0xFluegel

I had a codex session stop in the middle because the auto-reviewer timed out with something like "auto-review not available at the moment". The capacity issues are very clearly observable since gpt6-astra launched.

Hamburg, Germany: the public transport services released a new app a few months ago which assumes that the request failed when the app has not finished receiving an answer to a route query after a given time span. The problem: It is normal to have a data cap that, when reached, limits the rate to 64 kbit/s. That is too slow for the answer to be fully transmitted in time...

At least the website works, even though it transmits the full site for every request...

> I live in Japan which people usually mock for still using fax machines

Greetings from Germany as well...

> I heard it talked about on a podcast […]

Do you remember which one? I'd like to get some more detail on this...

> … the vast majority of right leaning sources (aside from fox) are effectively blacklisted.

This reminds me of a past blog post, a few weeks ago or so, from a journalist who spoke about how strongly right leaning "journalists" often think of their writing as political influencing and counter-acting the left journalism.

I don't know how true that really is but such a tendency could be a good faith explanation for such a rule (as long as it has exceptions).

Since it's a student project to learn about VHDL and synthesis the pong clone had a few additions like simple math-expression based textures and we ignored trying to get close to the exact movement behaviour. But you actually get to control it with rotation based input. :)

In regards to ROM. It was an FPGA, meaning it was reprogrammed on every startup. But all behaviour was built as logic without accessing any memory.

I still have a VHDL version lying around from my old student project. :)

> Looking now at the demos at https://www.qt.io/qt-examples-for-webassembly, a lot of the same problems are jumping out at me. A complete lack of accessibility features, poor handling of scroll events, large load times, unfocusable fields, lack of keyboard controls, etc...

I found that hard to believe. But, clicking that link and trying the pizza demo... Yikes. It's not exaggerated at all.

Germany has that, too. Any USB-Stick, CD, etc. that a normal user can write any data onto.

The sensor needs a lot more energy. This produces two types of IR cams: actively and passively cooled. The passive cooling is easy to take around with you but only possible with a smaller sensor.

As for the price per pixel: I don't know from memory.

Can you give me a reference? I am somewhat interested and my current information is that they are generally trying to work around the disadvantage of giving up scale compared to traditional farms.

> And an unhandled exception crashes the program, which is often worse than an invalid value.

Please elaborate. Why is working with garbage better than not working at all? Would it not also rather make it easier to find these out-of-range cases with fuzzing?

I want to like that idea. Unfortunately, I can counter with: When you (as an outsider) see two very similar projects, one claiming to be from 2018 the other from 2020, what will you look into/try out first?

Is it though? It surely is authenticated but is it authorized? By whom? Certainly not by the user.

I would call that system secure. It does not just rely on an obscure password but is actually restricted by a list of whitelisted networks.

The failure in that case is only that the admin didn't consider that normal work might be done from home at some point or that the middle or upper manager thinks that he should be able to freely administrate his critical infrastructure from anywhere...

I can only half agree with you on that. Yes, I also dislike APIs that make wrong or unsafe use easy and correct use more bothersome but seemingly no different in behaviour (until it goes BOOM), but I also find that soooo many people simply don't have the awareness that they are interfacing with another system that interprets their data in a potentially unsafe way. And these people will misuse any API like this.

I can see how that happens immediatly...

    var domain = addr.split(['@','.']).take_last(2)
    var tld = domain[1]
    // check TLD against wacky rules...

See, boss, it's easy. What's next?

GDPR has a section that defines a way for an EU institution to "grab" cases, but that's timely and not used much (or at all, I don't know about the numbers).

Thanks for clarifying. You convinced me.

> Anyone doing secure comms at this level, and is talking about families of mathematics always gives me the impression they don't really know what they're doing.

Why is that? Do you assume that making competent choices for encryption algorithms (for which you try to understand the math problems involved) and trying to market the systems security means that they also try to implement it themselves? Or is the "family of mathematics" a sign for incompetence that I just don't recognize?

Could also simply mean that they are chaining different encryption algos from libraries (à la TrueCrypt) which would indeed add to the security level or even at worst not harm it. (This assumes that each step is not broken...)

Can you elaborate on why Catch would be impossible or unergonomic? It's a test framework, right? Why would one even use it instead of the normal rust test framework?

Also, how is it ergonomic? I'd agree if you say that it's (one of?) the best for C++, and it's also my go-to option, but ergonomic?

Is it less effective, though? From what I remember about successes e.g. against organized crime or other groups that operated partially or totally over the internet in the news over the last years, it was mostly the simple and dull ~~physical~~ police work that leveraged mistakes of operational security, but not an evaluation of minable data (even when learned in hindsight that it existed).

E.g. looking at the logs of relevant servers and waiting for someone to login without their VPN at some point.

My german bank Haspa has a limit of 12. My guess for the reason -- apart from "we didn't have any problems, yet" -- is that part of the relevant infrastructure is ancient and uses a hardcoded max length...

> While 32 chars is plenty long for a password, [...].

It might be plenty long for completely random characters, but when one uses a word sequence, the 32 characters are easily reached.

In current VR technology that's going to be hard and annoying:

> Text is tough in VR. It’s hard to read, given the resolution of today’s HMDs and it’s hard to write, since you’re typically blind in a headset and it’s annoying to be tied to a keyboard at a desk when you’d rather walk and move around in VR. I tolerate these problems with RiftSketch by making the text in the editor extremely large. I can only see 20 lines of code at a time in VR whereas my physical desktop has a 4K monitor where I’m usually looking at 140 lines of code per file with several files open side-by-side. (https://uploadvr.com/riftsketch/)

And that isn't even trying to give you a pen & paper experience. (<- something I'd also like to have.)

Any finite sampling of any distribution (biased or not) can show bias. A feasible process that doesn't introduce or exagerate bias is the biggest problem when trying to get representative statistics.

And bias can certainly be of different magnitude. A weighted dice can have a bias that shows up after only 10 rolls and another one can only show it's bias after 1000.

It'll be even worse when the search engine later magically ranks their's higher even when searching for the original.