cuts probably make sure that letters don't become blurry.

I think I'll try this one on my 1080p monitor for a day and see how it goes.

I wittnessed this very fuckery with my own eyes just two weeks ago:

It was an excel file created in pandas that threw an error on english commands but worked if the code included german ones.

I don't know how or why that happens, I just know I need to get away from it.

When doing calculations on paper, you can write your interim results anywhere. That's what excel copies. It's also why text in excel files will gladly overflow into adjacent cells.

Excel is literally 2D programming. Us mortal developers who can only put lines below one another are incapable of comprehending it, so we only get to ask the wise finance people how their enigma works.

On a serious note, I dread excel. If your PC is set to german, excel will translate the VBA keywords to german. But if you want to type them, you have to do that in english and then have excel translate them.

I don't want to accept that crap like this is the standard.

I'm in the position of that user. I have a boss telling me to not turn Team Microsoft's problems into mine. She's happier if I take someone from that team along my troubleshooting journey. It now takes two people's time to arrive at the same conclusion. And most of the time that conclusion is that microsoft screwed up and the only thing we can do is reinstall.

The best stories are always those that end with a rewrite in rust. This one is no exception, thanks for sharing.

Maybe there's another part that leads to this. The world is rapidly growing in complexity, so people may not know exactly what to do after finishing high school. Anecdotally, I found that to be the case in germany.

We get state-funded universities and people often start over in a different subject after realizing they didn't pick the correct one the first time around. I did that too. My and my friends' parents said this was very uncommon in their time, usually due to financial considerations.

I could definitely imagine that there's a similar situation in the US. You get to pick once and if you pick wrong, it's worse than not going to college at all.

Any other input on that? Mine is purely anecdotal and I didn't find any hard data to support or disprove my theory.

Im that case, how does one transfer catch up to the next one? Shouldn't the seek time from switching files have the same impact on both and keep the first one in the lead?

Microsoft does something cool for once and it's an accident?

Some context for those wondering: Those are reserved aliases for google. If you send a mail go abuse@company.com and they use google, that mail will be receiced by google (and the company if they have that set up). It's intended for reporting spam sent out by company.com email addresses.

Not entirely sure how it works, but I assume sending newsletters to that causes some automatic sanctions for the company that are a massive headache to resolve.

I'd like to read that but can't seem to find it. Do you remember the title or have a link?

You post it to HN and claim that it's safe. If you aren't proven wrong in the next 2 hours, it probably is.

I really like using LLMs to learn stuff because they can explain anything at the exact level I need. Hallucination is a big problem with that and RAG pretty much solves it. If I give chatGPT a good stackoverflow post and tell it to dumb it down for me, it does very well. RAG just automates that process with the added benefit of not letting the LLM decide which information to retrieve, which should greatly reduce the chance of accidentally biasing the model with your prompt.

Let's hope they did get it and this is just marketing blabber to avoid admitting any mistakes.

Unfortunately that opens up a whole new can of worms. There are already sites that guess how fake amazon reviews are, maybe we need something similar for github.

I hope an analysis like you proposed could yield some insights to patterns or maybe even enough data to do some machine learning on.

The point of textual is, from what I can tell, making simple and fast UIs very easy to develop. Running in a terminal also makes keyboard input easy, so why reinvent that wheel?

Being able to turn that into a website with little extra work is nice and I think the web absolutely needs a bit more simplicity.

I used firefox without any issues, but today was the first time I tried to make a call instead of joining a meeting. The site said that my browser didn't support that (meetings work perfectly, so that's obviously BS) and that I should use edge or chrome.

So edge and chrome are the only options. Still better than having audio stop working in almost every call and constant crashes.

Onedrive uses sharepoint as a backend, so maybe there's a bug in there that causes both issues.

I can highly recommend uninstalling teams and using the web version instead.

It's a bit less broken and crashes much less often, but the most important thing is that you can just F5 it when it does, which is much faster than starting up the desktop app.

Possibly related: A few weeks ago I was told to use the autosave option of MS word. The login dialog went through, but then nothing happened. Classic Microsoft. I reinstalled Onedrive, autosave said it's on and I go ahead with my work. That was friday 3PM. Come monday, I find that the last 2 hours of friday are gone and spend about half an hour looking for it. I Ctrl+S about every 5 seconds so I can't imagine that there's no copy of it anywhere. But there wasn't one, it was just gone.

What happened was that onedrive was stuck trying to upload my VMs (I had those in my _local_ documents folder, which onedrive claimed ownership of during reinstallation without telling me). Onedrive never showed my excel file in the list of files it was uploading, so it could have either tried to do the VMs first and never got to the excel file or quietly errored out on that without ever uploading it.

In any case, I'm still dealing with the fallout of the fatal mistake I made: Trusting a microsoft product to do the absolute fucking basics.

Maybe I'll learn this time.

You do better than this so your customers don't start hosting themselves.

I read up on that a year or two ago. Then it was mainly IBM mainframes running z/os.

You start differentiating between free and gratis.

Good read, but it focuses a lot on conveying information to the user. The issue nowadays is often the other direction: Interacting with GUIs often feels clunky because there are unnecessary steps, moving buttons on websites or weird mouse rituals that require dragging something a very certain way.

Improving on that should be more of a focus imo.

Pattern reproduction is very close to speech in my opinion. Formal grammars even have it in the name and approaches like https://news.ycombinator.com/item?id=37125118 show that LLMs are indeed very fit for that purpose.

I think I have to walk that claim about math back and try to phrase what I meant differently: LLMs have a hard time with problems that don't translate well into the text space, i.e. abstract problems. Math used to be one of those because early tokenizers were designed just with text in mind and LLMs weren't good enough to overcome those limitations. OpenAI put in a lot of effort into their tokenizers to make GPT3.5 and GPT4 better at math specifically.

The second paper you linked is very interesting and I think it supports my original assertion of text space and latent space being close. The first graph shows GPT3.5 doing much better at pattern reproduction and language tasks while humans still hold an advantage in the more abstract tasks like story analogies. Higher order relations being a key thing that's measured maybe makes this task a bit too perfect for arguing my case, but it does show that humans have an advantage in more abstract situations.

I think any problem that can be viewed as being mostly a form of translation is a good one for LLMs and if you can express a problem as that, you can get better results.

To get back to the main point: latent space and text space, or feature space in general, being close is what I believe causes all of this. Happy to hear counterexamples.

Arch. My PC sometimes hangs for no reason, which it also did with windows. It used to be a non-issue because it was too loud to run it overnight, but now it's quiet so I'm trying to figure that out. Every nvidia driver update has about a 10% chance to break something. That was also the case on windows so I just didn't update it there and had great success with recommending people to find a version that works and never touching it again.

I don't recall ever having an issue that was fixed by restarting an application though. If it doesn't work, I read the error message to figure out why it doesn't work and fix that. No "just try again, it might work this time" like on windows, which I enjoy.

SMB really is hell. I write business software and our admins do what they can to avoid using it because it's very hard to know when it fails.

But from a dev perspective, that's probably a driver level issue because that's where you want to handle or propagate errors instead of crashing.

While it's nice to imagine what this could be used for, it just looks like frontend dev hell with so many moving parts.